User Modeling for Building Personalized Web Assistants ABSTRACT

合集下载

虚拟人英语作文模板

虚拟人英语作文模板英文回答：Virtual Humans: A Paradigm Shift in Human-Computer Interaction。

Virtual humans, also known as digital humans or artificial humans, are computer-generated characters that mimic human behavior and appearance. They are designed to provide a more natural and immersive user experience in a wide range of applications, including customer service, education, healthcare, and entertainment.History and Development。

The concept of virtual humans dates back to the early days of computer graphics. In the 1960s, researchers began experimenting with creating human-like characters for use in flight simulators and other training applications. However, it was not until the 1990s, with the advent ofpowerful personal computers and advanced 3D graphics software, that virtual humans became a viable technology.Components of a Virtual Human。

如何定义用户模型（persona）

如何定义用户模型（persona）一、什么是用户模型？Persona（[p?:'s?un?]）：(Persona是用户模型的的简称)是虚构出的一个用户用来代表一个用户群。

一个persona可以比任何一个真实的个体都更有代表性。

一个代表典型用户的persona的资料有性别、年纪、收入、地域、情感、所有浏览过的URL、以及这些URL包含的内容、关键词等等。

一个产品通常会设计3～6个用户模型代表所有的用户群体。

Persona Web人物角色介绍用户模型（任人物角色）不是用户细分用户模型看起来比较像用户市场细分。

用户细分通常基于人口统计特征（如性别，年龄，职业，收入）和消费心理，分析消费者购买产品的行为。

用户模型更加关注的是用户如何看待、使用产品，如何与产品互动，这是一个相对连续的过程，人口属性特征并不是影响用户行为的主要因素。

用户模型是为了能够更好地解读用户需求，以及不同用户群体之间的差异。

用户模型（人物角色）不是平均用户某个人物角色能代表多大比例的用户？首先，在每一个产品决策问题中，“多大比例”的前置条件是不一样的。

是“好友数大于20的用户”？是“从不点击广告的用户”？不一样的具体问题，需要不一样的数据支持。

人物角色并不是“平均用户”，也不是“用户平均”，我们关注的是“典型用户”或是“用户典型”。

创建人物角色的目的，并不是为了得到一组能精确代表多少比例用户的定性数据，而是通过关注、研究用户的目标与行为模式，帮助我们识别、聚焦于目标用户群。

用户模型（人物角色）不是真实用户人物角色实际上并不存在。

我们不可能精确描述每一个用户是怎样的、喜欢什么，因为喜好非常容易受各种因素影响，甚至对问题不同的描述就会导致不同的答案。

如果我们问用户“你喜不喜欢更快的马？”用户当然回答喜欢，虽然给ta一辆车才是更好的解决办法。

所以，我们需要重点关注的，其实是一群用户他们需要什么、想做什么，通过描述他们的目标和行为特点，帮助我们分析需求、设计产品。

Personalization of Content in Virtual Exhibitions

Personalization of Content in Virtual ExhibitionsB. Bonis1, J. Stamos1, S. Vosinakis2, I. Andreou3 and T. Panayiotopoulos11Department of Informatics, University of Piraeus, Piraeus, Greece{bonisb, jstamos, themisp}@unipi.gr2Department of Product and Systems Design Engineering, University of the Aegean,Syros, Greecespyrosv@syros.aegean.gr3Department of Technology Education and Digital Systems, University of Piraeus,Piraeus, Greecegandreou@unipi.grAbstract. Presentation of content is an important aspect of today’s virtualreality applications, especially in domains such as virtual exhibitions. The largeamount and variety of exhibits in such applications raise a need for adaptationand personalization of the environment. This paper presents a contentpersonalization framework for Virtual Exhibitions, which is based on asemantic description of content and on information implicitly collected aboutthe users through their interaction. The proposed framework uses stereotypes toinitialize user models, adapts user profiles dynamically and clusters users intointerest groups. A 3D virtual museum has been implemented as a case study,and an evaluation has been conducted.Keywords: Virtual Reality, Virtual Museums, User Modeling, Personalization1 IntroductionThis advent of Virtual Reality in recent years has enabled the development of novel interactive and immersive applications that emphasize on the presentation of content and satisfying user experience. Examples of such applications are 3D Virtual Exhibitions, i.e. interactive 3D environments containing a large collection of media objects. While the graphical representation of the exhibits is an important aspect of Virtual Exhibitions, the large amount and variety of these elements presents another challenge to the designers of such environments, namely to adapt the presentation of a collection according to user interests, i.e. content personalization.A number of Virtual Exhibition applications are currently available and run standalone [1], or over the Internet [2,3,4], representing real or fictional museums and exhibitions. It can be noted that most of today’s exhibition environments emphasize on the content and aim to the deeper understanding of entities and concepts through user navigation and interaction in 3D. Most of these applications contain, nevertheless, static collections arranged in predefined positions, and the design of the virtual space and its contents is based entirely on the author’s categorization. Therefore, the user’s role is limited to a passive observer and the presentation of large2 B. Bonis, I. Stamos, S. Vosinakis, I. Andreou and T. Panayiotopouloscollections may fall into the obstacle of navigational difficulties in 3D environments [5], which eventually leads to the reduction of user interest and to the disability to explore and search for the desired content [6].The problem of presenting and categorizing large quantities of content has been effectively addressed in Web [7,8,9,10] and multimedia applications [11], where user modeling has been used to personalize content presentation based on the users’ own interests. The authors claim that virtual environments could also benefit from user modeling and adaptation techniques that make assumptions about user interests and intentions concerning the application, and construct the virtual space accordingly. Such a personalized space may reduce the navigational burden while it still retains the metaphor of being immersed in a 3D environment. Although the concept of personalizing content is not new, there have been few approaches towards user modeling in 3D environments. In analogy to the theory of content adaptation on the Web [12], content personalization in 3D could be supported by a) a process of recognizing user interaction patterns in the 3D environment [13], b) a mechanism that makes assumptions about user preferences based on these interactions, and c) respective modifications of the environment that reflect user needs and increase her satisfaction.As a first step towards this theory, Chittaro and Ranon have proposed AWE3D [14], an architecture for presenting personalized 3D content on the Web. In this approach, the system monitors and records user interactions with the environment and applies a rule set on the recorded interactions in order to modify the user model, and personalizes the environment based on assumptions concerning user preferences. They present a 3D E-Commerce application as a case study. The work of dos Santos and Osorio [15] also focuses on content adaptation in 3D environments. They use a rule based approach with certainty factors to modify the content and structure of the environment according to user interests. A distance learning environment has been created as a prototype. Both approaches adopt a rule based mechanism to generate and update the user models, and the responsibility of defining and validating these rules lies entirely on the designer. Any modification of the content will require respective changes in the rules.Celentano and Pittarello [13] propose an approach to facilitate adaptive interaction with the virtual environment, which is based on the following: a structured design of the 3D interaction space, the distinction between a basic virtual world layer and an interaction layer, and the recording of the environment’s usage by the user in order to find interaction patterns. The aim is to facilitate the system’s usage by monitoring user behavior and predicting future needs for interaction purposes. When the system recognizes the initial state of an interaction pattern, it executes the final state without letting the user engage in the intermediate ones.The authors propose a novel approach towards adaptive virtual environments, in which objects are dynamically distributed among rooms and users may experience personalized presentations based on their previous interaction with the system. The proposed framework is supported by a semantic graph, defined by the designer, which describes the nature of the exhibits by hierarchically categorizing the content, and drives the user modeling process. Additionally, the authors support the claim that users’ experience in a virtual exhibition is enhanced through communication and collaboration with other users in a shared environment [16]. In this context, thePersonalization of Content in Virtual Exhibitions 3 proposed framework supports dynamic clustering of user groups based on the similarity of user models, which may lead to the formation of e-societies with similar interests. The authors have implemented a science fiction virtual museum as a case study of the proposed framework and have performed a user evaluation of the application. Evaluation results indicate the impact of the user modeling process in adapting the presentation of content to the user preferences and a significant degree of user satisfaction from the system in general.The rest of the paper is structured as follows: section 2 presents the proposed framework for content personalization in virtual exhibitions, while section 3 presents a science fiction virtual museum as a case study and section 4 presents an evaluation of the implemented system. Finally, section 5 draws the conclusions and states the future work.2 A Framework for Content PersonalizationThe proposed framework for designing and implementing virtual exhibitions with content personalization is based on four methods that enhance a static 3D environment with dynamic characteristics: user model generation, content selection and presentation, user model update and clustering. In the next paragraph a full user interaction session with an application designed under this framework is presented. When a new user enters the environment for the first time, a user model is assigned to her based on stereotypes [17] that correspond to her selection of an avatar, i.e. her graphical representation in the 3D environment. While the user is browsing the environment, her navigation and interaction with content are monitored and the recorded behavior is utilized to make assumptions about her interests and preferences, which are then incorporated into the user’s profile. At any time, the user can ask to be transported to a personalized environment, which reflects her assumed preferences and recommends new content that might be of her interest. The user can also join communities with similar preferences, visit other personalized environments, and exchange opinions about the content. User interest groups are proposed by the environment through an automated clustering process.From the designer’s point of view, the framework can be employed to construct new dynamic virtual exhibitions without having to define explicit rules for content personalization and adaptation. The designer has to provide the 3D content, i.e. the rooms and objects of the environment, the semantic graph, i.e. an ontological description of the content, and the user stereotypes that contain templates of initial user preferences concerning the content. A presentation process then creates the exhibition rooms and distributes the exhibits dynamically based on the above data. The personalized environments also depend on the interaction history of the respective users. Furthermore, exhibitions generated using the proposed framework can be easily adapted or enhanced by altering or inserting new 3D content and making appropriate changes in the semantic graph and the user stereotypes.4 B. Bonis, I. Stamos, S. Vosinakis, I. Andreou and T. PanayiotopoulosFig.1. The proposed architectureThe proposed framework is based on a thin client/server architecture (Fig.1), in which the users interact with the environment on the client side and user modeling takes place on the server side. This architecture follows the paradigm of decentralized user modeling architectures [18], and allows clustering of users into groups with similar preferences, whilst it is also necessary for the support of multi-user environments and for immediate adaptation and expansion of exhibition data by the designer.The rest of this section describes the proposed framework and the related architectural components in detail.2.1 The Semantic GraphA vast number of applications that utilize user modeling methodologies try to address the user’s need for quick and efficient access to a subset of information that meets her interests and preferences, without having to search through a larger set of objects. A widely used term in the literature for describing these applications is recommender systems [7,10,11,17,19,20,21]. A distinctive characteristic of these systems compared to information retrieval and filtering systems and to search engines is the output of individualized information based on a priori knowledge about the content and assumptions about user preferences.A thematically uniform set of objects can be grouped together and categorized based on a number of criteria; relations can be determined between objects and categories or between categories themselves, e.g. relations of affinity and inheritance. For example, in an art exhibition, exhibits can be grouped with respect to their creators, the epoch or the style. These categories can be generalized into broader categories or specialized into subcategories. A categorical hierarchy of this type forms a tree with nodes being the categories and edges being the relations between them. The entirety of the categories can thus be represented as a forest (a set of distinctPersonalization of Content in Virtual Exhibitions 5 trees). Nodes at higher levels in each tree imply categories with broader meaning and relations between nodes can be viewed as inheritance relations. Nodes of different trees can be connected implying a semantic union relation. Because of the union relations this categorization scheme is used as a graph instead of a forest in the context of determining related categories. The actual objects are attached to the categorization trees using connection(s) with one or more lowest-level category nodes (the leafs of the categorization trees) via an instance relation. Thus, the resulting structured hierarchical semantic taxonomy forms a directional graph, the Semantic Graph (SG), in which nodes (distributed into levels) stand for objects and concepts, edges represent the relations between them [19,22,23] and the levels represent the degree of generalization.Nodes are divided in two categories, the object nodes, which represent the actual elements of the environment, and the categorization nodes, which represent object categories. The latter are distributed hierarchically based on generalization relations; a broader term lies at a higher level than a more specialized one.Figure 2 presents a sample part of a semantic graph that is used for categorizing cars. The dashed line is a union relation that connects two categories belonging to different trees. In the example the node Type is divided into two subcategories, the nodes Professional and Racing that are also divided into two subcategories each. The EVO 5 has been used as a WRC racing car and it was manufactured by Mitsubishi, thus it can be connected with the nodes WRC and Mitsubishi. The union relation connects the respective nodes to imply the association between Ferrari manufacturer and F1.Fig. 2: A part of a semantic graph for categorizing carsLet A be the node at the lower end of a categorization edge and B the node at the upper end. All edges (which represent the aforementioned relations) have a numerical weight in the range of (0, 1], which is the degree of membership of the object or concept that node A represents to the set that the node B portrays. The use of the degree of membership is analogous to the respective term from the Fuzzy Set Theory [23]. For example, the Terminator II movie can be said to be a well known sci-fi movie of the 90’s, so the degree of membership of this movie in the Sci-fi and 90’s categories is significant. In the case of union edges, the weights represent the degree of association between the connected nodes. The choice of weights can be made by using expert knowledge or by using machine learning techniques. The degrees of6 B. Bonis, I. Stamos, S. Vosinakis, I. Andreou and T. Panayiotopoulosmembership and the degrees of association are used during the execution of the personalization-recommendation algorithm.In the proposed framework, the SG is the core element, because it drives most computational procedures:•modeling user preferences in the environment,•providing recommendations to users,•clustering users with similar preferences to groups, and•dynamically distributing the content into separate rooms and determining the connections between them.The authors argue that the SG facilitates access to the exhibition’s content repository, by reflecting a natural content interpretation, effectively serving the users’ informational needs and preferences.2.2 Dynamic Content Generation and RenderingThe hierarchical structure of the SG is used for the creation of the virtual exhibition’s spatial structure. Each categorization node can be represented by a set of conceptually relevant rooms that are connected via doors, following the graph’s edges and the relations of kinship between semantic nodes. This approach provides a multidimensional navigation paradigm, in which rooms are connected based on their semantic similarity. A user can navigate inside the exhibition environment and browse the content by following paths that correspond to any of the concepts that can characterize it. E.g. a room with objects from Steven Spielberg movies is connected with the room of US directors (at a higher level). By dynamically structuring the environment, the effort of designing and creating every room in the exhibition is reduced to a minimum.The designer provides a number of template rooms that are used by the client application to construct the actual exhibition rooms. The system requires a default template room and, optionally, any number of thematic rooms related to existing categories of the semantic graph. Each template room is a 3D model of a section of the exhibition environment, in which two types of objects are inserted: doors and exhibit containers. When the client application has to construct a thematic exhibition room it searches for a predefined template that matches the respective category. In the case that no such template exists, the default room is used. Then, the system dynamically links the room with related ones using the existing doors, each labelled by the name of the room it leads to. Finally, the exhibits are dynamically placed in the containers defined by the designer. A default exhibition is constructed by creating an entrance room that is connected to all the general (top-level) category rooms of the semantic graph.The collection of objects that will populate each room is generated by traversing the graph from the respective categorization node to the set of object nodes. The number of objects can be significantly large, especially when browsing the content of higher-level nodes. In that case, objects are distributed into a set of interconnected rooms.Personalization of Content in Virtual Exhibitions 7 2.3 PersonalizationAll available objects in the proposed framework are categorized by the SG in a semantic taxonomy. When and if a user is interested in a certain object, it can be assumed that she is also interested in one or more categories to which the object belongs (she likes the movie series Star Wars because she likes sci-fi movies). If this belief is reinforced during the interaction with the system, a recommendation set with members originating from this particular category would probably be a preferable choice. The user model is an instance of the SG, with all nodes being given a numerical value (positive or negative), which represents the degree of interest that the user is assumed to have for the term or object that the node portrays. The process of calculating these degrees of interest is analytically explained in the next paragraphs. User Stereotypes. As stated in [24], the explicit creation of a user profile may annoy users that are unwilling to state their interests and lead to user models that do not actually reflect the user preferences. Therefore, an indirect method for generating an initial user model has been utilized, in which a set of stereotypes [17] is used to initialize the model of a new user. At registration time, the user selects an avatar from a provided library, and the system creates an initial user model and thus makes an assumption about her preferences and interests based on her selection. Each avatar is related with assumptions about the user, which are lexical values of properties defined by the designer. A possible set of properties can be age, sex, education or anything that is considered to characterize the users and the respective values can be young, old, female, male, high, low etc. The stereotypes are rules that relate each value of each property with estimated degrees of interest for a set of nodes in the SG.The degree of interest in the categorization nodes is calculated as follows. Initially all the categorization nodes have a zero value. For each value of each property, the degree of interest declared in a stereotype (if any) is added to the respective categorization node. This initial user model is used for the formation of a recommendation set (which will be explained in the following paragraphs) prior to the user interaction with the 3D environment. This approach deals with the new user problem [20], with a reduced accuracy nevertheless. This is an initial estimation, however, and as the user interacts with the system, information is accumulated in the user model, updating it and thereby increasing its accuracy.The aforementioned methodology is based on researches [25,26] that state the possible relation between the choice of an avatar by the user and the users’ intrinsic characteristics and personality trades. Expert knowledge and/or user segmentation researches can be used to create the set of stereotypes.Data Acquisition and User Model Updating. The algorithm that updates the user model based on her interactions uses a forward propagation mechanism. Initially, it collects the summing degree of interest for each object node of the SG that the user has interacted with. This is the result of a monitoring process that takes place on the client side. A variety of acquisition methods can be used, such as measuring the time spent by the user observing an object, taking into account the type of interaction, and providing a rating system that lets users express their preferences. The framework8 B. Bonis, I. Stamos, S. Vosinakis, I. Andreou and T. Panayiotopoulossupports various types of acquisition methods and interpretations. At the end of this step, all object nodes have a degree of interest ranging from negative values (dislike for the particular object) to positive values (positive interest).For algorithmic uniformity, each union relation in the SG is substituted with a dummy node (union node) placed at one level higher than the maximum level of the categorization nodes it connects. These nodes are connected to the union node with dummy relations that have weights equal to w , where w 2 equals to the degree of association of the union relation . For the rest of the paper we refer to both categorization and union nodes with the term semantic nodes .Let SN be a semantic node and DI (SN )t the degree of interest in node SN at time instance t . Let CSN = {CSN 1, CSN 2, …, CSN N } be the set of children nodes of node SN , with DI(CSN i )t , i ∈ [1,N] being the respective degrees of interest. Let also W i be the weight of the edge that connects CSN i to SN . Then, the degree of interest at time instance t +1 for node SN is calculated by:∑=++⋅=N i t it i t SN DI W CSN DI SN DI 11)())(()(Calculations begin at level 0 (the object nodes level) and they proceed level by level until the maximum level of the SG. Object nodes maintain the values from the initial step. The resulting degrees of interest comprise the updated user model.Adaptation of the Environment . For the creation of a personalized room, a set of objects is chosen to be recommended to the user by populating the room with them. This task is realized using a backward propagation mechanism based on the user model. This mechanism is computationally equivalent to the forward propagation mechanism described before, having only the opposite direction. Thus, instead of the CSN , the ASN = {ASN 1, ASN 2, …, ASN K } set is used which represents the set of ancestor nodes of node SN and N is replaced by K . During the backward propagation procedure an estimated degree of interest is assigned to the object nodes. The choice of objects that will be part of the user’s personalized room depends on the ratings of the respective object nodes. Consequently, the simplest approach is to choose the M top rated nodes of level zero (the level with the object nodes), where M is the capacity of the room, whilst variations of the proposed room can be produced by employing methods, such as to insert a few random elements to increase the diversity, or to avoid selecting objects that the user has already interacted with.Besides the selection of the recommendation set, a personal room should also have connections to other rooms of the virtual environment, allowing the user to further explore the exhibition. The set of rooms that are connected to the personalized room should also reflect the user’s preferences. As mentioned earlier, every categorization node is represented as a set of rooms. So, to connect the personal room with the rest of the environment, L doors, where L is the door capacity of the personal room, are dynamically created. These doors are connected to a single room of each room set from the L top rated categorization nodes of the user’s model.Personalization of Content in Virtual Exhibitions 9 2.4 Clustering: Formation of user communitiesIn internet based applications, added grouping capabilities can promote the formulation of e-communities, thus increasing the sense of immersion in the virtual environment, enhancing communication opportunities, and satisfying the need for social interaction and awareness. To create a group of users with similar interests and preferences in the proposed framework, the user models must be compared. By assigning a unique index to every node of the SG, a user model can be taken as a numerical vector with each item being the degree of interest in the corresponding node. The dissimilarity between a pair of vectors can be computed using multidimensional Euclidean distance, and thus a group can be created based on distances smaller than a given threshold. This technique can provide satisfactory results in finding users with similar interests and presenting the respective personalized rooms to promote user communication. The user can choose a personalized room from a subset of recommended rooms, the owners of which have been clustered by the system, based on their interests.3 Case Study: A Science Fiction MuseumIn order to assess the proposed framework, a Virtual Museum concerning science-fiction movies, has been developed. It lets users navigate in thematically different rooms (e.g. containing characters, vehicles etc.), enter a personalized room with various exhibits that the user might enjoy, according to her interests, or enter online exhibitions of users with similar interests. The exhibition’s content (3D models) has been created using external modeling tools and imported to the environment, while an authoring tool has been implemented to facilitate the process of construction, manipulation and maintenance of the SG and the system’s database management. The authoring tool can be used by the designer of the virtual exhibition, effectively reducing the effort of creating all the needed elements. The Virtual Museum contains 87 exhibits with a SG consisting of 34 nodes distributed over 3 levels.As the users navigate inside the virtual museum, they look at exhibits that fall inside their field of view. In order to perceive an exhibit, a user must be oriented towards it and be within a predefined distance. Furthermore, the users have the option of manipulating an exhibit, i.e. rotating it in order to gain a full perspective. Presentation of additional information concerning the exhibit is achieved through description pages in a provided panel. The users are also able to provide feedback about an exhibit by rating it and entering their personal message/comment. At the same time, users can view comments written by others, chat with them, observe their actions in the virtual space (via their avatars) and enter their personalized room.The initial assumptions concerning user preferences are based on stereotypes related to the avatar gallery, and consequent interaction in the exhibition’s space provides the appropriate degrees of interest for the profile update. Three types of interaction are monitored (ordered by the degree of influence upon the rating of an object, ascending):10 B. Bonis, I. Stamos, S. Vosinakis, I. Andreou and T. Panayiotopoulos•viewing time: the time spent looking at an exhibit. It is considered that the more time the user looks at an exhibit, bounded by an upper limit, themore interested she is in it.•manipulation of an exhibit: interaction with an exhibit has a stronger impact than viewing time, as the user expresses her preferences moreexplicitly.•commenting and rating exhibits: users can write comments that others can read and rate the exhibits accordingly. There are three available ratingvalues: positive, negative and indifferent, which can reveal the user’sopinion directly.User interactions produce degrees of interest that are used by the server to assembly the personalized room and to distribute its contents dynamically upon user request. Users have the added ability to be instantly transported to any exhibition room (including the personalized room), in order to access rapidly a desired category/exhibit room.Fig. 3. A screenshot of the case studyThe museum’s 3D content is modeled in VRML and is loaded/rendered by the client application, which is implemented in Java, using the Java3D API. At the server side, the exhibition and user data were stored into an SQL Server 2000 database and their retrieval/manipulation is coordinated by the server application, implemented in Delphi programming language. The client-server communication is achieved with the use of TCP/IP sockets. Figure 3 presents a screenshot of the virtual museum application.。

建筑方案优化模型英文

建筑方案优化模型英文Optimizing Architectural Plans: A Model for Excellence Introduction:Architectural design is a complex and iterative process that involves numerous considerations and trade-offs. From aesthetics to functionality, architects strive to create spaces that are visually pleasing, functional, and sustainable. However, achieving an optimal architectural plan often requires multiple iterations and adjustments. To streamline the design process and enhance efficiency, an optimization model tailored specifically for architectural plans can be implemented. This model can help architects strike a balance between various factors and enable them to deliver exceptional designs that meet client requirements while minimizing resource utilization.Objective of the Model:The primary aim of the architectural plan optimization model is to identify the most efficient and effective design that fulfills all the specified criteria. These criteria may include client preferences, budget constraints, building regulations, and environmental considerations. By leveraging mathematical algorithms and optimization techniques, architects can evaluate different design alternatives and identify the most optimal solution. The model takes into account various factors such as cost, energy efficiency, structural integrity, and spatial utilization, enabling architects to make informed decisions and save valuable time during the design process.Methodology:The optimization model for architectural plans follows a systematic approach to ensure accurate and reliable results. The process begins with defining the problem statement and identifying the key objectives and constraints. Once the problem is established, architects can proceed to gather relevant data and inputs. This includes input parameters such as project requirements, building codes, environmental regulations, and cost constraints. Additionally, architects can integrate existing building databases and simulation tools to enhance the accuracy and realism of the model.Next, architects need to convert the architectural plan into a mathematical representation known as a design variable. This design variable encapsulates the key elements of the design, such as building dimensions, orientation, material selection, and structural components. The mathematical model is then formulated based on the design variable and input parameters, which involves defining objective functions and constraints that reflect the desired outcomes and limitations.To optimize the architectural plan, various optimization algorithms can be employed. These algorithms employ techniques such as genetic algorithms, simulated annealing, or particle swarm optimization to explore the solution space and identify the most optimal design. By iteratively evaluating different design alternatives, the model helps architects narrow down the search space and converge on the best architectural plan.Benefits and Applications:The implementation of an optimization model for architecturalplans offers several benefits to architects and design firms. Firstly, it facilitates faster and more accurate decision-making, enabling architects to explore a wide range of design possibilities efficiently. This not only saves time but also ensures that the design meets all requirements and constraints.Secondly, the model enhances the design quality by considering multiple objectives simultaneously. Architects can balance conflicting factors such as cost and sustainability, allowing for a more holistic and well-rounded design approach. The optimization model also enables architects to evaluate and incorporate environmental factors, such as energy consumption and carbon footprint, ensuring the creation of sustainable and eco-friendly structures.Furthermore, the model can aid in cost estimation and budget control by providing insights into the cost implications of different design choices. Architects can identify cost-effective design alternatives and avoid potential cost overruns or budget constraints. Conclusion:In conclusion, the implementation of an optimization model for architectural plans offers immense potential for improving the efficiency and effectiveness of the design process. By enabling architects to evaluate multiple design alternatives and consider various factors simultaneously, the model promotes the creation of exceptional architectural plans. It not only saves time and resources but also enhances design quality, sustainability, and cost-efficiency. As the architectural industry embraces digitization and advanced technologies, the integration of optimization models willbecome an essential tool for architects in delivering outstanding designs that meet the evolving needs of clients and society.。

关于对人设的建议英语作文

关于对人设的建议英语作文Title: Constructive Suggestions for Character Development。

In the journey of life, crafting a well-rounded and admirable persona is of paramount importance. A strong character not only shapes one's interactions with the world but also serves as a compass guiding one's decisions and actions. However, character development is an ongoing process, requiring introspection, effort, and openness to growth. Here are some constructive suggestions for enhancing and refining your character:1. Self-Reflection and Awareness: Take time to introspect and understand your strengths, weaknesses, values, and beliefs. Self-awareness forms the foundation of character development as it enables you to identify areas for improvement and leverage your strengths effectively.2. Embrace Growth Mindset: Cultivate a growth mindsetthat views challenges as opportunities for growth rather than obstacles. Embracing a mindset of continuous improvement fosters resilience, adaptability, and a willingness to learn from both successes and failures.3. Practice Empathy and Compassion: Develop empathy by seeking to understand others' perspectives, emotions, and experiences. Actively listening to others without judgment and showing genuine compassion cultivates meaningful connections and fosters a sense of community and belonging.4. Integrity and Honesty: Uphold principles ofintegrity and honesty in all your interactions. Be truthful, reliable, and accountable for your words and actions, even when faced with difficult choices or temptations. Integrity forms the bedrock of trust and credibility in relationships.5. Cultivate Resilience: Life is fraught withchallenges and setbacks. Cultivate resilience by developing coping mechanisms, maintaining a positive outlook, and reframing adversity as an opportunity for growth and learning. Resilience enables you to bounce back strongerfrom setbacks and persevere in the face of obstacles.6. Practice Gratitude: Cultivate an attitude of gratitude by acknowledging and appreciating the blessings, opportunities, and supportive relationships in your life. Practicing gratitude fosters humility, resilience, and a sense of abundance, leading to greater overall well-being and satisfaction.7. Embrace Diversity and Inclusion: Value and respect the diversity of perspectives, cultures, and identities in your community and beyond. Actively seek out opportunities to learn from others' experiences and perspectives, and challenge your own biases and assumptions. Embracing diversity and inclusion fosters empathy, creativity, and collaboration.8. Seek Personal Development: Commit to lifelong learning and personal development. Set goals for self-improvement in various aspects of your life, whether it be acquiring new skills, pursuing passions, or fostering meaningful relationships. Engage in activities thatchallenge you to step out of your comfort zone and expand your horizons.9. Practice Self-Care: Prioritize your physical, mental, and emotional well-being by practicing self-care rituals. Nurture yourself through activities that rejuvenate and replenish your energy, such as exercise, meditation, hobbies, or spending quality time with loved ones. Remember that self-care is not selfish but essential for maintaining balance and resilience.10. Lead by Example: Be a role model for others by embodying the values and principles you espouse. Lead with integrity, empathy, and authenticity in your interactions with others, inspiring them to cultivate similar virtues in their own lives. Remember that small acts of kindness and integrity can have ripple effects, contributing to a more compassionate and harmonious world.In conclusion, character development is a lifelong journey of self-discovery, growth, and refinement. By embracing these constructive suggestions and committing tocontinuous self-improvement, you can cultivate a character that not only reflects your values and aspirations but also inspires others and contributes positively to the world around you.。

英语作文my model

英语作文my modelIn the vast canvas of life, we all seek inspiration and guidance from those who excel in their fields. They are our models, the beacons that guide us towards our own aspirations. My model, too, is someone who has inspired me deeply, someone who, with their actions and achievements, has shaped my understanding of success and personal growth.My model is not a celebrity or a public figure, but a person who has quietly excelled in their chosen field, someone who has demonstrated the power of hard work, perseverance, and unwavering dedication. They are a scientist, someone who has spent years in laboratories, meticulously conducting experiments, analyzing data, and contributing to the advancement of knowledge.What I admire most about my model is their unwavering commitment to their work. They are someone who has never been swayed by the glamour or fame of other professions. Instead, they have chosen to stay true to their passion, topursue knowledge and understanding, and to make adifference in the world through their research. This kind of dedication and focus is truly admirable.Moreover, my model is someone who has always been approachable and generous with their time and knowledge. They are someone who is happy to share their experiences and insights, someone who is eager to mentor and guide those who are interested in following in their footsteps. This humility and generosity have made them not just a role model in their field, but also a mentor and a friend to many.The impact of my model on my life has been profound. Their dedication to their work has taught me the importance of staying true to one's passions and pursuing what one loves. Their humility and generosity have taught me the value of sharing knowledge and helping others. And their unwavering commitment to their field has taught me the importance of perseverance and hard work.In conclusion, my model is someone who has inspired meto pursue my own passions and dreams, someone who has taught me the importance of dedication, humility, and generosity. They are a role model not just in their professional life, but also in their personal life, and their example has been a constant source of inspiration and guidance for me.As I continue to grow and develop in my own life, I strive to emulate the qualities and attributes of my model.I work hard to stay true to my passions and interests, and I always try to be approachable and generous with my time and knowledge. I know that with the help of my model's example, I can continue to grow and learn, and I amgrateful for the impact they have had on my life.In the end, we all need someone to look up to, someone who can inspire us and guide us towards our own success. My model has been that person for me, and I am forevergrateful for the example they have set. I hope that by sharing their story, I can inspire others to find their own models and to strive to emulate their qualities andattributes. After all, the power of a good model is immeasurable, and it can change lives for the better.。

Analyzing User Modeling on Twitter for Personalized News Recommendations

Analyzing User Modeling on Twitter forPersonalized News RecommendationsFabian Abel,Qi Gao,Geert-Jan Houben,and Ke TaoWeb Information Systems,Delft University of TechnologyMekelweg4,2628Delft,the Netherlands{f.abel,q.gao,g.j.p.m.houben,k.tao}@tudelft.nlAbstract.How can micro-blogging activities on Twitter be leveragedfor user modeling and personalization?In this paper we investigate thisquestion and introduce a framework for user modeling on Twitter whichenriches the semantics of Twitter messages(tweets)and identiﬁes top-ics and entities(e.g.persons,events,products)mentioned in tweets.Weanalyze how strategies for constructing hashtag-based,entity-based ortopic-based user proﬁles beneﬁt from semantic enrichment and explorethe temporal dynamics of those proﬁles.We further measure and com-pare the performance of the user modeling strategies in context of apersonalized news recommendation system.Our results reveal how se-mantic enrichment enhances the variety and quality of the generateduser proﬁles.Further,we see how the diﬀerent user modeling strategiesimpact personalization and discover that the consideration of temporalproﬁle patterns can improve recommendation quality.Keywords:user modeling,twitter,semantics,personalization.1IntroductionWith more than190million users and more than65million postings per day, Twitter is today the most prominent micro-blogging service available on the Web1.People publish short messages(tweets)about their everyday activities on Twitter and lately researchers investigate feasibility of applications such as trend analysis[1]or Twitter-based early warning systems[2].Most research initiatives study network structures and properties of the Twitter network[3,4].Yet,little research has been done on understanding the semantics of individual Twitter activities and inferring user interests from these activities.As tweets are limited to140characters,making sense of individual tweets and exploiting tweets for user modeling are non-trivial problems.In this paper we study how to leverage Twitter activities for user model-ing and evaluate the quality of user models in the context of recommending news articles.We develop a framework that enriches the semantics of individual Twitter activities and allows for the construction of diﬀerent types of semantic 1/2010/06/08/twitter-190-million-users/Joseph A.Konstan et al.(Eds.):UMAP2011,LNCS6787,pp.1–12,2011.c Springer-Verlag Berlin Heidelberg20112 F.Abel et al.user proﬁles.The characteristics of these user proﬁles are inﬂuenced by diﬀer-ent design dimensions and design alternatives.To better understand how those factors impact the characteristics and quality of the resulting user proﬁles,we conduct an in-depth analysis on a large Twitter dataset of more than2million tweets and answer research questions such as the following:how does the seman-tic enrichment impact the characteristics and quality of Twitter-based proﬁles (see Section4.2)?How do(diﬀerent types of)proﬁles evolve over time?Are there any characteristic temporal patterns(see Section4.3)?How do the diﬀer-ent user modeling strategies impact personalization(personalized news article recommendations)and does the consideration of temporal patterns improve the accuracy of the recommendations(see Section5)?Before studying the above research questions in Section4-5,we will summarize related work in Section2and introduce the design dimensions of Twitter-based user modeling as well as our Twitter user modeling framework in Section3.2Related WorkWith the launch of Twitter in2007,micro-blogging became highly popular and researchers started to investigate Twitter’s information propagation patterns[3] or analyzed structures of the Twitter network to identify inﬂuential users[4]. Dong et al.[5]exploit Twitter to detect and rank fresh URLs that have possibly not been indexed by Web search engines tely,Chen et al.conducted a study on recommending URLs posted in Twitter messages and compare strate-gies for selecting and ranking URLs by exploiting the social network of a user as well as the general popularity of the URLs in Twitter[6].Chen et al.do not investigate user modeling in detail,but represent Twitter messages of a user by means of a bag of words.In this paper we go beyond such representations and analyze diﬀerent types of proﬁles like entity-based or hashtag-based proﬁles.Laniado and Mika introduce metrics to describe the characteristics of hashtags –keywords starting with“#”–such as frequency,speciﬁcity or stability over time[7].Huang et al.further characterize the temporal dynamics of hashtags via statistical measures such as standard deviation and discover that some hashtags are used widely for a few days but then disappear quickly[8].Recent research on collaborativeﬁltering showed that the consideration of such temporal dy-namics impacts recommendation quality signiﬁcantly[9].However,the impact of temporal characteristics of Twitter-based user proﬁles on recommendation performance has not been researched yet.Neither hashtag-based nor bag-of-words representation explicitly specify the semantics of tweets.To better understand the semantics of Twitter messages published during scientiﬁc conferences,Rowe et al.[10]map tweets to conference talks and exploit metadata of the corresponding research papers to enrich the semantics of tweets.Rowe et al.mention user proﬁling as one of the applications that might beneﬁt from such semantics,but do not further investigate user modeling on Twitter.In this paper we close this gap and present theﬁrst large-scale study on user modeling based on Twitter activities and moreover explore how diﬀerent user models impact the accuracy of recommending news articles.Analyzing User Modeling on Twitter for Personalized News Recommendations3Table 1.Design space of Twitter-based user modeling strategiesdesign dimension design alternatives (discussed in this paper)proﬁle type (i)hashtag-based,(ii)topic-based or (iii)entity-basedenrichment(i)tweet-only-based enrichment or (ii)linkage and exploitation of external news articles (propagating entities/topics)temporal constraints(i)speciﬁc time period(s),(ii)temporal patterns (weekend ,night ,etc.)or (iii)no constraints3Twitter-Based User ModelingThe user modeling strategies proposed and discussed in this paper vary in three design dimensions:(i)the type of proﬁles created by the strategies,(ii)the data sources exploited to further enrich the Twitter-based proﬁles and (iii)temporal constraints that are considered when constructing the proﬁles (see Table 1).The generic model for proﬁles representing users is speciﬁed in Deﬁnition 1.Deﬁnition 1(User Proﬁle).The proﬁle of a user u ∈U is a set of weighted concepts where with respect to the given user u for a concept c ∈C its weight w (u,c )is computed by a certain function w .P (u )={(c,w (u,c ))|c ∈C,u ∈U }Here,C and U denote the set of concepts and users respectively.In particular,following Table 1we analyze three types of proﬁles that diﬀer with respect to the type of concepts C :entity-,topic-and hashtag-based pro-ﬁles –denoted by P E (u ),P T (u )and P H (u )respectively.We apply occurrence frequency as weighting scheme w (u,c ),which means that the weight of a con-cept is determined by the number of Twitter activities in which user u refers to concept c .For example,in a hashtag-based proﬁle w (u,#technology )=5means that u published ﬁve Twitter messages that mention “#technology”.We further normalize user proﬁles so that the sum of all weights in a proﬁle is equal to 1: c i ∈C w (u,c i )=1.With p (u )we refer to P (u )in its vector space model representation,where the value of the i -th dimension refers to w (u,c i ).The user modeling strategies we analyze in this paper exploit Twitter mes-sages posted by a user u to construct the corresponding proﬁle P (u ).When constructing entity-and topic-based user proﬁles,we also investigate the impact of further enrichment based on the exploitation of external data sources (see Table 1).In particular,we allow for enrichment with entities and topics ex-tracted from news articles that are linked with Twitter messages (news-based enrichment).In previous work [11]we presented strategies for selecting appro-priate news articles for enriching users’Twitter activities.A third dimension we investigate in the context of Twitter-based user model-ing is given by temporal constraints that are considered when constructing the4 F.Abel et al.proﬁles(see Table1).First,we study the nature of user proﬁles created within speciﬁc time periods.For example,we compare proﬁles constructed by exploiting the complete(long-term)user history with proﬁles that are based only on Twit-ter messages published within a certain week(short-term).Second,we examine certain time frames for creating the proﬁles.For example,we explore the diﬀer-ences between user proﬁles created on the weekends with those created during the week to detect temporal patterns that might help to improve personalization within certain time frames.By selecting and combining the diﬀerent design dimensions and alternatives we obtain a variety of diﬀerent user modeling strategies that will be analyzed and evaluated in this paper.3.1Twitter-Based User Modeling FrameworkWe implemented the proﬁling strategies as a Twitter-based user modeling frame-work that is available via the supporting website of this paper[12].Our frame-work features three main components:1.Semantic Enrichment.Given the content of Twitter messages we extractentities and topics to better understand the semantics of Twitter activities.Therefore we utilize OpenCalais2,which allows for the detection and identi-ﬁcation of39diﬀerent types of entities such as persons,events,products or music groups and moreover provides unique URIs for identiﬁed entities as well as for the topics so that the meaning of such concepts is well deﬁned.2.Linkage.We implemented several strategies that link tweets with externalWeb resources and news articles in particular.Entities and topics extracted from the articles are then propagated to the linked tweets.In[11]we showed that for tweets which do not contain any hyperlink the linking strategies identify related news articles with an accuracy of70-80%.er Modeling.Based on the semantic enrichment and the linkage with ex-ternal news articles,our framework provides methods for generating hashtag-based,entity-based,and topic-based proﬁles that might adhere to speciﬁc temporal constraints(see above).4Analysis of Twitter-Based User ProﬁlesTo understand how the diﬀerent user modeling design choices inﬂuence the char-acteristics of the generated user proﬁles,we applied our framework to conduct an in-depth analysis on a large Twitter dataset.The main research questions to be answered in this analysis can be summarized as follows.1.How do the diﬀerent user modeling strategies impact the characteristics ofTwitter-based user proﬁles?2.Which temporal characteristics do Twitter-based user proﬁles feature?2Analyzing User Modeling on Twitter for Personalized News Recommendations5types of proﬁlesparison between diﬀerent user modeling strategies with tweet-only-based or news-based enrichment4.1Data Collection and Data Set CharacteristicsOver a period of more than two months we crawled Twitter information streams of more than20,000users.Together,these people published more than10mil-lion tweets.To allow for linkage of tweets with news articles we also monitored more than60RSS feeds of prominent news media such as BBC,CNN or New York Times and aggregated the content of77,544news articles.The number of Twitter messages posted per user follows a power-law distribution.The ma-jority of users published less than100messages during our observation period while only a small fraction of users wrote more than10,000Twitter messages and one user produced even slightly more than20,000tweets(no spam).As we were interested in analyzing also temporal characteristics of the user proﬁles,we created a sample of1619users,who contributed at least20tweets in total and at least one tweet in each month of our observation period.This sample dataset contained2,316,204tweets in total.We processed each Twitter message and each news article via the semantic enrichment component of our user modeling framework to identify topics and entities mentioned in the the tweets and articles(see Section3.1).Further,we applied two diﬀerent linking strategies and connected458,566Twitter messages with news articles of which98,189relations were explicitly given in the tweets by URLs that pointed to the corresponding news article.The remaining360,377 relations were obtained by comparing the entities that were mentioned in both news articles and tweets as well as by comparing the timestamps.In previous work we showed that this method correlates news and tweets with an accuracy of more than70%[11].Our hypothesis is that–regardless whether this enrichment method might introduce a certain degree of noise–it impacts the quality of user modeling and personalization positively.4.2Structural Analysis of Twitter-Based ProﬁlesTo validate our hypothesis and explore how the exploitation of linked external sources inﬂuences the characteristics of the proﬁles generated by the diﬀerent user modeling strategies,we analyzed the corresponding proﬁles of the1619users6 F.Abel et al.from our sample.In Figure1we plot the number of distinct(types of)concepts in the topic-and entity-based proﬁles and show how this number is inﬂuenced by the additional news-based enrichment.For both types of proﬁles the enrichment with entities and topics obtained from linked news articles results in a higher number of distinct concepts per pro-ﬁle(see Fig.1(a)and1(b)).Topic-based proﬁles abstract much stronger from the concrete Twitter activities than entity-based proﬁles.In our analysis we utilized the OpenCalais taxonomy consisting of18topics such as politics,entertainment or culture.The tweet-only-based user modeling strategy,which exploits merely the semantics attached to tweets,fails to create proﬁles for nearly100users(6.2%,topic-based)as for these users none of the tweets can be categorized intoa topic.By enriching the tweets with topics inferred from the linked news articles we better understand the semantics of Twitter messages and succeed in creating more valuable topic-based proﬁles for99.4%of the users.Further,the number of proﬁle facets,i.e.the type of entities(e.g.person, location or event)that occur in the entity-based proﬁles,increases with the news-based semantic enrichment.While more than400twitter-based proﬁles(more than25%)feature less than10proﬁle facets and often miss entities such as movies or products a user is concerned with,the news-based enrichment detects a greater variety of entity types.For more than99%of the entity-based proﬁles enriched via news articles,the number of distinct proﬁle facets is higher than10.A comparison of the entity-and topic-based user modeling strategies with the hashtag-based strategy(see Fig.1(c))shows that the variety of entity-based pro-ﬁles is much higher than the one of hashtag-based proﬁles.While the entity-based strategy succeeds to create proﬁles for all users in our dataset,the hashtag-based approach fails for approximately90users(5.5%)as the corresponding people neither made use of hashtags nor re-tweeted messages that contain hashtags. Entity-based as well as topic-based proﬁles moreover make the semantics more explicit than hashtag-based proﬁles.Each entity and topic has a URI which deﬁnes the meaning of the entity and topic respectively.The advantages of well-deﬁned semantics as exposed by the topic-and entity-based proﬁles also depend on the application context,in which these proﬁles are used.The results of the quantitative analysis depicted in Fig.1show that entity-and topic-based strategies allow for higher coverage regarding the number of users,for whom proﬁles can be generated,than the hashtag-based strategy. Further,semantic enrichment by exploiting news articles(implicitly)linked with tweets increases the number of entities and topics available in the proﬁles signif-icantly and improves the variety of the proﬁles(the number of proﬁle facets).4.3Temporal Analysis of Twitter-Based ProﬁlesIn the temporal analysis we investigate(1)how the diﬀerent types of user pro-ﬁles evolve over time and(2)which temporal patterns occur in the proﬁles. Regarding temporal patterns we,for example,examine whether proﬁles gen-erated on the weekends diﬀer from those generated during the week.SimilarAnalyzing User Modeling on Twitter for Personalized News Recommendations7Fig.2.Temporal evolution of user proﬁles:average d 1-distance of current individual user proﬁles with corresponding proﬁles in the pastto the click-behavior analysis by Liu et al.[13],we apply the so-called d 1-distance for measuring the diﬀerence between proﬁles in vector representation:d 1(p x (u ),p y (u ))= i |p x,i −p y,i |.The higher d 1(p x (u ),p y (u ))∈[0..2]the higher the diﬀerence of the two pro-ﬁles p x (u )and p y (u )and if two proﬁles are the same then d 1(p x (u ),p y (u ))=0.Figure 2depicts the evolution of proﬁles over time.It shows the average d 1-distance of the current user proﬁles with the proﬁles of the same users created based on Twitter activities performed in a certain week in the past.As suggested in [13],we also plotted the distance of the current user-speciﬁc proﬁle with the public trend (see Fig.2(a)),i.e.the average proﬁle of the corresponding weeks.For the three diﬀerent proﬁle types we observe that the d 1-distance slightly decreases over time.For example,the diﬀerence of current proﬁles (ﬁrst week of January 2011)with the corresponding proﬁles generated at the beginning of our observation period (in the week around 18th November 2010)is the highest while the distance of current proﬁles with proﬁles computed one week before (30th December 2010)is the lowest.It is interesting to see that the distance of the current proﬁles with the public trend (i)is present for all types of proﬁles and (ii)is rather constant over time.This suggests (i)a certain degree of individualism in Twitter and (ii)reveals that the people in our sample follow diﬀerent trends rather than being inﬂuenced by the same trends.Hashtag-based proﬁles exhibit the strongest changes over time as the aver-age d 1-distance to the current proﬁle is constantly higher than for the topic-and entity-based proﬁles.Figure 2(b)discloses that entity-based proﬁles change stronger over time than topic-based proﬁles when news-based enrichment is en-abled.When merely analyzing Twitter messages one would come to a diﬀerent (possibly wrong)conclusion (see Fig.2(a)).Figure 3illustrates temporal patterns we detected when analyzing the indi-vidual user proﬁles.In particular,we investigate how proﬁles created on the weekends diﬀer from proﬁles (of the same user)created during the week.For topic-based proﬁles generated solely based on Twitter messages,it seems that for some users the weekend and weekday proﬁles diﬀer just slightly while for 24.9%of the users the d 1-distance of the weekend and weekday proﬁle is maximal (2is the maximum possible value,see Fig.3(a)).The news-based enrichment reveals8 F.Abel et al.arbitrarily chosen daysence betweenday/night diﬀerenceFig.3.Temporal patterns:comparison between weekend and weekday proﬁles by means of d1-distance((a)-(c):topic-based proﬁles)however that the diﬀerence of weekend and weekday proﬁles is a rather common phenomenon:the curve draws nearer to the average diﬀerence(see dotted line); there are less extrema,ers for whom the d1-diﬀerence is either very low or very high.Hence,it rather seems that the tweets alone are not suﬃcient to get a clear understanding of the users concerns and interests.Fig.3(b)further supports the hypothesis that weekend proﬁles diﬀer sig-niﬁcantly from weekday proﬁles.The corresponding distances d1(p weekend(u), p weekday(u))are consistently higher than the diﬀerences of proﬁles generated on arbitrarily chosen days during the week.This weekend pattern is more signiﬁ-cant than diﬀerences between topic-based proﬁles generated based on Twitter messages that are either posted during the evening(6pm-3am)or during the day (9am-5pm)as shown in Fig.3(c).Hence,the individual topic drift–i.e.change of topics individual users are concerned with–between day and evening/night seems to be smaller than between weekdays and weekends.The weekend pattern is coherent over the diﬀerent types of proﬁles.Diﬀer-ent proﬁle types however imply diﬀerent drift of interests or concerns between weekend and weekdays(see Fig.3(d)).Hashtag-based and entity-based proﬁles change most while the types of entities people refer to(persons,products,etc.)Analyzing User Modeling on Twitter for Personalized News Recommendations9 do not diﬀer that strongly.When zooming into the individual entity-based pro-ﬁles we see that entities related to leisure time and entertainment become more important on the weekends.The temporal analysis thus revealed two important observations.First,user proﬁles change over time:the older a proﬁle the more it diﬀers from the current proﬁle of the user.The actual proﬁle distance varies between the diﬀerent types of proﬁles.Second,weekend proﬁles diﬀer signiﬁcantly from weekday proﬁles. 5Exploitation of User Proﬁles for Personalized News RecommendationsIn this section,we investigate the impact of the diﬀerent user modeling strategies on recommending news articles:1.To which degree are the proﬁles created by the diﬀerent user modeling strate-gies appropriate for recommending news?2.Can the identiﬁed(temporal)patterns be applied to improve recommenda-tion accuracy?5.1News Recommender System and Evaluation Methodology Recommending news articles is a non-trivial task as the news items,which are going to be recommended,are new by its very nature,which makes it diﬃcult to apply collaborativeﬁltering methods,but rather calls for content-based or hybrid approaches[13].Our main goal is to analyze and compare the applicability of the diﬀerent user modeling strategies in the context of news recommendations.We do not aim to optimize recommendation quality,but are interested in comparing the quality achieved by the same recommendation algorithm when inputting diﬀerent types of user proﬁles.Therefore we apply a lightweight content-based algorithm that recommends items according to their cosine similarity with a given user proﬁle.We thus cast the recommendation problem into a search and ranking problem where the given user proﬁle,which is constructed by a speciﬁc user modeling strategy,is interpreted as query.Deﬁnition2(Recommendation Algorithm).Given a user proﬁle vector p(u)and a set of candidate news items N={p(n1),...,p(n n)},which are rep-resented via proﬁles using the same vector representation,the recommendation algorithm ranks the candidate items according to their cosine similarity to p(u).Given the Twitter and news media dataset described in Section4.1,we consid-ered the last week of our observation period as the time frame for computing recommendations.The ground truth of news articles,which we consider as rel-evant for a speciﬁc user u,is obtained via the Twitter messages(including re-tweets)posted by u in this week that explicitly link to a news article published by BBC,CNN or New York Times.We thereby identiﬁed,on average,5.5relevant news articles for each of the1619users from our sample.For less than10%of the10 F.Abel et al.Fig.4.Results of news recommendation experimentusers we found more than20relevant articles.The candidate set of news ar-ticles,which were published within the recommendation time frame,contained 5529items.We then applied the diﬀerent user modeling strategies together with the above algorithm(see Def.2)and set of candidate items to compute news recommendations for each user.The user modeling strategies were only allowed to exploit tweets published before the recommendation period.The quality of the recommendations was measured by means of MRR(Mean Reciprocal Rank), which indicates at which rank theﬁrst item relevant to the user occurs on aver-age,and S@k(Success at rank k),which stands for the mean probability that a relevant item occurs within the top k of the ranking.In particular,we will focus on S@10as our recommendation system will list10recommended news articles to a user.We tested statistical signiﬁcance of our results with a two-tailed t-Test where the signiﬁcance level was set toα=0.01unless otherwise noted.5.2ResultsThe results of the news recommendation experiment are summarized in Fig.4 and validateﬁndings of our analysis presented in Section4.Entity-based user modeling(with news-based enrichment),which produces according to the quan-titative analysis(see Fig.1)the most valuable proﬁles,allowed for the best recommendation quality and performed signiﬁcantly better than hashtag-based user modeling(see Fig.4(a)).Topic-based user modeling also performed better than the hashtag-based strategy–regarding S@10the performance diﬀerence is signiﬁcant.Since the topic-based strategy models user interests within a space of 18diﬀerent topics(e.g.,politics or sports),it further required much less run-time and memory for computing user proﬁles and recommendations than the hashtag-and entity-based strategies,for which we limited dimensions to the10,0000most prominent hashtags and entities respectively.Further enrichment of topic-and entity-based proﬁles with topics and entities extracted from linked news articles,which results in proﬁles that feature moreAnalyzing User Modeling on Twitter for Personalized News Recommendations11 facets and information about users’concerns(cf.Section4.2),also results in a higher recommendation quality(see Fig.4(b)).Exploiting both tweets and linked news articles for creating user proﬁles improves MRR signiﬁcantly(α=0.05). In Section4.3we observed that user proﬁles change over time and that recent proﬁle information approximates future proﬁles slightly better than old proﬁle information.We thus compared strategies that exploited just recent Twitter activities(two weeks before the recommendation period)with the strategies that exploit the entire user history(see Fig.4(c)).For the topic-based strategy we see that fresh user proﬁles are more applicable for recommending news articles than proﬁles that were built based on the entire user history.However,entity-based user modeling enables better recommendation quality when the complete user history is applied.Results of additional experiments[12]suggest that this is due to the number of distinct entities that occur in entity-based proﬁles(cf.Fig.1): long-term proﬁles seem to reﬁne preferences regarding entities(e.g.persons or events)better than short-term proﬁles.In Section4.3we further observed the so-called weekend pattern,er proﬁles created based on Twitter messages published on the weekends signiﬁ-cantly diﬀer from proﬁles created during the week.To examine the impact of this pattern on the accuracy of the recommendations we focused on recommending news articles during the weekend and compared the performance of user proﬁles created just by exploiting weekend activities with proﬁles created based on the complete set of Twitter activities(see Fig.4(d)).Similarly to Fig.4(c)we see again that the entity-based strategy performs better when exploiting the entire user history while the topic-based strategy beneﬁts from considering the week-end pattern.For the topic-based strategy recommendation quality with respect to MRR improves signiﬁcantly when proﬁles from the weekend are applied to make recommendations during the weekend.6ConclusionsIn this paper we developed a user modeling framework for Twitter and inves-tigated how the diﬀerent design alternatives inﬂuence the characteristics of the generated user proﬁles.Given a large dataset consisting of more than2mil-lion tweets we created user proﬁles and revealed several advantages of semantic entity-and topic-based user modeling strategies,which exploit the full function-ality of our framework,over hashtag-based user modeling.We saw that further enrichment with semantics extracted from news articles,which we correlated with the users’Twitter activities,enhanced the variety of the constructed pro-ﬁles and improved accuracy of news article recommendations signiﬁcantly.Further,we analyzed the temporal dynamics of the diﬀerent types of proﬁles. We observed how proﬁles change over time and discovered temporal patterns such as characteristic diﬀerences between weekend and weekday proﬁles.We also showed that the consideration of such temporal characteristics is beneﬁcial to recommending news articles when dealing with topic-based proﬁles while for entity-based proﬁles we achieve better performance when incorporating the en-tire user history.In future work,we will further research the temporal speciﬁcs。

你做过什么模型吗英语作文5句话

你做过什么模型吗英语作文5句话English: I have worked on various models in my academic and professional career. In my academic studies, I have created statistical models to analyze data and predict outcomes for research projects.In my professional career, I have developed financial models to assess investment opportunities and forecast business performance. Additionally, I have also worked on predictive models for machine learning applications, which have involved building and testing algorithms to make data-driven predictions. Overall, my experience with different types of models has allowed me to gain valuable insights and skills in data analysis and modeling.中文翻译: 我在学术和职业生涯中曾经涉及过各种模型。

在我的学术研究中，我曾创建统计模型来分析数据并预测研究项目的结果。

在我的职业生涯中，我开发了金融模型来评估投资机会并预测业务绩效。

另外，我还参与了机器学习应用的预测模型工作，包括构建和测试算法以进行数据驱动的预测。

总的来说，我对不同类型的模型有了丰富经验，这使我获得了在数据分析和建模方面宝贵的见解和技能。

1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。

User Modeling for Building Personalized Web AssistantsZheng Chen1Liu Wenyin1Fan Lin2Peng Xiao2 Liu Yin1Bin Lin1Wei-Ying Ma11Microsoft Research Asia, 49 Zhichun Road, Beijing 100080, P.R. China{zhengc, wyliu, binlin, wyma}@2Dept. Computer of Sci. and Tech., Tsinghua Univ., Beijing 100084, P.R.Chinalfan@, xiaopeng@ABSTRACTWe proposed several new approaches to extracting features from the user visited web pages for modeling the user’s long- and short-term preferences and interests. The constructed user models are used to build personalized web assistants which help the user navigate the Web by providing interest summarization, related pages suggestion, and highlighting important content in a web page. Our experimental results show that the proposed techniques improve the user’s web browsing experiences significantly.1.IntroductionThe amount of information that a single user could access has been rapidly increasing with the great facility of the World Wide Web. However, due to low search precision, the performance of most popular search engines is unsatisfactory. In this paper, we present a new method to help the Internet users search their visited information and the web pages potentially interesting to them. Besides the explicit features directly extracted from the contents of the visited pages, several implicit features are also computed based on the user’s navigation paths. In certain situations, some of the semantics of the web pages in the beginning of the navigation path can be inherited by the following web pages in a short session. We refer to such implicit semantic features as inherited features.To test the effectiveness of our proposed methods, we have implemented a personal WWW assistant system. Through the user’s navigation paths, the inherited semantic features are extracted and combined with other explicit semantic features extracted from the web pages for modeling the user’s long-term and short-term interests. The user models are then used to build three personalized assistants (interesting information summarizer, related pages advisor, and interesting part highlighter) to provide the user better web navigation experience.2.Explicit and Implicit Features of Web PagesIn this section, we present how to extract and utilize explicit features and implicit features to represent an end-user’s interests and intentions. Explicit features are the features that can be directly extracted from the content of visited information. We use empirical rules similar to Zheng et al. (2001) [1] to compute explicit features from the sources including filenames, URLs, alternative text, surrounding text, and anchor text. And the implicit features are extracted from the user’s navigation paths.The following example demonstrates why implicit features are more illustrative than explicit features. When the user needs to find the information, he/she may submit a query to a search engine. He/she then follows several hyperlinks, starting from one of the search results and reaching the final target web-page which contains desired information. Frequently we found that the query words and the hypertexts over the hyperlinks are good resource for representing the meaning of the final target web-page, though the content of the target web-page may not always contains the query words. In this case, the target page could not be found if only using explicit features. But the query words, which are inherited by the target page on the navigation path, are implicit but highly related to the semantics of the page. A semantic hierarchy can be built based on the navigation pattern. For example, a user maygo to to find information about the Sony digital camera. He may follow a sequence of hyperlinks (“digital cameras”), /digitalimaging/cybershot.htm (“cyber-shot”), and /digitalimaging/cybershot.htm(“DSC-P50”) to get the information about Sony DSC-P50, the target page containing desired information. In this case, not only “DSC-P50” can be used to index the final target web-page, but “digital camera” and “cyber-shot” are also relevant text features to index the page. These two keywords can be inherited by their subsequent pages on the navigation paths. This is the reason why this kind of implicit features is called “inherited features”. These kinds of features not only are semantic extension of the target web-page, but also represent the semantic hierarchy of the web-pages on the navigation paths.In order to extract the “inherited features” from the navigation paths, one important task is to determine whether a web-page can inherit the features from its precedent pages or not. For this, we have developed two methods. One is based on the structure of the website; the other is based on the content similarity between the visited web-pages.3.ExperimentsWe have performed several experiments to test the effectiveness of the three assistants and the importance of inherited features in modeling the user’s interests. In order to extract implicit features from the user’s behavior on the Web, we developed a capture tool to automatically collect the user’s actions in the Microsoft Internet Explorer (IE) environment. The five common user’s actions are recorded: browse, click, query, save, and open/close. The actions of five users’ web surfing over a week were collected, which contain approximately 1,000 web-pages.The system is configured into two levels: The first level is the baseline, which only employs explicit features; the second level is the full system, which uses both explicit features and implicit features. After the week of collecting data from the five users, we asked them to search their visited web-pages using keywords on our system. From Figure 1, we can see that the precision of the full system significantly outperforms the baseline system.Figure 1. Comparison of the implicit features and explicit features4.ConclusionIn this paper we described several new approaches to extracting features from the user’s visited web pages for modeling his preferences. In particular, the implicit features associated with how the user interacts with the information were used to improve the performance. The concept of feature inheritance was introduced to associate relevant features between web pages on the navigation paths. Our experimental results have indicated that the precision of the explicit feature and implicate features are significantly outperforms the pure explicit features. Reference[1]Zheng Chen, Liu Wenyin, Feng Zhang, Minjing Li, Hongjiang Zhang. Web Mining for Web Image Retrieval.In: Journal of the American Society for Information Science and Technology, Vol. 52, No. 10, pp. 831-839, August, 2001.。