行政院国家科学委员会专题研究计划成果报告

合集下载

1行政院国家科学委员会专题研究计划成果报告人体手势之运动分析与辨识

¦æ¬F°|°ê®a¬ì¾Ç©e-û·|±MÃD¬ã¨s-p¹º¦¨ªG³ø§i¤HÅé¤â¶Õ¤§¹B°Ê¤ÀªR»P¿ëÃÑEnergy Spectrum Based HMM for Gesture Recognition -p¹º½s¸¹¡G NSC89-2218-E-011-012°õ¦æ´Á--¡G89¦~8¤ë1¤é¦Ü90¦~7¤ë31¤é¥D«ù¤H¡G³\·s²K±Ð±Â°ê¥ß¥xÆW¬ì§Þ¤j¾Ç¹q¾÷¨t-p¹º°Ñ»P¤H-û¡G·¨¼w¶³°ê¥ß¥xÆW¬ì§Þ¤j¾Ç¹q¾÷¨tAbstractVision based gesture recognition has been a new way of communication between machine and human being. In this paper, we develop an energy spectrum based HMM for both gesture representation and recognition. A symbol sequence is derived from the gesture via vector quantization. The Baum-Welch algorithm is employed to establish the Hidden Markov Model (HMM) during the training phase. The probability of each model producing the symbol sequence associated with the input gesture is computed and the one with maximum probability is recognized.Experiment shows that the recognition rate of the developed energy spectrum based approach outperformed the Fourier descriptor based method as expected.Keywords:gesture recognition, energy spectrum based HMM, vector quantization.I.IntroductionWith the development of the computer technology, it is now possible to take alternative ways to communicate with the computer in addition to the traditional keyboard and mouse. Speech recognition and gesture recognition are the most actively studied approaches [1].Both glove [2, 3] and vision based [4, 5] methods was conducted for gesture recognition. While the former is more precise in the positioning of the 3D information, the latter is more friendly hence is more practical, especially when the machine to be communicated is far from or isolated in an environment.Takahashi and Kishino used data gloves to establish the hand shape code that describe the shape, orientation, and motion of the hand [2]. Cui and Weng combined the action and the shape of hands and extracted the feature vectors for recognition with moderate recognition rate in extremely long time [6]. Liang and Ouhyoung used data gloves to fetch the stretch of the 10 joints of a hand, the orientation of the palm and the motion vector of the hands [3]. Starner and Pentland developed a vision based method with the shape, orientation and trajectory of hands and used a feature vector with only eight elements for HMM [5]. Chen, Huang and Juang applied the hand shape variation and the motion trajectory from 2D images for HMM [4] In this research, the vision based Hidden Markov Model (HMM) will be employed for both hand gesture representation and recognition. Vector quantization will be taken to reduce the size of codebook and the energy spectrum of the symbol sequence will be used as the input for the HMM. In this paper, the features extraction associated with a gesture and the HMM that use energy spectrum as input for training is developed in Section II. Experiment results and discussions are given in Sections III and IV, respectively. II.Energy spectrum based HMMThe selection of features and the application of these feature as input to train the HMM are described.Feature selection is important in both the representation and recognition of gestures. A good type of feature has to satisfy the RST invariance, the ease of computation and high recognition rate. Energy spectrum derived from Fourier descriptor (FD) is such a feature. Instead of using the FDs directly as in [4], we use the energy spectrum to stress the more significant Fourier coefficients to improve the recognition rate. This has been proved to be quite effectivein the experiment: the overall recognition rate has been improved from 94% to 98%.For the recognized ten gestures, twenty training samples used for each gesture and thirty images for each sample, we have a huge size of 6,000 samples. Vector quantization is used to reduce the size and the computation time for recognition. The size of the codebook is selected to be 64 that compromises between the execution time and the average quantization error.HMM is a good vehicle to describe the gesture, in which both state and observation stochastic processes are involved. Ergodic discrete HMM is employed to characterize the evolution of the gesture thoroughly.An HMM λ is usually expressed asλπ=(,,)A B (1)where π is the initial probability vector, A a ij =[] is the state transition matrix and B b ij =[] is the observation matrix, respectively.The energy spectrum obtained from the FDs is employed to train the HMMs to obtain the parameters A and B for each gesture. For the observation sequence of 30 images per gesture and 10 gestures to be tested, the dimensions of A and B are 1010× and 1030×, respectively. Baum-Welch algorithm is employed for training , which is a backward procedure that introduces a backward variable βt i (), two probabilityvariablesγt i () and ξt i j (,), and iteratively update the estimated parameters to achieve the maximum likelihood of P O ()λfor the given observation sequence O whereβλt t t T t i P O O O q i ()(,,,)==++12L , (2)γλt t i P q i O ()(,)== (3)ξλt t t i j P q i q j O (,)(,,)===+1 (4)for the state q and the gesture i, j at time t .Then the parameters areπγi t i =() (5)a i j i ij t t T t t T ==−=−∑∑ξγ(,)()1111 (6)b j j jk t O ktTt t Tt====∑∑γγ()()11(7)Experiments are then conducted based on thisframework.III.Experiment resultsTen gestures, including hand shape change, rotation, and translation are tested. An I-cam USB camera with 160120× at 30 frames per second is employed to fetch the gesture image and a PC Pentium III-500 with 128MB RAM is used to do both training and recognition. The gestures tested are shown in Fig.1 and the total time spent for recognition is 6.228 seconds, including preprocessing time 6.12 seconds. That is, the time used for vector quantization and recognition is only 0.108 second. Some typical recognition results are shown in Table 1 in which the values are calculated using the formula log ()10P O i λ for the given P O i ()λ. Same approach has been taken except that FDs are used directly. The recognition rates for both FDs and the proposed energy spectrum with the 10 gestures are compared as shown in Table 2.IV.Discussion and conclusionAn energy spectrum based HMM for gesture recognition has been developed. Vector quantization was employed to reduce the data size and Baum-Welch algorithm was employed for training. Ten gestures, including hand shape change, rotation, and translation were tested. Experiments showed that the energy spectrum based HMM improved the recognition rate from 94% with FD based to 98%. This is attributed to the stress of the more significant coefficients using the energy spectrum based HMM. The recognition rate of gesture 2 is as low as 93.3%. Comparing gestures 2 and 10, one can find that the beginning two fifth of the two gestures are almost the same. This explains why gesture 2 is taken as gesture 10. However, the situation is not reflective. That is, gesture 10 is not taken as gesture 2. This is because the symbol sequence generated by gesture 2 remains almost unchanged since the hand shape remains unchanged in gesture 2 thus the probability of mistakenly taking the shape varying gesture 10 as gesture 2 is low.Fig.1 Gestures under test. (From top to bottom: gestures 1 to 10)Table 1 Recognition results using energy spectrum based HMMTable 2 Recognition rate using the FD and the energy spectrum approachesReferences[1] Vladimir I. Pavlovic, Rajeev Sharma andThomas S. Huang, “Visual interpretation of hand gesture for human-computer interaction: A review”, IEEE Trans. on Pattern Analysis and Machine Intelligence, vol.19, no.7, pp.677~pp.694, July, 1997. [2] T.Takahashi and F.Kishino, “A hand gesturerecognition method and its application”, Systems and Computers in Japan, V ol. 23, No.3, pp.38-48, 1992.[3] Liang, R. H. and M, Ouhyoung, “A real-timecontinuous gesture recognition system for sign language”, In Proc. 2 nd Int. Conf. on Automatic Face and Gesture recognition,pp.558-565, 1998.and Gia-Ying Juang, “Gesture recognition using hidden Markov models”, Conference on Computer V ision ,Graphics and Image Processing, pp.477~pp.484, 1999.[5] T. Starner and A. Pentland, “Real-timeAmerican Sign Language recognition form video using hidden Markov models”, Proc.of International Symposium on Computer V ision, pp.265-270, 1995.[6] Cui, Y., J.J. Weng, “Hand sign recognitionfrom intensity image sequences with complex background”, In Proc. IEEE 2nd Int. Conf. on Automatic Face and Gesture Recognition, pp.259 –264, 1996.。

行政院国家科学委员会专题研究计划成果报告

¦æ¬F°|°ê®a¬ì¾Ç©e-û·|±MÃD¬ã¨s-p¹º¦¨ªG³ø§i¥xÆW¨ÅÀù¯f²z¦¨¦]¤§¬ã¨s-§íÀù°ò¦]¦b¨ÅÀù§Î¦¨¤¤¤§-«-n©Ê (²Ä¤T¦~)-p¹º½s¸¹: NSC 88-2314-B-002-365°õ¦æ¦~--: 87¦~08¤ë1¤é¦Ü88¦~7¤ë31¤é¥D«ù¤H : ³\ª÷¥É¥x¤jÂå¾Ç°|¥Í¤Æ©Ò¤¤¤åºK-nÃöÁäµü¡G¨ÅÀù/§íÀù°ò¦]p53,TSG101³\¦hªº¬ã¨s³ø§iÅã¥Ü¦b¨ÅÀù²Ó-M¤¤¦³«Ü°ª¤ñ¨Òªº¬V¦âÅé11p15¤§LOH(Loss of Heterozygosity )ÅÜ²§¡A Åã¥Ü¦¹°Ï°ì¤§°ò¦]ÅÜ²§»P¨ÅÀù¤§§Î¦¨¦³Ãö¡F¦ÓTSG101¬O 1997¦~¤~µo ²{¤§§íÀù°ò¦]¡A ¦¹°ò¦]´N ¦ì©ó11p15¡A¦Ó¥B³Ìªñªº³ø¾É«ü¥X ¡A¨ÅÀù²Ó-M¤¤½T¹ê·|§t¦³¦¹°ò¦]¤£¥¿±`ªºmRNA ªí²{¡A ¦]¦¹¦b³o-Ó-pµeùØ¡A §Ú-Ì¥Î¨Ó¦Û¥x¤jÂå°|ªº¤@§å¨ÅÀùÀËÅé°µTSG101ªº¤ÀªR (Table 1)¡A¤S¥Ñ©ó³o§åÀËÅé¡A§Ú-Ì¤w¸g¤ÀªR¹L¨äER¡A erbB2(HER2/neu)¤Îp53ªºªí²{±¡§Î(Table2, 3)¡A±N¨Ó§Ú-Ì¥i¥H¦P®É©Î¤À¶}¨Ó¤ÀªR TSG101¤Î³o¨Ç°ò¦]»PÁ{§É¯fª¬ªº¬ÛÃö©Ê¡A³o±N´£¨Ñ¤@-Ó§ó-Ñ¯S²§©Êªºgenetic markers ¨ÓÀ°§U¤F¸Ñ¦¹Àù¯g¥i¯àªº-P¯f¾÷Âà¡A ¦Ó¥B¥i¥H´£¨Ñ±N¨Ó¬ã¨s T SG101ªº¥Í²z ¥\¯àªº¤è¦V ¡A ¥H¤Î³o ¨Ç°ò¦]¬Û¤¬¶¡ªº§@¥Î¡C -^¤åºK-nKeywords¡G Tumor Suppressor Gene p53 and TSG101/Breast CancerLoss of heterozygosity (LOH) on chromosome 11p15 occurs frequently in breast cancer indicating that this region may have a role in the pathogenesis of breast cancer. TSG101 was identified as a tumor susceptibility gene by homozygous functional inactivation of allelic loci in mouse 3T3fibroblasts.The human homologue of this gene was then isolated and mapped to 11p15.Moreover, abnormalities of TSG101transcripts in human breast cancer and prostate cancer from western country have been reported recently. To determine whether abnormal TSG101 expression has correlation with unique characteristics of breast cancer from Taiwan, TSG101 gene status of breast cancer specimens from National Taiwan University Hospital (NTUH) and several breast cancer cell lines have been studied by reverse transcription-polymerase chain reaction (RT-PCR ) and directly sequencing the cDNA of this gene (Table 1). We haveaccomplished the study about variants ofestrogen receptor (ER), over expression of erbB2(HER2/neu) and the mutation of p53 in same pool of breast cancer specimens mentioned above for the past two years.Those genetic alterations do provide someclues about breast carcinogenesis (Table 2, 3).More genetic analysis will help to developeffective genetic markers for this cancer’sprogression. Therefore, our specific aims ofthis proposed study are the following.Screening a large pool of breast cancer samples with known clinical stages for mutation in the TSG101 gene. The status of ER, erbB2(HER2/neu), p53 and TSG101 in the tumor samples will be correlated with their clinical stages. Moreover, any positivecorrelation between phenotype and mutation would provide scientifically importantdirection to explore inter-relationshipbetween these genes.Background TSG101 was identified as a tumor susceptibility gene by homozygous functional inactivation of allelic loci in mouse 3T3 fibroblasts. The human homologous of this gene was then isolated and mapped to 11P15(6,8). Some previous reports indicate that the major function of TSG101 is the following. TSG101 proteins of putative DNA-binding and transcriptional activation domains, it has suggested that the 43kD TSG101 protein may act to control gene expression and regulate the cell cycle.On the other hand, the coiled-coil domain of TSG101 can interact with stathmin. This finding suggests that this gene may control cell growth and differentiation (4,9,12,14,15,16). Other reports indicate thatTSG101 transcripts are frequently abnormal in human cancer cell, including breast cancer (1,2,3,13,17), prostate cancer (11), and leukemia (5). The major type of abnormality of TSG101 is aberrant splicing but not mutations (7,10). The relaxation of RNA splicing fidelity of TSG101 may be an oncodevelopment marker in cancer. We,therefore, plan to screen the status of TSG101 in a large pool of breast cancer samples that has been analyzed for ER,erbB2(HER2/neu) and p53 alterations in my laboratory. Clinical data including the pathologic stage, histologic type and follow-up will be collected. Then, the statistical analysis will be use to clarity the association between genetic alterations of ER,erbB2(HER2/neu), p53 as well as TSG101and clinical outcome.ResultTo detect the aberrant transcripts of TSG101 gene in breast cancer as well as the normal counterpart.According to the structure of TSG101 gene,two pairs of primer (p1/p2 and p3/p4) have been used to analyze any of the TSG101truncated transcripts by RT-PCR. Further restriction enzyme mapping or sequencing of RT-PCR products have been performed to dissect those truncated transcripts as well.The results were listed in table 3.DiscussionIn this preliminary study the abnormal TSG101 transcripts have been detected in tissues of breast cancers but not normal counterparts. These abnormalities of TSG101may play some role in carcinogenesis of breast. Therefore , clinical data including the pathologic stage, histologic type and fellow-up will be collected. Then, the statistical analysis will be used to clearify the associations between genetic alterations of ER, neu, p53 as well as TSG101 and clinical outcome.Table 1:HER-2/neu Row-P=0.018P53+Column Total56.3%43.7%100% HER-2/neuRow -P=0.031PR+Column Total 56.4%43.6%100%p53Row-P=0.004ER+Column Total74.2%25.8%100%Note:HER-2/neu¡G¡Ï¡÷overexpression¡Ð¡÷low level of expression/no expressionp53¡G¡Ï¡÷ point mutation¡Ð¡÷ wild-typeTable 2:p53 Row-P=0.00193Systemic recurrence+Column Total74.1%25.9%100%HER-2/neuRow -P=0.5617Systemic recurrence +Column Total 56.1%43.9%100%Note:HER-2/neu¡G¡Ï¡÷overexpression¡Ð¡÷low level of expression/no expressionp53¡G + ¡÷ point mutatation¡Ð¡÷ wild-typeSystemic recurrence¡G¡Ï¡÷three year’s follow-up¡Ð¡÷no recurrence on follow-up dateREFERENCES1. Benard J. Ahomadegbe JC. TSG101 and breast ca Ali, I.U., Lidereau, R., Theillet, C. & Callahan, R. Reduction to homozygosity of genes on ncer: a correctly named tumor-suppressor gene? Bulletin du Cancer. 84(12):1141-2, 1997Dec.2. Driouch K. Briffod M. Bieche I. Champeme MH. Lidereau R. Location of several putative genes possibly involved in human breast cancer progression. Cancer Research. 58(10):2081-6, 1998 May 15.3. Hofferbert S. Brohm M. Weber BH. Search for TSG101 germ-line mutations in BRCA1/BRCA2-negative breast/ovarian cancer families Cancer Genetics & Cytogenetics. 102(1):86-7, 1998 Apr 1.4. Koonin EV. Abagyan RA. TSG101 may be the prototype of a class of dominant negative ubiquitin regulators Nature Genetics. 16(4):330-1, 1997 Aug.5. Lin PM. Liu TC. Chang JG. Chen TP. Lin SF. Aberrant TSG101 transcripts in acute myeloid leukaemia. British Journal of Haematology. 102(3):753-8, 1998Aug.6. Lee MP. Feinberg AP. Aberrant splicing but not mutations of TSG101 in human breast cancer. Cancer Research. 57(15):3131-4, 1997 Aug 1.7. Li L. Li X. Francke U. Cohen SN. The TSG101 tumor susceptibility gene is located in chromosome 11 band p15 and is mutated in human breast cancer. Cell.88(1):143-54, 1997.8. Li L. Cohen SN. Tsg101: a novel tumor susceptibility gene isolated by controlled homozygous functional knockout of allelic loci in mammalian cells. Cell.85(3):319-29, 1996 May 3.9. Ponting CP. Cai YD. Bork P. The breast cancer gene product TSG101: a regulator of ubiquitination?. Journal of Molecular Medicine. 75(7):467-9, 1997 Jul.10. Steiner P. Barnes DM. Harris WH. Weinberg RA. Absence of rearrangements in the tumour susceptibility gene TSG101 in human breast cancer Nature Genetics. 16(4):332-3, 1997 Aug.11. Sun Z. Pan J. Bubley G. Balk SP.Frequent abnormalities of TSG101 transcripts in human prostate cancer. Oncogene. 15(25):3121-5, 1997 Dec 18.12. Thomson TM. Khalid H. Lozano JJ. Sancho E. Arino J. Role of UEV-1A, a homologue of the tumor suppressor protein TSG101, in protection from DNA damage. FEBS Letters. 423(1):49-52, 1998 Feb 13. 13. Wang Q. Driouch K. Courtois S. Champeme MH. Bieche I. Treilleux I. Briffod M.Rimokh R. Magaud JP. Curmi P. Lidereau R. Puisieux A. Low frequency of TSG101/CC2 gene alterations in invasive human breast cancers. Oncogene. 16(5):677-9, 1998Feb 5.14. Watanabe M. Yanagi Y. Masuhiro Y. Yano T. Yoshikawa H. Yanagisawa J. Kato S. A putative tumor suppressor, TSG101, acts as a transcriptional suppressor through its coiled-coil domain. Biochemical & Biophysical Research Communications.245(3):900-5, 1998 Apr 28.15. Xie W. Li L. Cohen SN. Cell cycle-dependent subcellular localization of the TSG101protein and mitotic and nuclear abnormalities associated with TSG101 deficiency. Proceedings of the National Academy of Sciences of the United States of America.95(4):1595-600, 1998 Feb 17.16. Zhong Q. Chen Y. Jones D. Lee WH. Perturbation of TSG101 protein affects cell cycle progression. [Journal Article] Cancer Research. 58(13):2699-702, 1998 Jul 1.17. Zhong Q. Chen CF. Chen Y. Chen PL. Lee WH. Identification of cellular TSG101 protein in multiple human breast cancer cell lines. Cancer Research.57(19):4225-8, 1997 Oct 1.Table 3 Truncated transcripts of TSG101 in pair specimens4。

行政院国家科学委员会专题研究计画成果报告[001]

瓶頸。（二）國內對於工業製品的需求必然有限。解決國內市場不足的方法可能只有組成多國或區域性貿易協會，有如拉丁美洲自由貿易協會、中美洲共同市場。拉丁美洲國家曾在這方面有所努力，但仍未能解決問題。（三）拉丁美洲工業運用高度的科技，因此只創造了有限的就業機會。這導致了兩個無法避免的不利結果：使國內消費品市場的大小受到限制；未能改善日益嚴重的失業問題。在這樣的情況下，拉丁美洲國家進入第二階段的進口替代工業化的第一個時期，將由本地製造商生產原先仰賴進口的各種資本及技術密集的製造品包括：耐久性消費財（如，汽車）、中間性財貨（如，石化及鋼鐵）、資本財（如，重機械）。這意味建立更技術性勞力的、資本的、技術密集的工業，能夠生產資本財和耐久性消費財，並能加工處理原料；這也意味持續對內導向的發展，包括維持前一階段保護性、控制取向的政策結構。此外，拉丁美洲繼續出口相對豐富的天然資源。6基於以上的特徵，我們又稱之為垂直性的進口替代工業化。拉丁美洲在一九六○年代末期都面臨赤字與通貨膨脹的問題，垂直性的進口替代工業化並不能解決與外在世界之經濟關係處於不平衡的狀況。7因之，拉丁美洲國家修正了第二階段進口替代的策略，包括了促進出口。換言之，以行政部門的措施，選擇性鼓勵特定工業或廠商，推動出口，但並未改變保護性的結構，也沒有走向市場自由化。促進出口需要給予補貼，這項補貼有來自公共部門，也有來自民間部門。這是在既有進口替代工業化的大架構外，給予額外的管制與獎勵。8 而出口促進策略使該地區主要國家工業製品出口都增加了一倍以上。伴隨此政策而來的是主要拉美國家與全球跨國企業策略的進一步聯繫，於此同時，垂直性進口替代仍繼續在資本財部分努力。巴西在此階段中由於先前創造了良好的投資環境而具有優勢，而墨國則繼續其「墨西哥化」政策，吸引外資的結果當然也有差異。9 在這個階段，經濟整合運動的發展，和擴大市場有密切的關係。從一九五○ 年代開始，拉丁美洲區域整合的運動即已出現，但遲至一九六○年代才有正式的組織。最早兩個主要組織分別是中美洲共同市場（ Central American Common

浅谈《历体略》

浅谈《历体略》[摘要]《历体略》是明末天文学西学东渐的产物,作者在书中试图对西方天文学和中国传统天文学进行融会。

透过此书可以对明末知识分子对待新知识和旧有传统的态度以及此书产生的社会环境做一些分析。

[关键词]王英明《历体略》天文学明末一、《历体略》简介《历体略》是明人王英明撰,成书于1612年或略早。

全书共分三卷:上卷六篇介绍了一些中国传统天文学知识;中卷三篇,介绍步天歌,也主要是关于传统天学;下卷七篇侧重介绍欧洲天文学知识。

此书可能是中国士人最早的一部试图融会中西历体知识的作品,也很可能是最早的一部通俗天文学著作。

作者王英明(?—1614),字子晦,自号太常吉星,开州澶渊人(今河南濮阳)。

万历三十四年(1606年)举人。

重刻《历体略》序言中提到他重“实学”,“才识渊博,于历律、兵屯、河防、水衡之事无弗沉究”,著述颇丰。

王英明是明末先进的知识分子的代表,此书在研究明清之际思想观念方面有重要价值。

二、著书背景1.实学思想。

序言和作者自序中提到这部书的编写,是王氏其重视“实学”的一贯态度使然,这就引出了我们对“实学”的思考。

自南宋末期起,封建社会母体中就开始酝酿一种自我否定因素,经济上表现为资本主义萌芽,意识形态领域则是出现了一股批判思潮,这种思潮就是实学思想。

明末实学思想正是这种思潮的延续,这也是西学东渐的一个思想基础。

2.政府控制减弱。

古代,天象观测和星占术称“天文”,历朝都禁止民间私习。

属于历法的内容称为“推步”,在明以前并不禁止私习,入明之后开始确立了全面禁令,在至少100余年的时间里这一禁令都被严格地执行着。

直到明代中后期,随着国家机器的腐朽,这一禁令才逐渐在事实上被废止,这是《历体略》成书的重要社会背景。

没有这一背景,西方的天文学知识不可能得到传布,《历体略》这样的私人天文学著作也不可能产生。

三、本书所体现的作者观念1.作者对待天文学的态度。

作者对待天文学的态度可以从自序中看出来,他提到“历者,帝王经世之洪范。

行政院国家科学委员会.

承辦單位：高雄醫學大學
18
行政院國家科學委員會
研究合作計畫合約（一）
• 智財權之權利歸屬 – 共有且權利均等 – 共有但歸某一方管理 – 歸屬某一方（常見為出資者） – 智財權之權利喪失之可能風險
• 後續研究的可能斷層 • 未來技轉授權的可能侵權責任 • 無法分享智財權的經濟利益
承辦單位：高雄醫學大學 19
30
行政院國家科學委員會
個案分析
承辦單位：高雄醫學大學
31
行政院國家科學委員會
個案六
‧Ｘ校Ａ教授與Ｂ公司進行建教（產學）合作研究計畫，且該建教合作合約中約定智財權皆歸屬Ｂ公司所有。執行該合作案時，Ａ教授運用某一國科會研究計畫之成果於該合作計畫。結案時，Ａ教授未標明國科會之成果及技術範圍，而整合成為該計畫之成果結案報告繳送Ｂ公司。俟後Ｃ公司經由Ｘ校技轉中心授權Ａ教授之國科會研發成果，Ｂ公司得知後具狀控告Ｃ公司、Ｘ校及Ａ教授侵害其權益，以及Ｘ校及Ａ教授違反保密協定。
‧行政院國家科學委員會補助專題研究計畫作業要點
‧行政院國家科學委員會補助產學合作研究計畫作業要點
承辦單位：高雄醫學大學 14
行政院國家科學委員會
國科會補助計畫之權益關係
國科會與學校
除經國科會認定歸屬國科會所有外，均歸學校所有。屬國科會出資部份，除經國科會認定歸屬國科會所有外，均歸學校所有。
個案六解說（一）
‧Ｂ公司之訴求是否可以成立？
–Ｘ校Ａ教授與Ｂ公司既已約定建教合作成果歸屬於Ｂ公司，雖該成果運用某一國科會研究計畫成果，Ｂ公司仍得主張該研究成果之權利，
具狀控告Ｃ公司、Ｘ校及Ａ教授侵害其權益，以及Ｘ校Ａ教授違反保密協定。
承辦單位：高雄醫學大學

行政院国家科学委员会补助专题研究计划作业要点

行政院国家科学委员会补助专题研究计划作业要点状态:有效发布日期:2005-05-18生效日期:2005-05-18发布部门:台湾发布文号:台会综二字第0940036292号一、行政院国家科学委员会(以下简称本会)为补助大专院校及学术研究机构执行科学技术研究工作，以提升我国科技研发水准，特订定本要点。

二、申请机构(即执行机构)：(一)公私立大专院校及公立研究机构。

(二)经本会认可之财团法人学术研究机构。

三、计划主持人(申请人)及共同主持人之资格：(一)申请机构编制内按月支给待遇之专任教学、研究人员，具有专门学识与研究经验，且有具体研究成绩，并具备下列资格之一者：1.助理教授级以上人员。

2.具博士学位之专任教学或研究人员。

3.担任讲师职务四年以上，并有著作发表于国内外著名学术期刊或专利技术报告专书者。

4.研究机构副研究员、技正或相当副研究员资格以上人员。

5.于教学医院担任主治医师二年以上或获硕士学位从事研究工作四年以上，并有著作发表于国内外著名学术期刊之医药相关人员。

具有前项计划主持人资格，且依相关规定被借调之人员，得由原任职机构提出申请。

(二)已退休之教学、研究人员，如为中央研究院院士、曾获得教育部国家讲座或学术奖、本会特约研究人员或杰出研究奖三次以上、财团法人杰出人才发展基金会杰出人才讲座、或其它相当奖项经本会认可者，且其原任职机构于申请研究计划函内叙明愿意提供相关设备供其进行研究并负责一切行政作业者，得申请一般型研究计划补助。

(三)实施校务基金制度之学校，于校务基金自筹经费范围内，依国立大学校院进用项目计划教学人员、研究人员暨工作人员实施原则聘任之专任教学、研究人员，按月支给待遇，经学校各级教评会审议通过遴聘，符合第(一)项计划主持人资格者，得申请专题研究计划补助。

(四)公立大专校院依公立大专校院稀少性科技人员遴用资格办法遴用具博士学位之核能、信息及航天等三类稀少性科技人员，得申请专题研究计划补助。

行政院国家科学委员会专题研究计画期中进度报告

Improvement on the growth of ultrananocrystalline diamond by using pre-nucleation techniqueYen-Chih Lee1, Su-Jien Lin1, Debabrata Pradham2,I-Nan Lin2*1. Department of Material Science and Engineering, National Tsing-Hua University, Hsin-Chu 300, Taiwan, R. O. C.;2. Department of Physics, Tamkang University, Tamsui251, Taiwan, R. O. C.AbstractUltrananocrystalline diamond (UNCD) films, which possess very smooth surface, were synthesized using CH4/Ar plasma. The Si-substrate was pre-nucleated using bias enhanced nucleation (BEN) technique under CH4/H2 plasma, so that the growth of UNCD films can be markedly enhanced. The growth rate of these UNCD films were observed to be correlated intimately with the deposition conditions, such as substrate temperature, microwave power, total pressure, CH4 ratio. When the nucleation process was carried out under methane and hydrogen (CH4/H2) plasma with negative DC bias voltage, no pretreatment on substrate was required prior to the formation of diamond nuclei. The growth kinetics of BEN induced nuclei was monitored by the evolution of the bias current to ensure the full coverage of diamond nuclei on the Si-substrate. The average grain size of BEN induced diamond nuclei is about 30 nm, with the nucleation site density more than 1011 sites/cm2. The growth rate of UNCD is markedly enhanced due to the application of BEN induced nuclei. Moreover, the growth rate of UNCD films was more significantly affected by the substrate temperature, but was less influenced by the microwave power. All of these UNCD films showed similar morphology, i.e., with grain size less than 10 nm and surface roughness around 20 nm. They also possess the same Raman spectra, i.e., the same crystallinity. However, the deposition rate can be increased from about 0.2 µm/hr to 1.0 µm/hr when substrate temperature increased from 4000C to 600o C.Novelty:The ultrananocrystalline diamond (UNCD) films were grown on bias-enhaned nucleation substrate to improve the of growth behavior.Keywords: UNCD, high speed growth, BEN, MPECVDSubmission of this paper has been approved by the co-authors.Corresponding author: Prof. I-Nan Line-mail: inanlin@.twTel. 886-2-26268907; Fax. 886-2-26207717Department of physics, Tamkang University;151 Yin-Chuan Rd. Tamsui, Taipei, Taiwan 251, R. O. C.Estimated word count: 3571words1Improvement on the growth of ultrananocrystalline diamond by using pre-nucleation techniqueYen-Chih Lee1, Su-Jien Lin1, Debabrata Pradham2, I-Nan Lin2*1. Department of Material Science and Engineering, National Tsing-Hua University, Hsin-Chu 300, Taiwan, R. O. C.;2. Department of Physics, Tamkang University, Tamsui251, Taiwan, R. O. C.AbstractUltrananocrystalline diamond (UNCD) films, which possess very smooth surface, were synthesized using CH4/Ar plasma. The Si-substrate was pre-nucleated using bias enhanced nucleation (BEN) technique under CH4/H2 plasma, so that the growth of UNCD films can be markedly enhanced. The growth rate of these UNCD films were observed to be correlated intimately with the deposition conditions, such as substrate temperature, microwave power, total pressure, CH4 ratio. When the nucleation process was carried out under methane and hydrogen (CH4/H2) plasma with negative DC bias voltage, no pretreatment on substrate was required prior to the formation of diamond nuclei. The growth kinetics of BEN induced nuclei was monitored by the evolution of the bias current to ensure the full coverage of diamond nuclei on the Si-substrate. The average grain size of BEN induced diamond nuclei is about 30 nm, with the nucleation site density more than 1011 sites/cm2. The growth rate of UNCD is markedly enhanced due to the application of BEN induced nuclei. Moreover, the growth rate of UNCD films was more significantly affected by the substrate temperature, but was less influenced by the microwave power. All of these UNCD films showed similar morphology, i.e., with grain size less than 10 nm and surface roughness around 20 nm. They also possess the same Raman spectra, i.e., the same crystallinity. However, the deposition rate can be increased from about 0.2 µm/hr to 1.0 µm/hr when substrate temperature increased from 4000C to 600o C.Novelty:The ultrananocrystalline diamond films were grown on bias-enhaned nucleation substrate to improve the growth.Keywords: UNCD, high speed growth, BEN, MPECVD2I. IntroductionThe unique combination of the physical and chemical properties of diamond film save drawn more attention among researcher to use diamond in many applications. However, the high roughness of microcrystalline diamond films made them inapplicable in specific applications. In the recent past, very smooth ultra nano-crystalline diamond (UNCD) films deposited by CH4/Ar mixture has been established. The detail mechanism for the formation of UNCD from CH4/Ar plasma has been reported[1, 2]. Recent application of nano-diamond films in bio-sensors[3], filed emission[4, 5] and bio-medical application[6] have shown the promising future of this nano-material. Even so, no detailed study has been performed on the growth rate and formation of nucleation sites by biased enhanced nucleation (BEN) method to grow uniform UNCD film on the silicon surface. Therefore, it is significant to understand more precisely on the deposition rate and deposition conditions influencing growth process of UNCD film.The substrate pretreatment strongly affects the nucleation and growth process of diamond films determining the initial deposition rate, crystal quality and surface roughness. High deposition rate is primarily important to grow thick diamond films normally required for application like SAW devices[7]. Moreover, a smooth surface of diamond film is another important requirement. Thus, suitable conditions need to be established to grow ultrananocrystalline diamond grains to directly obtain a smoother film. One of the most effective methods of diamond nucleation is bias enhanced nucleation (BEN) method[8, 9]. The formation of nano-diamond phase on silicon acts as nucleation center for the growth of either nano-crystalline or microcrystalline diamond depending on the deposition parameters used.One of the main objectives of the present work was to systematically investigate the nucleation behavior of UNCD on silicon surface using a BEN technique. As BEN method does not involve any scratching by diamond abrasives, it avoids the confusion of presence of any residual diamond particle on the substrate. Another objective of current study is to find a suitable deposition condition for the high and uniform growth of UNCD. The effect of microwave power, substrate temperature, CH4 to Ar ratio and total pressure on the growth rate is reported in this article.3II. ExperimentalThe bias enhanced nucleation (BEN) diamond films were grown in a 2.45 GHz ASTeX microwave plasma enhanced chemical vapor deposition (PECVD) system on N-type mirror polished Si (100) substrates. A microwave power of 1.5 kW (ASTeX 5400), total pressure of 55 torr and 300 sccm H2 flow rate were used during biased treatment. Different substrates were biased treated for different time intervals (0 to 15 minutes) at constant biased voltage (-125 V) and the resulted bias current – time relationship was measured. Silicon substrates after BEN process were used for the deposition of UNCD in an IPLAS MPCVD system. Table I presents the detail experimental deposition conditions used for UNCD growth. In Series - P, C, T and MW, chamber pressure, CH4/Ar ratio, temperature and microwave power was varied respectively, keeping rest of the parameters constant.Surface morphology of samples was examined with a field emission scanning electron microscope (JEOL 6010). Crystal quality of UNCD films was investigated by Raman Spectroscopy using 514 nm argon laser beam (Renishaw). Surface topography and roughness was measured with atomic force microscopy (PARK).III. Results and discussion(a) Nucleation processThe formation of nanodiamond phase for the nucleation of diamond growth during BEN is known for last few years. Therefore BEN time is crucial to create uniform nucleation center on the silicon. Figure 1 shows the nucleation of nano-diamond films deposited after different BEN time intervals. The SEM images show uniform island growth at the beginning after 5 minutes of BEN (Fig 1a) and subsequent increase of coverage in nano-diamond grains clusters on the surface after 7 minutes (Fig 1b). After 8 minutes of BEN, the whole silicon surface is covered by cluster of nano-diamond crystals. The average size of nano-diamond cluster is around 150 nm and size of each diamond grain in the clusters is ~ 20 nm (Fig 1c). A saturation of nano-diamond growth occurs after 8 minutes of BEN, covering whole area of silicon substrate. At 10 min of BEN, cluster size of nano-diamond is decreased but diamond grain of size 50 nm started to appear (Fig 1d). The surface morphology is almost same after 10 minutes BEN. This4establishes the minimum time (8 minutes) required for the creation of high nucleation centers and uniform nano-diamond layer on silicon. The grain size is also found to be smallest (~20 nm) after 8 minutes of BEN. The thickness was about 250 nm.The measurement of bias current during BEN in our study indicates direct correlation in the formation of nano-diamond phase on silicon, which is shown in fig. 2 for the trend of bias current versus time at a constant negative bias voltage of 125 V. When there is no methane flow, bias current is about -52 mA. This current starts to decrease in the first 3 minutes, which may be due to methane content increases and changes plasma condition. The subsequent increase of bias current which has been attributed to the enhancement in electron emission from the highly emissive diamond formed on silicon substrate surface and the bias current become saturated after 10 minutes indicating no more nano-diamond coverage increase in surface.(b) Growth processFigure 3 shows the effect of various parameters on the deposition rate of UNCD. The deposition rate of diamond film is found to depend significantly on the temperature and the ratio of CH4 to Ar in the reactant gas compared to microwave power and total pressure. The effect of chamber pressure in the deposition rate is linearly increases from 100 torr to 150 torr. There is no much effect of pressure on the deposition rate. However, there is a striking increase in deposition rate when CH4/Ar ratio was increased from 0.5 % to 2.0 %. The substrate temperature is another important parameter for increasing deposition rate. The deposition rate is found to increase around 4 times with increase in temperature from 400o C to 600o C while all other parameters were held constant. The deposition rate of UNCD increases with microwave power from 600 W to 750 W. However, deposition rate comes down and become almost constant in the microwave power range of 900 W to 1200 W. We have grown a UNCD film having highest deposition rate of ~1 µm/hr at 750 W with combination of parameters: 150 torr pressure, 600 o C temperature and 1 % methane.Figure 4 shows typical SEM images of UNCD film. Diamond grains of size less than 10 nm have been grown under above described deposition conditions. Unlike the agglomeration observed after bias enhanced nuclation, UNCD film shows a uniform and5smooth surface having numerical diamond crystallites density of as high as 1012 /cm2. This high nucleation density was due to formation of high nucleation centers during BEN and the growth in CH4/Ar plasma. The inset shows a cross-sectional image of UNCD film. The surface roughness of the diamond films is strongly affected by deposition method. AFM analysis shows that the surface roughness of the BEN film using CH4/H2 source gas under continuous bias (-125 V) in 8 minutes period was about 13.2 nm. This surface roughness reduces to 10 nm after deposition of UNCD using CH4/Ar source gas. The decrease of surface roughness is well matched with decrease of diamond grain size measured by SEM. Size of diamond grains were less than 10 nm after UNCD deposition on a 20 nm diamond crystallites film formed during negative biasing.Raman technique is one of the important non-destructive characterization techniques to study the properties of any type of carbon forms. There are four main peaks normally observed at around 1140 cm-1, 1330 cm-1, 1470 cm-1 and 1560 cm-1 in visible Raman spectrum of UNCD films[10, 11]. Figure 5 shows the Raman spectra of UNCD films deposited on silicon at different experimental conditions. These Raman spectra are found to be very similar to as reported in literature[10, 11]. The broad peak at 1330 cm-1 and 1560 cm-1 are commonly termed as D-band and G band respectively. The peaks at 1140 cm-1 and 1470 cm-1 are sometimes assigned to nano-crystalline diamond films[12, 13].However, there is certain ambiguity in these two peaks. Ferrari et. al.[14] and Kuzmany et. al.[15] have assigned these two peaks at 1140 cm-1 and 1470 cm-1 to trans-polyacetylene segments present at the grain boundaries and surfaces of diamond films. However, these two peaks are most commonly observed in UNCD or NCD film[10, 11]. Since the sp2-bonded carbon is highly sensitive to visible Raman spectroscopy than sp3-bonded carbon, sharp peak at 1332 cm-1 is not observed. In our study, the peak height of 1140 cm-1 and 1470 cm-1 suggests the increase in trans-polyacetylne percentage with substrate temperature. Raman spectra of series –P, -C and -MW samples are almost same indicating not much change in crystallinity of UNCD films by varying pressure, CH4/Ar ratio and microwave power.6IV. ConclusionUNCD film of diamond grain less than 10 nm was grown on BEN treated silicon surface. Nucleation density of ~1012 grains/cm2 was obtained in the growth process. Silicon substrate was biased for different time interval to study the formation of nucleation center. Our study has shown that a minimum 8-minute of bias enhanced nucleation was needed for uniform growth of UNCD film. Agglomeration of diamond crystallites obtained in BEN diamond growth was not observed after the growth of UNCD in a CH4-Ar medium. AFM study depicted the improvement in smoothness of UNCD film to 6.83 nm from 10.83 nm obtained by BEN NCD film. Raman spectra have shown the peak at respective positions that normally observed in UNCD films.V. AcknowledgmentThe authors would like to thank National Science Council, R.O.C. for the support of this research through the project No. NSC 93-2112-M-032-010.VI. References[1] D. Zhou, T. G. McCauley, L. C. Qin, A. R. Krauss, and D. M. Gruen, J. Appl. Phys.83 (1998) 540.[2] D. M. Gruen, Annu. Rev. Mater. Sci. 29 (1999) 211.[3] A. Hartl, E. Schmich, J.A. Garrido, J. Hernando, S.C.R. Catharino, S. Walter, P. Feulner, A. Kromka, D. Sreinmuller, M. Stutzmann, Nature Materials, 3 (2004) 736.[4] W. Zhu, G.P. Kochanski, S. Jin, Science282 (1998) 1471.[5] K. Wu, E.G. Wang, Z.X. Cao, Z.L. Wang, X. Jiang, J. Appl. Phys.88 (2000) 2967.[6] M. D. Fries, Y.K. Vohra, Diam. Rel. Mater. 13 (2004) 1740.[7] F. Benedic, M.B. Assouar, F. Mohasseb, O. Elmazria , P. Alnot , A. Gicquel, Diam. Rel. Mater. 13 (2004) 347.[8] S. Yugo, T. Kanai, T. Kimura, T. Muto, Appl. Phys. Lett. 58 (1991) 1036.[9] Q. Chen, Z. Lin, J. Appl. Phys. 80 (1996) 797.[10] X. Xiao, J. Birrell, J. E. Gerbi, O. Auciello, J. A. Carlisle, J. Appl. Phys. 96 (2004) 2232.[11] Y. Hayashi, T. Soga, Tribology International 37 (2004) 965.[12] W.A. Yarbrough, R. Messier, Science 247 (1990) 688.7[13] R.J. Nemanich, J.T. Glass, G. Lucovsky, R.E. Shroder, J. Vac. Sci. Technol. A 6 (1988) 1783.[14] A.C. Ferrari, J. Robertson, Phys. Rev. B, 63 (2001) 121405.[15] H. Kuzmany, R. Pfeiffer, N. Salk, B. Gunther, Carbon 42 (2004) 911.89Table I: Experimental deposition conditions for UNCD growth on BEN silicon surface. Materials Pressure(Torr)CH 4/Ar Ratio (%) Temperature (o C) MW Power (W) Series-P100~150 1 % 400 1200 Series-C150 0.5 ~ 2 % 4001200 Series-T150 1 % 400 ~ 600 750 Series-MW150 1 % 600 600 ~ 1200Figure captionsFig 1. SEM images of diamond films grown on silicon surface after different BEN time intervals, (a) 5 min., (b) 7 min., (c) 8 min. and (d) 10 min of BEN.Fig. 2. Bias current from the electrode to the substrate holder during the bias enhanced nucleation as a function of time.Fig. 3. Effect of pressure, CH4 to Ar ratio, substrate temperature and microwave power on the deposition rate of UNCDFig. 4. FE-SEM image of UNCD grown under deposition condition of 750 W microwave power, 150 torr total pressure, total flow of 200 sccm Ar-CH4 (CH4-1 %) andsubstrate temperature 600 o C in 3 hr deposition period. Inset shows a cross-sectional view.Fig. 5. Visible Raman spectra UNCD films obtained under different experimental conditions.。

行政院国家科学委员会补助专题研究计画作业要点

行政院國家科學委員會補助專題研究計畫作業要點95年10月4日本會第543次主管會報修正通過一、行政院國家科學委員會(以下簡稱本會)為補助大專院校及學術研究機構執行科學技術研究工作，以提升我國科技研發水準，特訂定本要點。

二、申請機構（即執行機構）：(一)公私立大專院校及公立研究機構。

(二)經本會認可之財團法人學術研究機構。

三、計畫主持人(申請人)及共同主持人之資格：(一)申請機構編制內按月支給待遇之專任教學、研究人員，具有專門學識與研究經驗，且有具體研究成績，並具備下列資格之一者：1.助理教授級以上人員。

2.具博士學位之專任教學或研究人員。

3.擔任講師職務四年以上，並有著作發表於國內外著名學術期刊或專利技術報告專書者。

4.研究機構副研究員、技正或相當副研究員資格以上人員。

5.於教學醫院或研究機構擔任主治醫師二年以上或獲碩士學位從事研究工作四年以上，並有著作發表於國內外著名學術期刊之醫藥相關人員。

具有前項計畫主持人資格，且依相關規定被借調之人員，得由原任職機構提出申請。

(二)已退休之教學、研究人員，如為中央研究院院士、曾獲得教育部國家講座或學術獎、本會傑出研究獎、財團法人傑出人才發展基金會傑出人才講座、或其他相當獎項經本會認可者，且其原任職機構於申請研究計畫函內敘明願意提供相關空間及設備供其進行研究並負責一切行政作業者，得申請一般型研究計畫補助。

(三)實施校務基金制度之學校，於校務基金自籌經費範圍內，依國立大學校院進用專案計畫教學人員、研究人員暨工作人員實施原則聘任之專任教學、研究人員，按月支給待遇，經學校各級教評會審議通過遴聘，符合第（一）項計畫主持人資格者，得申請專題研究計畫補助。

(四)公立大專校院依公立大專校院稀少性科技人員遴用資格辦法遴用具博士學位之核能、資訊及航太等三類稀少性科技人員，得申請專題研究計畫補助。

四、專題研究計畫分為下列二種：(一)一般型研究計畫：符合計畫主持人資格者，得依研究專長或參考本會學門規劃項目申請本項計畫。

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)。

行政院國家科學委員會專題研究計劃成果報告網際網路關係行銷之研究The Study of the Relationship Marketing on Internet 計劃編號: NSC 90-2416-H-415-002-SSS執行期限:90年8月1日至91年7月31日主持人: 陶蓓麗副教授國立嘉義大學管理研究所一、中文摘要隨著網際網路無遠弗屆的影響，顧客與企業關係建立與維繫的方式也受到衝擊，企業瞭解線上顧客關係的建立與維繫已是迫切的議題。

少有研究調查網際網路在關係行銷的角色。

本研究目的是瞭解網際網路上的顧客關係，探討影響線上顧客關係的因素。

本研究提出網際網路關係維繫的模型，影響線上顧客關係的因素包括社會、顧客、網站、及互動四個方向，共十一個構面，信任及關係承諾是衡量關係建立與維繫意願的指標。

研究結果驗証了網際網路顧客關係的存在，也証實社會、顧客、網站及互動四方面的因素會影響線上顧客關係，研究建議網站應建立自己的社群、導入顧客關係管理、注重專業領域的投入及網路的互動環境。

研究也發現年紀輕、收入低者較易對網站產生信任感，其間接社會束縛的力量也較大；而教育程度低者，雖然顧客專業技能較差，但較願投入精力以建立彼此關係，也較受間接社會束縛力量的影響。

此外，在各種不同類型的網站中，入口網站與購物網站最不易取得顧客的信任與關係承諾。

研究結果可作為企業界未來訂定網際網路關係行銷政策時的參考，企業界若能瞭解維繫顧客關係的要素，將使企業能更有效的運用網際網路於行銷策略上。

關鍵字：關係行銷、網路行銷、顧客關係AbstractInternet is transforming the customers and sellers relationships. Understanding why customers are receptive to a company on Internet is an emerging issue in relationship marketing. Little research has been done to investigate the enabler role of Internet in relationship marketing theory and practice. The purpose of this study is to explore the customer relationship on the Internet. The study also examines the factors influencing online customer relationship.A model for online customer relationship is proposed. This study suggests that four broad drivers—society, customer, web site, and interaction—affect online customers’receptivity to relationship maintenance. Trust and relationship commitment are used to measure the desire of relationship establishment and maintenance. Results indicate that the customer relationship on the Internet exists. All four broad drivers, which included society, customer, web site and interaction, are significant predictors in terms of improving customer relationship on Internet. The study suggests that companies should establish their own online community, introduce customer relationship management, focus on their area of expertise and aware of the unique interactive feature on Internet. Results also find that young and low-income customers respond to indirect social bonding and tend to trust companies. Though low-educated customers are low in customer expertise, they are more likely to devote their energy for relationship establishment and tie to indirect social bonding. Also, portal and shopping web sites are less trustful and less likely to be committed. The results enable a firm to understand the roles of Internet on relationship marketing and to develop Internet strategy effectively. It serves as a basis for the future growth of Internetmarketing and can guide those wishing to champion Internet marketing in their organization.Keywords: Relationship Marketing, InternetMarketing, CustomerRelationship二、研究背景與目的自八○年代以來，關係行銷觀念的提出對行銷領域產生了深遠的影響。

許多學者指出關係行銷從根本改變了行銷理論與實務，也改變了市場競爭的遊戲規則，關係行銷是成功行銷的關鍵。

另一方面，近年來網際網路已成為一種新興的行銷媒介，在我國上網人口穩定成長之際，如何有效維持既有的上網人口，一直是各界積極思考的議題。

企業若要在競爭激烈的網路商戰中獲得生存，便要將以往關係行銷之觀念與技巧應用在網路商務上，促使顧客與網站之緊密聯繫，網站運用顧客所提供的資訊，給予更乎合個人需求的服務，這種經營方式往往能對線上使用者產生「套牢效果」，透過關係建立，使顧客不會輕易轉移至其他網站，網站與顧客間長期關係得以建立。

在消費者市場、工業市場、金融市場等領域，關係行銷方面的研究相當多[1, 2,3,4]，但有關網際網路上顧客關係維繫方面的研究，仍是付之闕如，本研究自關係行銷的角度出發，探討在網路上何種因素會影響顧客關係的建立，何種因素會影響線上顧客關係的維繫，以做為日後企業在擬定網路行銷策略與建立顧客關係管理系統時的參考。

茲將本研究的相關研究議題條列如下：(一)、顧客對各種線上顧客關係的感受程度為何？(二)、影響線上顧客關係建立與維繫的因素為何？(三)、何種線上顧客關係的因素較其他因素更能建立與維繫顧客的關係？(四)、探討人口特徵、網站類型與線上顧客關係建立與維繫之間的關聯。

三、研究模式本研究的研究對象為曾使用網路用者。

調查對象男性比例略多於女性，年齡以20-29歲者居多，教育程度以大專院校佔大多數，顯示本研究的樣本特徵與國內其它研究所描述的網際網路族群特徵大致相符。

四、資料分析研究問題一探討顧客對各項線上顧客關係因素的感受程度，研究顯示顧客對各項線上顧客關係因素感受程度的差異不大，均屬普通，其中以溝通效能及顧客專業技能最高，而隱私權的重視最差，顧客的關係特定投資是倒數第二。

顯示顧客對自我專業技能的評估均較高，但大都不願投入太多時間與精力於其中。

同時，顧客對網站隱私權的重視評價也最差。

研究問題二探討影響線上顧客關係的因素，本研究首先以因素分析萃取出十一個獨立變數並據此而計算出十一個因素分數，信任及關係承諾則根據各自的衡量項目之平均值計算而得，本研究以逐步迴歸分析探討十一個因素對信任及關係承諾的影響。

研究發現，在採用0.05的進入及0.10的移出標準下，十一個因素中共有九個因素對線上顧客信任造成顯著的影響，第三者推薦及顧客的關係特定投資並未納入研究模式內，研究結果支持假說H1a, H2a, H4a, H6a, H7a, H8a, H9a, H10a及H11a，無法支持假說H3a及H5a。

另一方面，在採用同樣的進出標準下，十一個因素中共有十個因素對線上顧客關係承諾產生顯著的影響，第三者推薦被排除於研究模式之外，研究結果支持假說H1b, H2b, H4b, H5b, H6b, H7b, H8b, H9b, H10b及H11b，無法支持假說H3b。

研究問題三進一步探討各種線上顧客關係因素的相對重要性，亦即何種因素較能改善線上顧客的信任與承諾感，依偏相關係數的高低排列出其重要程度。

對信任具有影響力的因素，依其重要程度排列為隱私權的重視、互動滿意度、顧客專業技能、網站專業技能、直接社會束縛、網站的關係特定投資、價值的相似性、溝通效能及間接社會束縛。

對關係承諾具有影響力的因素，依其重要程度排列為互動滿意度、溝通效能、價值的相似性、隱私權的重視、直接社會束縛、顧客專業技能、間接社會束縛、網站的關係特定投資、網站專業技能、及顧客的關係特定投資。

研究問題四探討人口特徵與線上顧客關係建立與維繫的關聯，本研究以Spearman等級相關係數來分析各特徵變數與信任及關係承諾間的關係，結果顯示網站蒞臨頻率與信任間存在顯著正相關，而年齡與信任間及月收入與信任間存在顯著負相關，其餘的特徵變數，教育程度、網站蒞臨歷史、以及網路使用歷史，與信任間並未發現有顯著關係；網站蒞臨頻率與關係承諾間存在顯著正相關，其餘的特徵變數與關係承諾間並未發現有顯著關係。

為探討不同的網站類型對信任及關係承諾的影響，本研究利用ANOVA分析後發現網站類型的不同，顧客對信任與關係承諾度也會不同，再經配對比較後發現，顧客對入口網站的信任程度要較多種不同類型的網站來得低；同時，顧客對入口網站與購物網站的承諾度要較多種不同類型的網站來得低。