本研究首先以LED閱讀燈相關專利測試，以位置權重概念將專利文件分為專利名稱、摘要、說明全文及申請專利範圍四個文字敘述位置，根據此四個位置敘述之內容對於專利的重要程度不同而設定數組專利文件之此四個位置之不同的位置權重值之組合，測試分析出一組較佳的組合做為本研究之用。再將此較佳之不同的位置權重值之組合代入本研究之加入位置權重概念計算之常態化數值之計算公式，經由專利語意分析之斷詞斷字系統半自動半人工分析所獲得之各關鍵技術字、零組件元件字及功能字之加入位置權重概念計算之常態化數值做為該詞彙之權重，取代原有的未加入位置權重概念計算之常態化數值，可以更佳改善以結合改良式餘弦相似度概念之創新專利搜尋方法搜尋相關專利，並進行專利分析之結果，及以改良式餘弦相似度結合專利類別歸屬之機率方法進行專利歸屬之結果。本研究將位置權重的概念加入以改良式餘弦相似度結合專利類別歸屬之機率方法進行專利歸屬，其為將專利各關鍵字之加入位置權重概念計算之常態化數值先代入改良式餘弦相似度之公式的權重值，將新投入專利與各技術類別或功能類別比對計算考慮位置權重之餘弦相似度，初步判斷該篇新投入專利可能歸屬的技術類別或功能類別，再將該篇新投入專利各關鍵字之加入位置權重概念計算之常態化數值代入專利類別歸屬之機率方法之公式的權重值，計算專利歸屬之機率值(Pj)，判斷該篇新投入專利歸屬的技術類別或功能類別，以期更進一步提高判斷新投入專利之歸屬的準確性及效率。
上述方法先以LED閱讀燈相關專利進行測試分析完成後，再以LED立/桌燈為研究之載具，將所測試出較佳之不同的位置權重值之組合代入本研究之加入位置權重概念計算之常態化數值之計算公式，計算專利各關鍵字之加入位置權重概念計算之常態化數值，結合改良式餘弦相似度概念之創新專利搜尋方法搜尋相關專利並進行專利分析，並以改良式餘弦相似度結合專利類別歸屬之機率方法進行專利歸屬，實際投入相關專利驗證其歸屬之第一層、第二層技術類別及功能類別，證明本研究之創新方法為可行。本文將先定義LED立/桌燈相關專利之第一層、第二層技術類別及功能類別，進而產生第一層及第二層技術/功能矩陣，並建立各第一層、第二層技術類別及功能類別之技術字及功能字字群。接著將所搜集之相關專利進行國際專利分類(IPC)分析、所屬國別分析、申請人分析及公司別分析，並由第一層及第二層技術/功能矩陣建立各第一層技術類別及第二層技術類別與公司別分析表、核心競爭力公司之IPC分析表、競爭者研發活動力趨勢表及各第一層及第二層技術類別之活動力趨勢表。上述分析之結果可提供給企業及工程師，以節省研發時間，作為新專利產生之參考，此外亦可使企業了解競爭者之核心競爭力。

First of all, the study takes a test of patents relating to LED reading lamp. With position weight concept, these patent documents are divided into four positions with written description, namely the title of patent, abstract, description and claim(s). In accordance with the contents of these four position descriptions, and judging the difference in importance of these these four position descriptions for the patent document, the paper sets several combinations of different position weights for these four positions of patent documents, and then tests and finds out a better combination for these four positions. The better combination of different position weights is substituted in the study’s calculation equation of normalized numerical value calculated by the added position weight concept. The normalized numerical values which are calculated by the added position weight concept, for the various key technical words, part/component words and functional words, are taken as the weights of terms, replacing the original normalized numerical values calculated without addition of position weight concept. The above mentioned key technical words, part/component words and functional words are obtained by the term segmentation and word segmentation system of patent semantic analysis semi-automatically and semi-manually. It can better improve the analytic results of the related patents searched by innovative patent search method under the combined improved cosine similarity concept, and the results of patent categorization achieved by combination of improved cosine similarity and probability method of patent categorization. In this study, the position weight method is added in the patent categorization judgment which is combining improved cosine similarity with probability method of patent categorization. By doing so, the normalized numerical value, calculated by the added position weight method, of each patent keyword is substituted in the weight value of improved cosine similarity equation. After comparing the newly added patents with the various technical categories or functional categories, improved cosine similarity which is added position weight values is calculated. It is preliminarily judged that the newly added patent may be categorized into which technical category or which functional category. The normalized numerical value, calculated by the added position weight concept, of each keyword of the newly added patent is substituted in the weight value of the equation for probability method of patent categorization, in order to calculate the probability of patent categorization (Pj), judge whether the newly added patent is categorized into which technical category or which functional category, and further enhance the accuracy and efficiency in judgment of the categorization of the newly added patent.
After employing the above method to test and analyze the related patents of LED reading lamp, the paper takes LED standing/desk lamp as the carrier of the study, and substitutes the better combination of different position weights in the study’s calculation equation of normalized numerical value calculated by the added position weight concept so as to calculate the normalized numerical value calculated by the added position weight concept of each keyword of patent. Combining with the innovative patent search method under the improved cosine similarity concept, the study searches the related patents, makes analysis of patents, and conducts patent categorization by combining improved cosine similarity with probability method of patent categorization. After practical addition to the related patents, it is verified that the added related patent is categorized to which one of 1st-layer, 2nd-layer technical category and functional category. Then, it is verified proving that the feasibility of the proposed innovative method of the study. The study first defines the 1st-layer, 2nd-layer technical category and functional category of LED standing/desk lamp, further produces the 1st-layer and 2nd-layer technical/functional matrix, and establishes word groups of technical words and functional words of the 1st-layer, 2nd-layer technical category and functional category. After that, the related patents searched carry out international patent classification (IPC) analysis, nationality analysis, applicant analysis and company analysis. According the 1st-layer and 2nd-layer technical/functional matrix, it can establish the 1st-layer technical category and the 2nd-layer technical category, as well as company analysis chart, IPC analysis chart of core competitive companies, research and development (R&D) vitality trend chart of competitors, and vitality trend chart of the 1st-layer and 2nd-layer technical categories. The above analytic results can be provided to enterprises and engineers to save their R&D time, and can serve as a reference for production of new patents. Besides, they can also let enterprises understand the core competitiveness of their competitors.

【1】「國際專利分類表第七版使用指南」，經濟部智慧財產局，民國 九十年。
【2】「國際專利分類之研究」，經濟部中央標準局。
【3】 Overview of the Classification System, http://www.uspto.gov/main/patents.htm 美國專利商標局(USPTO)。
【4】 The EPO guide to patent information on the internet, http://epart.epo.org/dwl/espacenet_manual.pdf 歐洲專利組織(EPO)。
【5】 Trappey, C. V., Trappey, A. J. C., and Wu, C. Y., “Clustering patents using non-exhaustive overlaps”, Journal of Systems Science and Systems Engineering, Vol.19, No.2, pp.162-181(2010).
【6】 Trappey, C. V., Trappey, A. J. C., and Wu, C. Y., “Automatic patent document summarization for collaborative knowledge systems and services”, Journal of Systems Science and Systems Engineering, Vol.18, No.1, pp.71-94(2009).
【7】 Singhal, A. and Salton, G., Automatic Text Browsing Using Vector Space Model, Technical Report, Department of Computer Science, Cornell University(1993).
【8】 Lawrence, S. and Giles, C. L., “Searching the Web: General and Scientific Information Access ”, IEEE Communication, Vol.37, pp.116-122(1999).
【9】 陳哲宏、陳逸南、謝銘洋、徐宏昇，「專利法解讀」，月旦出版公司，第14-15頁，民國八十三年。
【10】Gruber, T. R.,“A translation approach to portable ontology specifications”, Knowledge Acquisition, Vol.5, pp.199-200(1993).
【11】Oleary, D. E., “Enterprise knowledge management”, IEEE Computer, Vol.31, pp.54-61(1998).
【12】杜家宏，「整合繁體和簡體中文及英文斷詞斷字系統之發光二極體檯燈專利分析研究」，碩士論文，國立台灣科技大學機械 工程學系，台北，民國一百年。
【13】吳俞灃，「LED水族燈之相關專利分析研究」，碩士論文，國立台灣科技大學機械工程學系，台北，民國一百零一年。
【14】Salton, G. and McGill, M., Introduction to Modern Information Retrieval, McGraw-Hill, New York(1983).
【15】Vallet, D., Fernandez, M., and Castells, P., “An Adaptation of the Vector-Space Model for Ontology-Based Information Retrieval”, IEEE Knowledge and Data Engineering, Vol.19, No.2, pp.261-272(2005).
【16】Wang, S., Sun, J., Gao, B. J., and Ma, J., “Adapting Vector Space Model to Ranking-based Collaborative Filtering”, CIKM'12 Proceedings of the 21st ACM international conference on Information and knowledge management, pp.1487-1491, New York, USA(2012).
【17】Gan, J. and Chen, L. C., “Research of improved TF-IDF Weighting algorithm“, International Conference on Information Science and Engineering (ICISE), pp.2304-2307, Hangzhou, China(2010).
【18】Zhang, G. H. and Odbal, “Sentence Alignment For Web Page Text Based On Vector Space Model”, International Conference on Computer Science and Information Processing (CSIP), pp.167-170, Shaanxi, China(2012).
【19】Hu, X., and Wu, Bin., “Automatic Keyword Extraction Using Linguistic Features”, 6th IEEE International Conference on Data Mining (ICDM), pp.19-23, Hong Kong, China(2006).
【20】Dong, H., and Yin, S., “A improved feature weighting algorithm for Chinese text classification”, 2010 International Conference on Computer Application and System Modeling (ICCASM), Vol.6, pp.433-436, Taiyuan(2010).
【21】Liu, X., Zheng, Q., Ma, Q., and Lin, G., “A Method for Thematic Term Extraction Base on Word Position Weight”, International Conference on Artificial Intelligence and Computational Intelligence (AICI), pp.266-277, Chengdu, China(2012).
【22】Cao, H. and Jia, H., “Tibetan text classification based on the feature of position weight”, 2013 International Conference on Asian Language Processing (IALP), pp.220-223, Urumqi, Xinjiang, China(2013).
【23】LEDinside of TRENDFORCE Corp., http://www.ledinside.com.tw/knowledge/20120801-22324.html
【24】林武憲，「LED整合中英文斷詞斷字系統之專利分析」，碩士論文，國立台灣科技大學機械工程學系，台北，民國九十九年。
【25】McGlen, R. J., Jachuck, R., and Lin, S., “Integrated Thermal Management Techniques for High Power Electronic Devices”, Applied Thermal Engineering, Vol.24, Issue 8-9, pp.1143-1156(2004).