透過您的圖書館登入
IP:3.133.86.172
  • 學位論文

自動化建構技術功效矩陣以供專利策略分析: 使用文字探勘技術

Automatically constructing technology effect matrix for patent strategy analysis: Using text mining techniques

指導教授 : 魏志平

摘要


專利對於企業或國家已經越來越重要,已經成為一個重要的資產。專利可 以作為解說一項特別技術的文件, 我們可以經由分析一大群文件來了解特定技 術領域的發展過程和狀態. 專利具有排他權, 可以保護個人或企業的知識資產 不為人所侵犯, 也因此引申出許多專利策略和佈局的議題。所以要維持或提昇企 業和國家的競爭力, 分析專利是一項不可或缺的工作。專利視覺化可以將專利的 原始資料變成有用的資訊,專利地圖的製作就是一個視覺化的過程,它們可以 讓人們能更快速了解分析現狀與解釋結果。其中一個重要的專利地圖就是技術 功效矩陣。它擁有兩個維度,一個歸納專利群集中使用的技術類別,另一個維 度就是這群專利想要達到哪些功效或是目標。擁有這樣的矩陣,可以清楚的知 道哪裡是還沒有很多技術開發的”新機會”,也可以知道哪邊是大部分技術集中研 發的熱區。因此可以經由此結果來做對應的策略,如迴避設計(design around), 開 發新技術與遠離專利地雷。但是技術功效矩陣的建立是很耗費人力與時間的, 從科技政策研究與資訊中心(STPI)對每個領域的研究來看,也花費很多財力去分 析它們的技術和功效。 因此,我們想要提出一個完全自動化的方法建立技術功 效矩陣,並以檢視我們系統的結果與真實資料做比較。我們使用文字探勘和自 然語言處理的技術去挖掘技術與功效相關的關鍵字來代表每個專利。我們並利 用 Fuzzy C-Means 將專利分成會重疊的技術和功效群組。我也會測試三種不同 的分群驗證(cluster validity)方法試圖找出最佳的技術和功效群數。我們的方法在 完全自動化的情況下,對於五個技術領域還保有不錯的結果,能有效協助專家 建構不同領域的技術功效矩陣。

並列摘要


The patent is getting more and more critical to enterprises, and it is becoming an important asset in the world. Patent can represent as a technical document, and we can understand the technology development status and formulate strategies by analyzing them. Patent map can visualize some vital information for people to comprehend the status easily and quickly, and the technology effect matrix is one of important patent map. It has two dimensions, one is what technology the patent use and the other is what effect the patent can reach. We can find what most patents develop on, called “hot spot” and what direction is still lack of developing, called “vacancy” or “technical opportunities” in specific technical field. T/E matrix needs many efforts to construct and there are still few of related researches, so we want to propose a method to automatically group patents into many clusters based on hidden technology and effect information inside them. And our method can extract tech and effect feature from patent in general, and use Fuzzy C-means to cluster them. We also apply three cluster validity methods to decide the tech and effect category number. We don’t just focus on one case; we evaluate our result on five real world technical fields that their T/E matrixes are from STPI. And we reach a good result finally.

參考文獻


Blei, David M., Ng, Andrew Y., & Jordan, Michael I. (2003). Latent Dirichlet Allocation. Journal of Machine Learning Research, 3, 993-1022.
Bolshakova, N., & Azuaje, F. (2003). Cluster Validation Techniques for Genome Expression Data. Signal Process, 83(4), 825-833.
Chen, Yi-Chun. (2009). Patent Valuability Prediction. (Master), National Tsing Hua University.
Cheng, Tien-Yuan. (2012). A New Method of Creating Technology/Function Matrix for Systematic Innovation without Expert. Journal of Technology Management & Innovation, 7(1).
Chuang, Shui L., & Chien, Lee F. (2005). Taxonomy Generation for Text Segments: A Practical Web-based Approach. ACM Transactions on Information Systems, 23(4), 363-396.

延伸閱讀