以AI語言模型優化專利關鍵字檢索

Abstract

當前的專利檢索技術，檢索者需要準確地列出相關關鍵詞，以提高檢索的精確性。專利文件通常使用較為抽象的概念詞彙撰寫，且同一技術元件可能會以不同的詞彙表達，這使得關鍵詞的列出往往無法精準匹配。即使投入大量的時間和人力，檢索者也很難找到所有相關的關鍵詞，以覆蓋所需的專利範圍。 本研究聚焦於應用 AI 語言模型於專利檢索的實踐，分析其與傳統專利檢索方法的差異，探討是否能夠透過 AI 語言模型優化現有專利檢索關鍵詞議題，以期節省檢索所需的人力與時間，實現更高效、更完整的專利查找方式。 AI 語言模型將文檢索轉化為基於 AI 向量值的相似度匹配。所以 AI 語言模型的檢索需要進行前置處理，首先需匯入專利資料並建立「專利 AI 向量資料庫」。當後續進行 AI 專利檢索時，查詢詞需轉換為 AI 向量值，隨後與「專利 AI向量資料庫」進行相似度比對，根據相似度的高低順序產生與專利案件的關聯的案件結果，最後再透過相關性判斷結果的關聯性及正確性。本研究使用兩個案例透過以上的方法分析及比較 TIPO 與 AI 語言模型的檢索結果，驗證 AI 語言模型在專利檢索提供「更準更全」的能力。期望透過 AI 語言模型的優化，為專利檢索帶來革新，並探討其商業化的可能性。

Current patent search techniques require searchers to accurately list relevant keywords to improve the precision of the search. Patent documents are often written using more abstract conceptual vocabulary, and the same technical component may be expressed with different terms, making it challenging for the listed keywords to match precisely. Even with considerable time and effort, it is difficult for searchers to find all the relevant keywords to cover the required range of patents. This study focuses on the practical application of AI language models for patent searching, analyzing the differences from traditional patent search methods, and exploring whether AI language models can optimize current issues with patent search keywords. The goal is to save the manpower and time required for searching, and to achieve a more efficient and comprehensive method of patent retrieval. AI language models transform text retrieval into similarity matching based on AI vector values. The retrieval of AI language models requires preprocessing; initially, a "Patent AI Vector Database" must be established. When subsequent AI patent searches are conducted, the query terms are transformed into AI vector values, followed by similarity matching with the "Patent AI Vector Database." Results are generated in the order of similarity, associating them with related patent cases. The final step involves manually reviewing and verifying the relevance and accuracy of the results. This study analyzes and compares the search results of TIPO and AI language models through two cases using the above methods, verifying the AI language model’s ability to provide "more accurate and comprehensive" patent retrieval capabilities. By optimizing AI language models, it is hoped to revolutionize patent search and explore its potential for commercialization.