TRIZ矛盾問題解構之智慧化分析模式研究

傳統TRIZ矛盾分析的研究多以人工方式來進行工程參數的判別和選擇，易造成矛盾分析解決問題的效率不佳，因而決定透過導入擴增智慧來輔助使用者在選擇工程參數的過程中，彌補個人主觀的判斷與心理慣性所造成的負面影響。本研究提出透過TRIZ理論學者Darrell Mann所提出的48個工程參數與矛盾分析為研究架構，透過發展定義矛盾的引導流程，建立辨識矛盾問題的引導步驟，找到矛盾問題點，定義與分類工程參數的正負向，作為改善參數與惡化參數的判斷依據，並透過自然語言處理中的Sentence-BERT模型，導入專利資料庫來進行語義相似度比較，搜索比對出與正負向工程參數的語義相似度最高的專利文本，分析專利文件中的文本與句子，來協助辨識敘述中的矛盾與正負向工程參數，讓工程參數在使用時能更客觀且更具有一致性，以利進行進一步的矛盾分析，達到運用擴增智慧來協助使用者篩選出最適合的工程參數的義涵。

關鍵字

TRIZ ；矛盾分析；自然語言處理； BERT ； Sentence-BERT ；專利探勘

並列摘要

The traditional TRIZ contradiction analysis is processed entirely by users to identify a problem and select the related engineering parameters. Because of different individual viewpoints on the problem, it may lead to the divergence of contradiction analysis and problem solving. Therefore, this study proposes to introduce "Augmented Intelligence" to assist users in making up for the negative effects caused by personal subjective judgment and psychological inertia in the process. Using the 48 engineering parameters proposed by Darrell Mann, we intend to develop the guiding procedure and mechanism for identifying contradiction problems, and define the possible positive and negative directions of engineering parameters based on the keywords of required functions. Applying the Sentence-BERT model in Natural Language Processing (NLP) techniques, we propose to import the patent database for cosine similarity comparison, then search and compare patent texts with high correlation with engineering parameters. Through semantic analysis of patent documents to help define improving engineering parameters and worsening engineering parameters, so that contradiction analysis can be more consistent in use, then facilitate further problem solving.