| Summary: | 碩士 === 國立臺北大學 === 資訊管理研究所 === 107 === In recent years, the business scale of Taiwan insurance market has been booming. In 2018, hundreds of thousands of employees work in the insurance industry, when they are aware of their needs, some people using search engines to search document and others start to ask questions in the "Community Question Answering" (CQA) and wait for other users to answer. Another common approach is to find similar questions directly in the history of question answering corpus in the CQA, but finding out similar questions for users in the large data sets is a huge challenge. This study builds a question and answer corpus from the CQA, and integrates many information retrieval strategies to complete the insurance domain question and answer system. The strategy includes query expansion, word embedding, text similarity, traditional BM25 retrieval method. Finally, this study improves the defect that the IDF in BM25 does not conform to the actual application scenario, and proposes to establish an insurance important word weight method to improve IDF.
|
|---|
