基於隱含主題模型與詞彙本體之線上問答服務回答機制

Abstract

隨著科技的發展與即時通訊軟體的普及，有越來越多的企業提供線上問答服務，每當顧客遇到產品上的問題時，他們能夠利用線上即時問答系統來諮詢或尋求服務，而客服人員也會在深入了解顧客情況後提供建議或服務，這樣的服務模式在目前已經越來越常見。 為了要協助改善線上問答服務和分析即時聊天訊息，本研究提出了一個基於隱含主題模型與詞彙本體的線上問答服務回答機制，能夠有效的處理非結構化且長度不一的即時訊息，還能夠針對目標問題找到合適的答案。本研究機制也建立產生常見問答集，來幫助客服部門調整常見問答集之參考，實驗結果顯示本研究所提出之機制能有效找出適合目標問題的答案，提供客服人員在線上問答服務時回答顧客問題之參考。

With the development of technology and the rapid popularity of instant messengers, online question-answering service is widely provided by lots of companies. Consumers who encounter problems on products can ask questions to customer service agents through online chatting system; then agents may provide suggestions or services. This kind of service approach is more and more common nowadays. In order to improve online question-answering service and analyze chat messages, we propose a novel Answer Finding Mechanism based on Latent Topic Model and self-define ontology, which can deal with unstructured and various-length instant messages effectively and can find appropriate answers corresponding to the target question. FAQs are also generated by our mechanism as an intermediate file that can be used to assist customer services. The experimental results show that our proposed mechanism can effectively find appropriate answers for online question-answering services. The discovered answers can be provided to customer services for references.