聊天機器人之研製－以PTT八卦板文章為知識庫

陳志達 ； 吳雅卉

《TANET2019 臺灣網際網路研討會》 2019年 (2019/12) Pp. 228-233

https://doi.org/10.6924/TANET.201909.0041

摘要

近年來，人工智慧一直是熱門的議題，其中聊天機器人的應用最為廣泛，且因許多軟體公司推出聊天機器人開發平台，提供眾多強大的功能，例如：自然語言處理、語音辨識、關鍵字分析等技術，甚至有連接其他通訊平台的服務，讓開發者可以更輕鬆、快速客製化出聊天機器人。因此有不少企業開始使用聊天機器人來為消費者服務，如:點餐訂購、客服服務、金融服務等，但這些聊天機器人大多數都是屬於Closed Domain的框架，使用者只能詢問特定領域的問題，且得到的回覆普遍都是制式性的回答，缺乏靈活性，使用者較不能感受到互動聊天的感覺。本研究欲實作一個Open-domain的聊天機器人，使用python語言開發，首先運用爬蟲技術抓取PTT八卦板的文章，將這些資料進行自然語言處理，再將處理完畢後的語料用來訓練聊天機器人，並建構檢索式模型（Retrieval-based model），藉此匹配問答，當使用者輸入語句時，可從模型當中找尋出與其相關之回答，並回覆給使用者。完成後的聊天機器人系統，將會發布在line平台上呈現，並達成問題檢索之有效性、快速性與趣味性之成果。

關鍵字

聊天機器人 ； 自然語言處理 ； Jieba ； Word2vec ； BM25

並列摘要

In recent years, artificial intelligence has always been a hot topic, among which application of chat robots (chatbot) are the most adopted. Many software companies have launched chatbot development platforms, providing many powerful functions, such as natural language processing, speech recognition, keyword analysis and other technologies. There are even services that connect to other communication platforms, making it easier and faster for developers to customize chatbots. Therefore, many companies have begun to use chatbots to serve consumers, such as ordering, customer service, financial services, etc. Not only save manpower and time, but also quickly analyze to understand the effectiveness of user behavior and marketing activities based on previous record of conversations, disclose information. However, most of these chatbots belong to the framework of Closed Domain. Users can only ask questions in specific areas, and the replies received are generally systematic answers and without flexibility and users can't feel the feeling of interactive chat. This study uses the python language to develop an Open-domain chatbot. First, we use the crawler technology to capture the PTT gossip articles, perform the natural language processing on those data, and then use the processed corpus to train the chatbot and construct it. The model is used to match the question and answer. When the user inputs query sentence, the relevant answer can be found from the model and replied to the user. The completed chatbot system will be published on the Line social media and will achieve the results of the validity, speed and fun of the problem retireval.

並列關鍵字

Chatbot ； Natural Language Processing ； Jieba ； Word2vec ； BM25