以BERT模型為基之美食聊天機器人之研製

近幾年的科技發展越來越快，社會的風氣也不再像以前一樣保守，許多人開始會在網路上發表自己的言論及看法。當我們使用過的產品或品嘗過的美食，都會想把自己的想法或看法寫在網路上跟大家一起分享，不管評論的好與壞。如果可以隨時分析評論的好壞，就可以讓我們更加快速地去檢視這一間店家，並將這一些資訊提供給業者或其他的使用者。由於現代人手機不離身，又加上平凡的使用LINE等通訊軟體，因此本研究希望可以建構出一套可以儲存使用者評論的機器人，並且將使用者輸入的評論，利用Bert-base-chinese模型進行分類藉由BERT的模型可增加美食正負評論分類的有效與正確性，而後將分類完成的資料存入資料庫內。系統再利用LINE在2016年推出的Messaging API建構出一套可以讓使用者新增美食評論使系統的參考性提高以及找尋店家評論資訊的聊天機器人。

關鍵字

聊天機器人； Bert ； LINE Messaging API

並列摘要

In recent years, technology has developed faster and faster, and the social behaviors are no longer as conservative as before. Many people begin to express their opinions and opinions on the Internet. When we have used products or tasted food, we all want to write our thoughts or opinions on the Internet and share them with everyone, no matter what the comments are good or bad. If we can analyze the quality of the reviews at any time, we can check the store quality more quickly, and provide this information to the web master or other users through mobile phones and communication software such as LINE. This research will build a set of Line chatbot that can store user comments, and use the BERT-base-chinese model for the comments entered by users. The model of bert can increase the validity and accuracy of the classification of positive and negative food reviews, and then store the classified data into the database. The system uses the Messaging API launched by LINE in 2016 to construct a set of chatbots that allow users to add food reviews to improve the reference of the system and find store review information.