Similar QA Pairs Detection System to Improve Chat Quality in Turkish Conversation Groups

Abstract

Nowadays, social media applications have become one of our essential communication tools. These applications can be used to communicate individually or in groups. Retrieving information by processing the text data produced in user groups increases the quality of the chat. Consequently, creating a system that will prevent the same question from being asked multiple times in groups with thousands of users will help increase efficiency in the chat. With the help of the application developed on the Slack platform in this study, users can search for similar questions in old messages and retrieve their answers. In cases where there is no similar question in the conversation, a mechanism has been developed to direct the question that is asked by a user to a user who has knowledge of the subject. The BERTurk model is fine-tuned for sentence classification and question answering tasks, which are natural language processing techniques. Since a large amount of training data is needed, datasets created in other languages are translated to Turkish and open-source datasets are used.