Developing a Turkish-language recommendation system based on user conversations

Abstract

Recommendation systems have recently been used in many fields. This study describes a restaurant recommendation system which is developed specifically for data that is collected from chat messages typed in Turkish. The proposed system aims to recommend best matching places to a group of users in a chat environment analyzing their conversations. In order to achieve this goal, a rule-based approach which composes of normalization, analysis and recommendation steps has been designed and implemented. Furthermore, an explanation module used for explaining why the system recommends selected places has been added. The system benefits from two data sources that are property data source and restaurant data source and a rule base. While the property source is a dataset contains features related to restaurant domain, the restaurant source has all places that can be recommended by the system. On the other hand, the rule base is a sequence of rules defined manually to extract information from chat messages in a more accurate way. The evaluation process of the system has been very difficult since no test data are available. To evaluate the system, both restaurant data source and chat messages are simulated manually.