Highlighting the user's primary criterion in text mining-based recommendation system

Mariem Briki, Sabrine Ben Abdrabbah, Nahla Ben Amor

Abstract: Recommendation systems (RS) are proposed to detect and suggest items that best meet the needs of users. Most of the existent works are based on recommendation algorithms are proposed to capture numerical data. Therefore, this may not provide consistent results (i.e. users may not have the same rating scale). The semantic web is proposed as an extension within the framework of RS in order to process data which are now the most used on the net, unstructured data (i.e. textual data). The unstructured data is used for the purpose of improving the personalization task in different application fields. As a matter of fact, textual data analysis has got the ability to reveal hidden information which enable us to further capture user's likes, preferences and interests and as a result a correct way to determine the user's profiles. However, the existing solutions unfit to take into account the multi-criteria feature of items which is essential to represent more complex user's preferences. This work proposes a text mining-based RS that exploit the users' comments in order to highlight the important criterion of the active user. Finally, the recommendation task is based on the criterion chosen thought exploiting the user's reviews. The choice of the criterion takes into consideration its importance regarding the percentage of its existence into the items reviews content. Several enhancement versions of this work were proposed that were tested on a real database extracted from the TripAdvisor website. The accuracy and quality of the recommendation has been improved considerably in this work by compared with traditional approaches.