Towards Context Integration in Content Based Recommender System for Smart Tourism

Recommendation systems (RS) are now essential in various sectors of daily life, especially in tourism, where they assist tourists in making better choices about which points of interest (POIs) to visit. However, these RSs face a number of challenges, including the risk of a cold start when a new POI is taken into account, and the problem of tourist dissatisfaction with recommended POIs. To address these issues, we focused on Content-Based Recommendation Systems (CBRS) that mitigate the problem of data sparsity and integrate contextual information from tourists during their visits. In this paper, we refined tourist feedbacks using contextual variables like “time” and “companion” during the visit. Next, we implemented a CBRS using the vector representation of POIs with the Term Frequency/Inverse Term Frequency (TF/IDF) method to compute similarity between tourist profiles and POI characteristics. With this type of similarity, our system can run three variants of CBRS in parallel: the first ignores the tourist context, the second incorporates the “temporal context”, and the third takes into account the “companion context”. Finally, to compare these three recommendation variants, we used an online evaluation to calculate the Click Through Rate (CTR) metric. According to our initial experiments, the CBRS with the integration of temporal context outperforms the other two implemented RS.