Utilizing Multilingual Social Media Analysis for Food Venue Recommendation

In this paper, we propose a food venue recommender system which utilizes social media information to determine crowd preferences in a multilingual context. To recommend an appropriate food venue for each language user, the system ranks venues in each region based on the popularity of the venue for that language user (by analyzing differences between locations and languages in geo-tagged tweets). In addition, explicit user rating data, as well as sentiment analysis data from nearby tweets are also used to represent the quality of the venues. To examine the feasibility and performance of our proposed method, approximately 26 million tweets from different European countries were collected and analyzed in this study. Afterward, we provide an evaluation of the ranked venues proposed by the system based on French language users in several European countries. Four different recommendation approaches were evaluated. These includes (1) a baseline approach (language independent popularity) (2) the proposed approach (language dependant popularity) (3) the proposed approach with explicit rating data and (4) the proposed approach with sentiment data. Overall, the results showed that our proposed approach out performed the baseline approach. However, the value of adding an explicit rating and sentiment data to the ranking process was not clear. The sentiment based approach out performed our proposed none-adjusted approach in only three out of five cities and the rating based approach only in two out of five cities.