The Planfree project studies how to improve recommendations for social leisure activities. Recommendations for social leisure activities are very different from the well-studied recommendations of products in e-commerce website or recommendations of movies. In our case, the people choice of places for leisure activities depends on some social aspects, such as the type of relationship that links them with their companions.
We started by focusing our attention to restaurants, i.e. places in which people can have dinner with friends, relatives and other people. Instead of collecting one rating for each restaurant, expressing the overall preference for the place, we let our users specify if they like or not a place according to different situations, that we call purposes:
- bringing out some visitors;
- having a romantic dinner with their partner;
- going out with friends;
- for a lunch break (where price/quality ratio is important).
We already demonstrated how these different purposes really identify different aspects of restaurants and produce different ranks. We collected ratings for Trento's restaurants and compared the average-based resulting ranks for each purpose, finding that they are very different . In the following table, the top 10 restaurants in our list according to an external source, i.e. TripAdvisor rank at the time of the study, are compared with their positions in the purpose-specific ranks. The total number of restaurants in our list is 23.
As can be seen, Visitors and Partner columns are very similar and most of the restaurants are still in the top 10 positions. When considering Friends, some places started to move below the middle of the rank, but the most surprising results are in the Lunch break column. The restaurant "Le due spade", which is in the first position for TripAdvisor, is in the last position in our rank and the restaurant "Da Andrea", which is considered the best in our rank, was not even present in TripAdvisor.
We are currently studying which personalized recommender algorithm better works with our data and better predicts users' tastes, keeping an eye also on its performance and applicability in an online context.
- Beatrice Valeri, Marcos Baez and Fabio Casati. ComeAlong: understanding and motivating participation to social leisure activities. International Conference on Social Computing and its Applications (SCA), Karlsruhe, Germany. Sept 2013.