Top-k Context-Aware Tour Recommendations for Groups

Cities offer a large variety of Points of Interest (POI) for leisure, tourism, culture, and entertainment. This offering is exciting and challenging, as it requires people to search for POIs that satisfy their preferences and needs. Finding such places gets tricky as people gather in groups to visit the POIs (e.g., friends, family). Moreover, a group might be interested in visiting more than one place during their gathering (e.g., restaurant, historical site, coffee shop). This task is known to be the orienteering under several constraints (e.g., time, distance, type ordering). Intuitively, the POI preference depends on the group, and on the context (e.g., time of arrival, previously visited POIs in the itinerary). Recent solutions to the problem focus on recommending a single itinerary, aggregating individual preferences to build the group preference, and contextual information does not affect the scheduling process. In this paper, we present a novel approach to the following setting: Given a history of previous group check-ins, a starting POI, and a time budget, find top-k sequences of POIs relevant to the group and context that satisfy the constraints. Our proposed solution consists of two primary steps: training a POI recommender system for groups, and solving the orienteering problem on a candidate set of POIs using Monte Carlo Tree Search. We collected a ground-truth dataset from Foursquare, and show that the proposed approach improves the performance in comparison to a Greedy baseline technique.