Towards a recommender system for bush taxis

To improve the transport efficiency and to reduce the traveller stress, we introduce a recommender system for bush taxis in Ivory Coast whose main objective is to propose to pedestrians potential means of transportation in their neighborhood whose destination match their own destination. The prediction of the next location relies on a mobility model called Mobility Markov Chain. One of the strength of the proposed recommender system is that it is fully automatic as a user does not need to explicitly express his next destination but rather the system tries to infer it based on his past mobility behavior. Moreover, the recommendation algorithm is biased towards suggesting the means of transportation that are the cheapest (if one is available). The preliminary evaluation of the recommender system conducted on one of the D4D dataset shows that approximately 99% of the time in less than 30 minutes, the system is able to suggest a mean of transportation that is at most 1 kilometer away from the current position of the user for an accuracy of the prediction of the next location that is between 30% and 50% depending on the complexity of the mobility behavior of the user considered. Keywords-Location-based service, Recommender system, Next place prediction, Mobility Markov chain.