POI Recommendation Based On Users' Contextual Behaviors

With increasing development of mobile internet in recent years,more and more people are attracted to the social network.Meanwhile,the technique of GPS is becoming more advanced,leading to rapid development of the social network.Some Location-based Social Network services,like Foursquare,Gowalla,Dianping.com and so on,have become the indispensable tools,bringing extreme convenience to people's daily life.These location-based services can not only allow users to explore interest points,but also can share other geographic positions by signing these interest points.At the same time,they can also share their visiting experience about these POIs.Therefore,POI recommendation produces tremendous value to both users and merchants.POI recommendation differentiates traditional recommendation in two aspects: the first one is that the data of users' behavior is sparser;the second one is that geographic information,social relationship and other contextual information influence a lot to the recommendation results.Consequently,this paper optimizes POI recommendation from above two aspects.On one hand,we combine comment information,visual features and so on to better mine users' interests,mitigating the unfavorable effect of collaborative filtering due to the extreme sparsity of users' behavior data.On the other hand,we integrate geographic information,social relationship and other context into POI recommendation to get more accurate recommendation results.In summary,the main work of this paper is as follows:(1)We analyze the background,the significance and the current state both domestically and abroad of current interest point recommendation.Then we discuss the development trend of interest recommendation,providing firm theoretical foundations to later research.(2)We find and analyze the effect over recommendation results with geographic information,social relationship and comments of users.This paper combines comment information and social relationship to further boost the performance of POI recommendation based on the original Geo MF model.(3)We work on the behavior pattern of users in the travelling circumstances,mining the effect over the span of time when they visit interest points and the visual features of photos generated by users.After that,we recommend users other interest points in which they are probably interested,and then we recommend users an itinerary with a set of ordered POIs under a series of constraints about the travel.Finally,this paper implements experiments on POI recommendation and itinerary recommendation respectively based on real datasets.Compared with current the-state-of-art methods,our method outperforms them with multiple different criterions,thus verifying the effectiveness and reliability of our method.