Research On The Point-of-interest Recommendation Algorithm Combining Temporal And Spatial Features

With the rapid development of Location-based Social Networks and the typical problem of information overload, the point-of-interest recommendation has get wide attentions of scholars at home and abroad. Point-of-interest recommendation in location based social networks can help users to find new places effectively and bring new life experience for them. It can also help the business owners to attract more potential customers and bring more profits to them.In order to explore the features of check-in data and improve location recommendation precision in depth, we have a further research of point-of-interest recommendation algorithms by analyzing the check-in data.Firstly, we propose a point-of-interest recommendation algorithm combining temporal features and collaborative filtering. Aiming at the temporal features, we get the feature of user check-in similarity as well as temporal features of non-uniformness and consecutiveness observed from large-scale check-in data in location-based social networks. We filter users using check-in similarity and utilize cosine similarity of sequential time slots to calculate the similarity between users, and then we introduce a recommendation algorithm combining temporal features and collaborative filtering which can send some recommendations to users at a specific time. The algorithm uses sequential time slots smoothing technology to tackle the problem of data sparseness.Secondly, we propose a point-of-interest recommendation algorithm combining two-dimensional spatial feature and kernel density estimation. Aiming at the spatial features, we have the spatial features of popularity and historical locations observed from large-scale check-in data in location-based social networks. Using two-dimensional spatial features is important to point-of-interest recommendation. So we design a method of two-dimensional Gaussian kernel density estimation with popularity by exploiting the two-dimensional spatial features. And we utilize the popularity based on sequential time slots to calculate the popularity of locations. Then, we normalize and then combine the above two algorithms into a uniform framework, namely, a point-of-interest recommendation algorithm combining temporal and spatial features.Lastly, we use the public check-in datasets which contain Foursquare and Gowalla to design the experiments, which verify that the proposed algorithms can obviously improve location recommendation precision and recall than the current same type of algorithms.