Research On Point-of-interest Recommendation Based On Spatial And Temporal Data Mining

With the continuous development and maturity of mobile Internet technology,the location-based social network has gradually emerged and become an important part of people's lives.Point-of-interest recommendation plays a vital role in location-based social network.It is not only a very important task in the field of recommendation,but also a significant applied research in spatial-temporal data mining.Since location-based social network often has problems such as sparse or inaccurate location information,this brings great challenges to the research on point-of-interest recommendation.In response to the new challenges and opportunities brought by noisy,short and rich social media texts,location prediction has become a new research hotspot,including user location prediction and message location prediction,which are the basic work of point-of-interest recommendation.Therefore,this article is oriented to location-based social network,and launches research based on spatial-temporal data mining.User location prediction and message location prediction are performed firstly,and point-of-interest recommendation is completed on this basis.The work in this article is done as follows:1)A method of user location prediction based on graph convolutional neural network is proposed.This method is based on TF-IDF to vectorize the representation of text features,uses user's interaction to construct a social network graph,and then uses graph convolutional neural network for label propagation and feature aggregation.This subject conducts experiments on the Geo Text dataset to verify the effectiveness of this method,and demonstrates the superiority of this method as a semi-supervised learning method in a small amount of labeled data scenarios.2)A model of message location prediction based on multi-feature fusion is constructed.The model divides the eight kinds of information in the message location prediction task into three feature categories,and the features of the three feature categories are represented by three different networks.Then the eight features are fused well by full connection.Finally,a softmax function is used to provide location prediction for message.This paper experiments on the open source dataset from the WNUT&2016,so as to verify the excellent performance of the built model.3)A point-of-interest recommendation algorithm integrating multiple impact factors is designed.The algorithm performs geographic influence modeling and social influence modeling on geographic information and social information,and combines temporal information and geographic information to model temporal and spatial influence,and then integrates the three influence scores in a weighted summation manner to obtain user's preference score.According to the user's preference score,each user is provided with a recommendation list containing Top-N points-of-interest.This paper carries out experiments on two public datasets to verify the effectiveness of the designed point-ofinterest recommendation algorithm.The experimental results show that the recommendation effect obtained by this model is better than that of the benchmark models.