Research On Mobile Recommendation Based On User Preference Evolution Under Multi-factor Constraints

With the popularity of smart mobile devices,location-based social networks(LBSNs)have become an important platform for users to share personal experiences and provide consumption strategies.For example,users can share their personal experiences,record their wonderful experiences,and meet new friends by check-in.This novel way of communicating facilitates the development of location-based social networking,but it also creates information overload,leaving users in a dilemma of choice.For this reason,POI recommendation system,an important application of mobile recommendation,emerges as the times require,and gradually becomes a popular research topic.However,user check-in needs to move from the current location to the location of the POI,and it takes extra time and money on the way during the travel,so that the user is not motivated to check-in.Single user check-in records are scarce,so the POI recommendation system for mobile recommendation also faces the problem of data sparsity.Making full use of user check-in data,mining and introducing more auxiliary information to compensate for the impact of data sparsity on recommendation is the focus of current point-of-interest recommendation research.This paper mainly starts from the needs of users and the influence of users’ selection of POIs,and uses various factors to obtain user preferences,so as to provide POI recommendations for mobile users.The main content and innovations of this paper are as follows:(1)A POI recommendation method based on the user’s latent intention constraints is proposed,which effectively alleviates the impact of data sparsity.This paper finds that users already have potential intentions when they need recommendation services.If the recommended POI does not meet the user’s potential intentions,they will be rejected.Further analysis finds that this latent intent varies over time and correlates with the category of POIs.On this basis,a recommendation method is proposed,which divides user preferences into two parts:time-related preferences and user unique preferences,and then uses embedding to map the two preferences and POI categories into a latent space,and adopts a weighted fusion method to determine the importance of both preferences in generating the user’s final preference.Experiments are carried out on the real data set and compared with the benchmark algorithm.The test results show that the proposed method in this chapter is superior to the benchmark algorithm in the two indicators of Recall and NDCG,and has a good effect on alleviating the impact of data sparsity on recommendation.(2)A POI recommendation method based on global and individual checkin pattern mining is proposed,which alleviates the impact of single user checkin data scarcity on recommendation.Specifically,firstly,the global check-in model is converted into global geographic influence modeling and global time influence modeling,and the global geographic and temporal influences are integrated by weighted fusion to obtain a global representation of POIs;then,according to the user’s personal check-in personalized representation of POIs for information acquisition;Then use a dynamic fusion method to determine the weight of the user’s global influence and personalized influence,and generate the final representation of the POI;finally,based on the user’s check-in information and the final representation of the POI,the user’s dynamic preference representation is obtained,and then generated recommended list.Experimental results on two location-based social network datasets show that the proposed method outperforms the benchmark algorithms in two metrics,Recall and NDCG.(3)A POI recommendation method based on group-aware dynamic graph representation learning is proposed,which constrains users to be influenced by other users within the group.Considering that users will join different groups at different times,they will be affected differently by partners in different groups.Since users with social relationships are more likely to visit POIs together,this paper builds dynamic groups based on social relationships and time.On this basis,a recommendation method using graph neural network is proposed.Based on the group-aware dynamic graph and context-aware dynamic graph,they capture the influence of group partners and obtain the user’s attention to POIs and context,and generate representations of user’ s preferences and POIs.Finally,the POI recommendation is transformed into a link prediction task between user nodes and POI nodes in dynamic graphs.Experimental results on two real location-based social network datasets show that the method proposed in this chapter achieves higher recommendation accuracy than recent representative algorithms.(4)A POI recommendation method based on latent group awareness and spatio-temporal constraints is proposed.Since there are explicit group relations and latent group relations among users,and the latent group relations are more flexible,so the research utilizes latent group user data to assist target user recommendation.Based on this situation,an adaptive group-aware topic model is proposed to mine latent communities,latent group relations,and topic distributions.On this basis,the spatio-temporal constrained context is further mined.Since the LDA model cannot model the check-in sequence,the check-in sequence is converted into the check-in predecessor and the check-in predecessor category,and the context information is fully utilized to generate POI recommendations.The experimental results on two real LBSNs datasets show that the method proposed in this chapter has higher recommendation accuracy than the baseline method.According to the above description of the research content and innovation,it can be clearly seen that the mobile recommendation research in this paper is based on the common sense that user needs and user behaviors are similar.The similarity of user behavior provides a basis for using other user check-in data to assist target user recommendation,and alleviates the adverse impact of data sparsity on recommendation.In the research,gradually restrict the range of user groups that affect target users,so that the mobile recommendation process is more in line with the actual situation of users choosing POIs.