Research On Group POIs Recommendation Fusion Of Users' Gregariousness And Activity In LBSN

With the development of mobile terminal services,location-based services have become one of the core services of the mobile Internet industry.In the current large data environment,how to establish human-land relationship in the multi-dimensional location-based social network(LBSN),to study the location of the user and the crowd,regional location,etc.will form an important reference value on the business activities.In current existing research,the use of LBSN for personalized recommendation has been relatively mature.However,the rise of social networking makes it easier for people to participate in group activities.How to extract effective information from mass information and recommend common and reasonable points of interest for everyone in the group according to group preference is the next goal of recommendation technology.Thus,group recommendation is generated.Group recommendation techniques are based on personalized recommendations.The sign-in data in the LBSN personalized location recommendation does not include item direct fraction and therefore the user rating strategy has been improved in this thesis.In addition,the inherent association of the check-in data is analyzed,which is used to improve the aggregation strategy in the group recommendation system.Following three aspects are the focus of this thesis:(1)the use of appropriate strategies for LBSN users to establish a model,and according to the model to calculate user similarity.In the user-based collaborative filtering technique,predicted score of the unattended item is obtained according to the score of the nearest neighbor,and the data provided by the LBSN doesn't give the corresponding score directly.Therefore,this thesis improves the scoring strategy by using the text retrieval idea.An experimental comparison between the improved scoring strategy and the traditional scoring strategy has been made,successfully verified the effectiveness of the improved scoring strategy.(2)After obtaining a single user rating,aggregating them to form a group score is needed.The existing two aggregation methods are the aggregation individual recommendation method and the aggregation individual preference model method.Both aggregation methods require appropriate aggregation strategies.The existing average strategy is the most widely used strategy,but lack of dynamic process.So the improved weighted average strategy is used.Weighting individual users,taking into account influence of users in the group and the amount of information they provide.Thus,a weighting method based on user cluster degree and activity is proposed.Weighting the user then obtaining the group item score,make the final group recommendation.Then the improved weight fusion strategy is compared with the traditional fusion strategy by experiment to verify that the former strategy improved the recommendation accuracy.