Research On The Algorithm Of Point-interest Recommendation Based On Learning To Rank

With the development of artificial intelligence technology,the rapid increase of information and the rapid development of social networks,the recommendation of interest points based on social networks has become a new research direction.At the same time,it also faces many problems,such as sparse user check-in matrix,low recommendation accuracy,insufficient information in the context and so on.In order to alleviate the cold start problem and improve the recommendation accuracy,this paper proposes fusion Matrix decomposition model of social information and geographic information and list level sorting learning algorithm based on List MLE algorithm,including:(1)In order to better mining the implicit feedback behavior,the BPR(Bayesian personalized rating)model is selected to optimize the process of matrix decomposition.In the traditional BPR model,the signed in and the non signed in interest points are regarded as the generation strategy of the partial order relationship,and the partial order relationship between the signed in interest points is ignored.In this paper,the frequency of the check-in and the score level are added as the generation strategy of the partial order relationship,so as to fully mine the user's preference for the interest points.(2)In order to improve the recommendation accuracy and alleviate the cold start problem,select social relations and improve the traditional calculation method of social relations.Traditional social relationship calculation is determined by user similarity.This paper designs a recommendation model based on social relationship by fusing and improving the calculation method of trust.(3)The location of social network is an important factor that affects users' choice of interest points.Finally,the BPR model is integrated with the model based on social relations and geographic information,and finally the recommendation results are generated for users.(4)Aiming at the problem of low recommendation accuracy and high model complexity of point level and pair level sorting learning technology in location-based social network,this paper compares and analyzes the advantages and disadvantages of loss function and algorithm complexity of list level sorting learning algorithm,selects List MLE algorithm with better complexity and loss function attribute to apply to point of interest recommendation,and scores based on Function structure of List MLE algorithm,social information into the scoring function calculation process.In order to solve the problem of users' attention to different positions of the list in the list level ranking learning,cost sensitive learning is applied to the scoring function to give different weights to different positions of interest points in the list of interest points.The experimental results show that under the real data sets of Gowalla,foursquare and yelp,the algorithm proposed in this paper improves significantly in mining implicit feedback and improving the recommendation quality compared with the baseline interestpoint recommendation algorithm,and the algorithm is superior to the traditional interest point recommendation algorithm in accuracy and recall.