Evolutional Computation Based Topic And Location Aware Recommender Algorithm

The purpose of the recommender system is to provide users with personalized online products or services to address the growing information overload problem.Since the mid-1990 s,a variety of recommender system technologies emerged,and software developers have designed a variety of recommender system software applications to a variety of recommendation scenarios.With the successful application of the recommender system in different areas,researchers and managers recognize that the recommender system provides tremendous opportunities and challenges for the enterprise,government,education and other areas.In the development of the recommender system,the two important aspects are: the continuous innovation of various recommendation algorithms and the full use of emerging data dimensions.For example,the collaborative filtering algorithm distinguishes a particular user may be interested in products through other similar users of which products interested in the analysis,rather than the analysis of interest itself.Then the hidden topic model is directly on the users and the distribution of the topic of the product analysis.With the development of mobile Internet,in some scenarios,geographic information in the recommender system occupies an increasingly important role,so the geographic information based recommender system has a boom development.Based on a large number of open data sets,this paper makes an in-depth study on the application of optimization theory based on evolutionary computation in the recommender system,including the use of evolutionary algorithms to optimize the specific targets in the recommender system,and effectively integrate the geographic information of users and commodities tag information and the use of new models to tap the distribution of users and item topic.The main work and contribution of this dissertation are as follows:-We proposed an algorithm based on evolutionary theory: in the recommender system,make full use of the user-item rating,the label of the items and the user's geographic information.The research on the rating is always accompanied by the development of the recommender system,and the algorithm is relatively mature.The commonly used algorithm is the matrix decomposition and the latent Dirichlet distribution.There are two types of processing,which are related to the geographical information mining.One method is geographical cluster of users.In this dissertation,the evolutionary theory is used to fuse the three methods.The information of the user's score and the information of the item tags are processed by the latent topic model.The distribution of users and items is obtained.Then,users are clustered according to the above subject distribution and the geographical distance as a penalty to establish the objective function,with the evolutionary algorithm to optimize.-A method of mining both rating information and tag information is proposed.In this dissertation,we use latent topic mining algorithm,and design a new latent topic mining algorithm based on tag information.Because only the items are tagged,in the classic tag based latent topic model,only the distribution of the topic of the items is calculated.We establish a new model under the scene of music and movie recommendation,the user and the regional distribution of the topic is mined.-A cold start method based on regional topics is proposed.Clustering of users is not only an effective way to deal with data sparse problems,but also ease the cold start problem,the general cold start recommendation is to recommend the most popular items to the cold start users.The commonly used cold start recommendation strategy is that the querying cluster is recommended the most popular items to all cold start users in this cluster,meaning that all cold start users in the same cluster are recommended with the same recommendation list.This article draws on this idea,the use of regional topic distribution,get the current regional candidate products,and recommended to the cold start users.