Research On Personalized Recommendation System Based On Linked Data

In recent years,with the rapid development of Internet technologies and social media,information loads have created tremendous challenges for consumer choice and decisionmaking.The personalized recommendation system can help people find items that they may be interested in in the complex information space and has been widely used.In practical applications,most recommendation systems generally use collaborative filtering recommendation,content-based(knowledge)recommendation,or mixed recommendation of these methods,but these methods often face data scarcity and cold-start problems in the actual application process.At present,the mainstream recommendation systems are mainly divided into two types.One is to improve the collaborative filtering recommendation system according to the newly constructed user and project characteristics,and to enhance the personalized recommendation effect.The second is to use matrix decomposition,dimension reduction,and context awareness technologies to solve data scarcity and cold start issues.From the roots,the limitation of information is the root cause of the production of the above problems.If the user/project information can be collected from multiple perspectives,the problem will be solved.However,in the traditional system,different data sources have great differences in terms of data structure and data format,and it is difficult to integrate and use them.The emergence of the concept of Linked Open Data(LOD),especially the development of related open data projects,provides an effective solution for the integration of heterogeneous data,and at the same time it provides a new solution to solves the data scarcity and cold start in the recommendation system.In this regard,this dissertation applies the linked open data to the framework of the recommendation system,links the background data in the recommender system with the rich data resources in the linked data cloud to resolves data scarcity and cold start problems from the data source,and Implicit feedback information of the linked open data to improve system recommendation accuracy.The main research content of this article is:(1)Research on personalized recommendation system framework based on associated data: This system framework was constructed and introduced from the data layer,fusion layer,algorithm layer,and application layer.compared with the traditional recommendation system and found that the recommendation system based on linked data can not only overcome data scarcity and cold start problems,but also implement cross-domain recommendations for users.However,there are two challenges in the implementation of the system.One is how to judge and link the same entity of context data and linked open data in the integration layer.the second is how to find,extract and use the implicit feedback information of the linked open data to enhance personalized recommendations in the algorithm layer.(2)Research on entity identity for data fusion: In-depth analysis of the concepts and types of entities in linked data,and found that the core of entity identity judgment in linked data fusion lies in the attribute matching between resource entities.In this regard,this dissertation proposes a method of entity identity determination based on attribute types: the attribute is divided into four types according to the relationship model between attributes and attributes,and the corresponding similarity calculation method is used according to the attribute type characteristics,and then The importance of the attribute type of the credential to entity matching was designed to determine the process of entity identity determination in the associated data fusion.Finally,the movie data set on Linked Movie Database and DBpedia was used for method validation,and the same entity on both data sets was successfully identified.(3)Research on personalized recommendation method based on linked data: In-depth analysis of the nature of association discovery,found that the more correlation paths between resource entities,the higher the degree of correlation between them,and the better the effect of association discovery.In this regard,this dissertation proposes a top N recommendation method based on the path features of the three-part diagram: taking into account the implicit feedback information of linked data can increase the number and type of recommended paths for users to alternative projects.This dissertation introduced the linked data into the collaborative filtering bipartite graph to form a trigraph data model of user-project-project attributes.After defining the model and the problem,extract the path features and property features from the three parts,and add the scoring feature to eliminate the influence of the epidemic deviation.Then use the feature ranking learning method to recommend topN recommendation to the user.Finally,compared with other algorithms on the fusion movieLens and DBpedia movie datasets,the experimental results show that the method improves the system's recommendation accuracy to a certain extent.