Research On Recommendation Technology Based On Heterogeneous Information Network Embedding

In recent years,recommendation system technology has been widely used due to its advantages in helping users find their matching information resources under various scenarios.And the applications of Heterogeneous Information Network and network embedding to recommendation systems become hot research topics.Heterogeneous information network incorporates various types of nodes and inter-node relationships,which will be used in recommendation systems with good results.Network embedding can not only obtain the low-dimensional vector representation of different nodes,but also preserve the structural and semantic information in the network.However,information in heterogeneous information networks is not effectively utilized,and most recommendation algorithms that combine network embedding in heterogeneous information networks do not exploit the heterogeneity of the network,resulting in the loss of part of the structural information in the network.To address the above issues,the main work of this thesis as follows.Firstly,this thesis proposed a recommendation model that is based on the combination of unweighted heterogeneous information network embedding and matrix decomposition algorithm.The model first uses the Path Sim algorithm to calculate the similarity to characterize the importance of different meta-paths,and it obtains the node sequence of each node by improving the random wandering based on the meta-path.Then the node sequences are learned through network embedding to the embedding representation of each node.After that,the embedding representations learned under different meta-paths are processed by fusion function,and finally the score prediction is performed through the extended matrix factorization model.The algorithm utilizes structural and semantic information in unweighted heterogeneous information networks.A lot of experiments are conducted on Douban Movie,Douban Book and Yelp datasets,and results show that the proposed algorithm has an excellent improvement in recommendation performance compared with others.Secondly,this thesis puts forward a recommendation model that is based on the combination of weighted heterogeneous information network embedding and matrix decomposition algorithm.First of all,user data and item data are used to add weight values of the relationship between corresponding nodes in the heterogeneous information network to construct a weighted heterogeneous information network.The calculated weight values can indicate the relevance between different objects.Therefore,more supplementary information will be utilized.Then the node sequences of each node are obtained in this network through random wandering based on weighted meta-paths,and the embedding representation of each node is learned through network embedding.At last,the matrix decomposition model is extended for scoring prediction.This method uses the attribute information of the object to obtain the weight value of the relationship between the objects.A large number of experiments are carried out on the Douban Movie,Douban Book and Yelp datasets with the comparison algorithm,and the results show that the algorithm proposed in this thesis can effectively improve the recommendation quality compared with other comparison algorithms.