Heterogeneous Information Network Embedding Recommendation Model Fusing Multiple Scores

In the era of information overload,people's daily life is increasingly inseparable from recommended technology.Therefore,the improvement of the performance of the recommendation system has great social significance.Because heterogeneous information networks have natural advantages in dealing with multiple nodes and different edge types,this thesis studies the recommendation system from the perspective of heterogeneous information networks.Since the existing heterogeneous information network recommendation models are mostly based on the semantic relevance of meta-paths,advanced semantics cannot be fully utilized effectively,especially the high-level semantic information of multiple implicit scoring of users and projects.The extraction is too simple and rough.Therefore,the main purpose of this thesis is to make full use of multiple implicit scoring and integrate multiple user evaluation data into the recommendation model based on heterogeneous information network,so that the existing recommendation model based on heterogeneous information network can not fully utilize advanced the problem of semantics ultimately improves the accuracy of the recommendation.In this thesis,the sentiment analysis technology is used to quantify the implicit evaluation information,which is convenient for the next recommendation model learning.Simultaneously,a multi-score fusion framework model is proposed,and the model training is carried out in a multi-score fusion manner.The method proposed adopts the framework model of “probability matrix decomposition + heterogeneous information network + network embedding + sentiment analysis + multi-score fusion” to achieve the purpose of learning the low-dimensional features of users and projects,and finally obtain the prediction result.The main research contents of this thesis are as follows:First,for the recommendation model based on heterogeneous information network cannot fully utilize the high-level semantics,this thesis first systematically expounds the relevant principles of recommendation technology,and then compares the advantages and disadvantages of the existing recommendation technology.In addition,the theory of heterogeneous information network,network representation learning,sentiment analysis and so on related to the proposed model is introduced in detail.Second,in order to alleviate the problem of insufficient utilization of multiple evaluation information in the process of user and project interaction,this thesis proposes a multi-score heterogeneous information network embedding recommendation model.The model is based on the implicit scoring fusion framework.Firstly,the meta-path instance is extracted from the heterogeneous information network,and the node features are generated through the network embedding.Then,implicit score is generated by quantifying the implicit evaluation information using emotion analysis.Finally,generate recommendation results by incorporating the aforementioned two models.At the same time,the HERec method has also been supplemented and improved in dealing with the inadequacy of text information.Last,in this thesis,the recommended model proposed is verified by crawling the movie data in the watercress as the experimental data set.The results show that the proposed model is better than other methods.What's more,under the condition that the data set is under the 20%,the performance of the proposed model is more superior.