| With the development of the Internet,recommendation algorithms play an increasingly important role in people's daily lives.Mining users' interests and needs from massive data,and make personalized recommendations to users has become an important research topic.Researchers have proposed a variety of recommendation algorithms through in-depth research,including collaborative filtering recommendation algorithms,content-based recommendation algorithms,and hybrid recommendation algorithms.Among them,a hybrid recommendation algorithm that comprehensively considers scoring and item content can fix limitations in a single recommendation algorithm.And the thinking of big data has also made researchers more committed to using multidimensional data to express user interests in detail and form more personalized recommendations.Therefore,one of the research purposes of this article is to comprehensively use multiple information to improve the recommendation effect.Through a large number of literature studies,we have found that fine-grained sentiment analysis of user review texts can enhance the expression of user interest preferences,and the use of knowledge maps can make good use of relationships between entities to represent heterogeneous information,and existing studies have separately proven that the use of fine-grained sentiment analysis of review text for recommendation and the use of knowledge graph representation learning for recommendation have a positive effect on the improvement of the recommendation effect,however,no research has been found to full use the two methods together to design the recommendation algorithm.Therefore,the main research content of this paper is based on in-depth research on recommendation based on fine-grained sentiment analysis and recommendation based on knowledge map,and propose an effective hybrid recommendation algorithm to improve the recommendation effect and verify it through experiments.The specific research content includes the following four parts:(1)This paper first studies the fine-grained sentiment analysis method.The finegrained sentiment analysis task is divided into two tasks: evaluation object extraction and sentiment determination.The evaluation object extraction task uses deep learning combined with conditional random domain models.Method,and introduced BERT pre-trained language model,and proposed to use BERT-Bi LSTM-CRF model to extract evaluation objects.Because BERT pre-trained language model has a better performance,the BERT model is used again in the sentiment determination task.Attention mechanism judges the emotional tendency of the subject;(2)In the design of the recommendation algorithm based on sentiment analysis,firstly use the LDA topic model to divide the attribute surface of the evaluation object,calculate the user's sentiment score on each attribute surface,and combine the idea of collaborative filtering by calculating the user's sentiment score Recommend similarity;(3)In the design of the recommendation algorithm based on knowledge map,firstly extract the reasoning path from the user to the project through the knowledge map,combine the Trans R method to learn the knowledge map triplet,and input the inference path as a vector into LSTM.Combining attention mechanism and pooling operation to capture the semantics of the inference path,and finally score prediction through the fully connected layer and sigmoid function.(4)Fusion recommendation based on sentiment analysis and recommendation based on knowledge map,and explore the fusion effect of the model.This paper uses the real data set in Douban movies to conduct experiments.The experimental results respectively prove the effectiveness of the fine-grained sentiment analysis proposed in this paper,and the hybrid recommendation algorithm designed by combining sentiment analysis-based recommendations and knowledge map-based recommendations can significantly improve the accuracy of recommendations and improve the effectiveness of recommendations. |
PDF Full Text Request