Cross-domain Recommendation Based On Overlapping And Non-overlapping Users

Cross-domain recommendation is a promising solution to solve the long-standing data sparsity problem in recommendation systems.It uses relatively rich information from the source domain to improve the recommendation accuracy of the target domain.In cross-domain recommendation algorithms,most of the existing methods only consider user rating information on items in different domains,user and item tag information,and user reviews on items,and cannot effectively use the emotional information implicit in the reviews.Fully mining and using the implicit emotional information helps to solve the cold start and data sparsity problems in the cross-domain recommendation process.Emotions have an important influence on a person's behavior and choices.It clarifies what people think and how they feel about something or someone in a specific situation,and it also affects people's decision-making bias.Thus,in order to solve the problem of cross-domain recommendation under different user scenarios,this thesis carried out the following two aspects of work.On the one hand,for the cross-domain recommendation problem based on overlapping user scenarios,the paper combines the emotional information implicit in the review to propose a cross-domain recommendation method based on sentiment analysis and latent feature mapping.Specifically,this model uses sentiment analysis on user review information in different domains.Unlike previous sentiment studies,we divide sentiments into three categories based on three-way decision ideas,namely,positive,negative,and neutral.Then use the Latent Dirichlet Allocation(LDA)to model the user's semantic orientation to generate the user's latent sentiment review features.Finally,the Multilayer Perceptron(MLP)is used to obtain the cross domain non-linear mapping function to transfer the user's sentiment review features.On the other hand,for the cross-domain recommendation problem in non-overlapping user scenarios,we propose a user matching method based on entropy weight method.Specifically,given the user name,personal description,and personal homepage of the two domains respectively,our goal is to match the string similarity through different string similarity calculation methods,and finally give it through the entropy weight method.Different attribute information is assigned corresponding weights to obtain the total similarity of user profile information,and obtain the final matching result of the user.Combined with the first work,the cross-domain recommendation problem in non-overlapping user scenarios is solved by the potential overlapping user sets obtained.In this thesis,cross-domain scenarios in the Amazon dataset are taken as an example to carry out experiments and analyses on the proposed method,and the experimental results are compared with those in other literatures,verifying that the research method in this thesis is effective in recommending cross-domain scenarios.This research work provides an alternative idea to solve the cold start problem in cross-domain recommendation.