Research On Personalized Recommendation Algorithm Based On Multi-modal Emotion Fusion

The recommendation algorithm has been widely used in recent years,but it lacks user personalization factors.Especially the user's emotional factor is less involved in the recommendation algorithm.This thesis adds the user's emotional factor to the recommendation algorithm to improve the accuracy of the recommendation.Emotional computing has developed rapidly in recent years.Because of the emotional differences of users,there are specific individual differences in the perception and understanding of emotions.The emotional tendency of users is determined to meet the needs of users.Users provide relevant personalized information,and at the same time,the acquisition of emotional value is calculated from a variety of modalities,including text emotional calculation and image emotional calculation.Therefore,this thesis proposes a personalized recommendation algorithm based on multi-modality.Taking the personalized recommendation system of movies as an example,the algorithm proposed in this thesis is discussed in detail.This thesis first proposes a multi-modal feature fusion algorithm with PSO adaptive weights,using Bi-GRU combined with attention to extract text emotional features,using attention combined with convolutional neural network to extract image emotional features,and then studying the shared semantic layer.Finally,when performing feature layer information fusion,the idea of PSO optimization is introduced,and multi-modal emotional features are weighted and fused,and the feature vector after particle swarm optimization and weighted fusion is used as the overall emotional vector.The weighted fusion of explicit and implicit emotion similarity is calculated by considering the content release time factor and the average emotional tendency.The emotion vector is calculated by the explicit and implicit emotion calculation formula proposed in this thesis,and then the personalization based on emotion prediction is studied.Recommendation module outputs top-n recommendation list in descending order of similarity.According to the algorithm proposed in this thesis,a personalized movie recommendation system is designed.The overall architecture of the proposed emotionbased movie recommendation system has three steps to deal with the recommendation task: movie emotion recognition,user emotion recognition and personalized recommendation.In movie emotion recognition,the text emotion features of the title,introduction and comment are extracted,and the emotion features of the movie poster image are extracted.Finally,the shared semantic layer vector,text and image emotion feature vectors that are optimized and weighted by particle swarm are used as the movie emotion vector.In user emotion recognition,text emotion features are extracted from the text of Weibo,and image emotion features are extracted from pictures of Weibo.The feature vector after particle swarm optimization and weighted fusion is also used as the user's emotion vector.In the personalized recommendation,the weighted fusion of explicit and implicit emotional similarity is calculated by considering the user's Weibo content release time factor and the user's average emotional tendency.The emotional vector of the movie and the user is calculated by the formula proposed in this thesis.The list of movie recommendations is output in descending order of similarity.The results prove that the accuracy of the adaptive weighted fusion feature algorithm is higher than that of the cascaded fusion feature algorithm.It can be seen that the introduction of the PSO algorithm to adaptively optimize the feature fusion weight method effectively improves the accuracy and robustness of the fusion feature;The proposed model and related benchmark test methods have personalized recommendation effects on the data set.The effectiveness of the multi-modal personalized recommendation model implemented in this thesis is better than that of the comparative experiment.According to the calculation of similarity,explicit and implicit emotional factors are added,and the effect of personalized recommendation is further enhanced.Therefore,the method in this thesis is better than the original method.In summary,the personalized recommendation algorithm proposed in this thesis is effective and practical for movie recommendation.The dimensional emotion model is successfully used to calculate the emotional similarity between movies and users.The multi-modal emotion recognition model proposed in this thesis is used for emotionbased personalized movie recommendation,which is an innovation that rarely exists in the previous personalized movie recommendation work.This thesis' s personalized recommendation algorithm has a good performance in terms of recommendation accuracy and recall rate.