Recommendation Research Based On Fusion Of Continuous Emotion Sequences And Movie Semantic Features

The development of the movie market is getting stronger and stronger,and the overload of movie information is a kind of trouble when people face the selection of movies.In recent years,the research on movie recommendation is gradually advancing.Collaborative filtering movie recommendation relies too much on user historical data,and has data sparsity and cold start problems.Content-based recommendation cannot be applied to new users,and movie features require a lot of manual annotation,and it is difficult to mine users’ new points of interest.The difficulty of recommendation based on knowledge graph is that it is difficult to define how to build a good knowledge network for different scenarios.Demographic-based movie recommendation has the risk of leaking user privacy,and the excessive granularity of recommendation cannot ensure the accuracy of recommendation and cannot achieve personalized recommendation.Although the hybrid recommendation method solves the limitations of a single recommendation,its implementation is too complicated.Movie review mining and summarization is one of the challenging tasks in natural language processing.In addition,only extracting movie introductions and movie reviews cannot effectively obtain the most essential feature of movies—movie semantic features.In most cases,users cannot have enough time to watch the whole movie,and in a movie,the most impressive part of the movie is often the core of the movie.Therefore,by extracting features from the movie And the emotional change sequence of the user watching the clip,the user’s interest in the movie can be modeled,and then the limitation of the traditional movie recommendation model for interest modeling can be solved.Based on the above discussion,the motivation of this paper is that the user’s interest in movies needs to be considered in combination with the semantic characteristics of the movie itself and the changes in the user’s own subjective feelings.To this end,this paper collects ten different types of wonderful movie clips,invites150 school students to watch multiple clips at random,and records their emotional changes in the process of watching clips in the form of video.Semantic-level feature extraction is performed for each movie segment using the latest movie feature extraction algorithms.Use the popular emotion recognition model to extract the sequence features of students’ emotional changes.The movie features are fused with the user’s movie-watching emotional change features,and the serialized recommendation model is used for modeling,and the user’s real score is fitted for training,in order to obtain a better interest degree prediction model.A comparative experiment is conducted to prove that the cross-domain feature fusion model designed in this paper is effective through ablation experiments and using some traditional recommendation models.