Movie Knowledge Graph Recommendation And Topic Research Based On Sentiment Analysis

Under the influence of global information technology,the film and culture industry has continued to develop,contributing greatly to the growth of social benefits on the one hand,and assuming the role of a cultural exchange carrier in the cultural exchanges of various countries on the other.With the development of mobile internet technology,many online film and television platforms have rapidly emerged,setting off an online film and television revolution.Thanks to breakthroughs in data transmission technology,more and more film resources can be distributed and shared online,and a large amount of film information can be quickly presented to users in one terminal.But at the same time,the problem of "information overload" caused by the dramatic increase in the number of online movies exists on all platforms,and users are often overwhelmed by the huge number of movies that come their way and are unable to quickly choose the ones that interest them.A movie recommendation system can analyse the movies users like and actively recommend them,which not only saves users’ time in searching for movie information,but also makes them dependent on the recommendation system and increases user loyalty.Therefore,in order to improve the problem of "information overload" of movies,this thesis takes movies as the main object of research,takes the emotion of movies as the main entry point,takes the IMDB movie dataset in Kaggle website as the data base,and integrates the LSTM model and KGCN model to propose a movie knowledge graph recommendation model based on sentiment analysis KGCN-S,on the one hand,compared the performance of KGCN-S model with other three knowledge graph recommendation models,and the results showed that KGCN-S model performed the best in four indicators,namely AUC,ACC,F1 and recall rate;on the other hand,the sentiment was dichotomized,trichotomized and quinichotomized according to the sentiment bias probability,and the knowledge graph was constructed respectively,and then KGCN-S model was applied to The accuracy of each sample data was calculated,and the ACC of the samples with three and five categories of sentiment characteristics was found to be significantly better than that of the samples with two categories of sentiment characteristics on the KGCN-S model by means of statistical tests.In addition,to explore the audience’s favourite movie themes,this thesis also classifies the movie dataset in two dimensions: sentiment and rating,and then performs LDA theme mining to mine multiple movie themes under different sentiments and ratings,from which the users’ preferred movie themes are summarised.Through the analysis of the above results,this thesis summarises the following conclusions: 1.sentiment and sentiment richness play a significant role in improving the recommendation performance of the movie knowledge graph;2.for documentaries,users are more likely to give good ratings;while for horror films,users are more likely to give poor ratings;3.users prefer relevant themes that are difficult to reach and imagine in daily life;4.for easy to reach and imagine in daily life related themes,users prefer to watch stories from films that are emotionally opposite to the themes.The research conducted in this thesis improves the movie "information overload" problem,enriches the movie knowledge graph system,improves the movie knowledge graph recommendation performance,summarises the thematic characteristics of users’ favourite movies,and finally provides feasible suggestions to movie recommendation platforms and movie creators based on the conclusions drawn from the research.