Yang, Yongmao. Hybrid movie recommendation system using a log likelihood content comparison approach and collaborative filtering user analysis. Master's Degree(Computer Engineering). Chiang Mai University. Library. : Chiang Mai University, 2025.
Hybrid movie recommendation system using a log likelihood content comparison approach and collaborative filtering user analysis
Abstract:
Recommendation systems assist users in filtering qualified data from vast datasets. And the recommendation algorithm based on content and collaborative filtering has been widely used in the current recommendation field. The content-based filtering approach is prone to the sparsity and the recommendation result is single and lacks innovation, while the collaborative filtering algorithm reduces the recommendation accuracy due to the sparsity problem. To overcome the weaknesses of these, most research proposed hybrid systems, where the sparse rating matrix is decomposed into a product of two low-dimension matrices to mitigate the sparsity issue. Unfortunately, the decomposition causes an information loss, which incurs error. we proposed three hybrid recommendation system for mitigating these problems, which include Hybrid Movie Recommendation System with User Partitioning and Log Likelihood Content Comparison, Hybrid Movie Recommendation System with Content-Based and Memory-Based Collaborative Filtering based on Deep Neural Network, and Hybrid Movie Recommendation System with Content-Based and Memory-Based Collaborative Filtering Using Similarity Graph.In Hybrid Movie Recommendation System with User Partitioning and Log Likelihood Content Comparison section, we proposed a novel hybrid recommendation system to solve cold-start and sparsity problems as well as compensating information loss to mitigate the error. The sparse rating matrix is decomposed by alternating least squares (ALS) into hidden user feature vectors and hidden movie feature vectors. We use users' age, gender and movie genre preference information to mitigate the cold-start issue in the hidden user feature vectors. Also, we replace the term frequency-inverse document frequency (TF-IDF) with log-likelihood in word weighting to mitigate the cold-start issue in the hidden movie feature vectors. Compared with the conventional (ALS) matrix factorization algorithm, our proposed hybrid system obviously gives less Root mean square error (RMSE).In Hybrid Movie Recommendation System with Content-Based and Memory-Based Collaborative Filtering based on Deep Neural Network section, we propose a movie recommendation system based on deep neural networks and user vocabulary preference features to alleviate cold start and sparsity issues, reduce prediction errors, and improve recommendation efficiency. We evaluate our model using hit rate (HR) and average reciprocal hit rank (ARHR) as indicators, achieving an HR of 0.76 and an ARHR of 0.38. The robustness of our model is demonstrated through comparisons with other studies.In Hybrid Movie Recommendation System with Content-Based and Memory-Based Collaborative Filtering Using Similarity Graph section, we propose a movie recommendation system based on the knowledge graph and a mixture of content-based and collaborative filtering, which will wander the path of the knowledge graph and calculate the recommendation index of movies to users in each path. We get the recommended list of the top N movies by sorting and calculating the Hit Rate and Average Reciprocal Hit Rank. The results show that the proposed recommendation model has good recommendation efficiency, where the best Hit rate and Average Reciprocal Hit Rank are 0.74 and 0.44