Synergizing Collaborative Filtering and Popularity Scores in a Machine Learning Approach for Content Suggestions

Today, it is common to rely on reviews from other people when making decisions on a variety of life decisions, including which movies to watch, which items to buy, and which books to read. Recommender systems are crucial in meeting this need. By projecting a user's expected rating for an item, these systems function as filtering algorithms. Recommender systems provide advice to consumers via a prioritized list of projected item ratings after filtering through large data repositories. Online book recommender systems focus primarily on book aficionados. People frequently take into account and put their faith in the comments and reviews written by previous readers when choosing which book to read. The following study proposes a cleverly combined hybrid recommender system that combines content-based and collaborative filtering techniques. The author uses clustering techniques to divide data points into various clusters in the context of collaborative techniques. For this, methods like K-means clustering and Gaussian mixture models are employed, with the best algorithm being selected through the use of the silhouette score. The clustering procedure is then guided by the selected algorithm. Additionally, to efficiently reduce the dimensions of a sparse dataset, the research applies the Matrix Factorization method, specifically using Truncated Singular Value Decomposition (Truncated-SVD). The system uses the TF-IDF vectorizer for Content-Based Filtering, which takes textual assertions as input and converts them into a matrix of related vectors. The Root Mean Square Error (RMSE), a metric that measures the difference between absolute and obtained values, is used in the study to assess the system's correctness. This measure sheds light on the recommender system's underlying accuracy.