Solving Cold Start Problem for Recommendation System Using Content-Based Filtering

A recommendation system is an effective supporting tool for a user to obtain the relevant result and avoid information overloading. However, the user cold-start problem has occurred when the system does not have adequate information about the user. In this project, we propose a content-based filtering method that applies the movie genres for finding the similarities among movies. We use the Term Frequency-Inverse Document Frequency (TF-IDF) as a text mining technique for retrieving the frequent movie genres and produce the TF-IDF into a matrix. Then, we use the matrix to generate recommendation models that apply with different similarity techniques which are Cosine similarity, Pearson correlation coefficient, and Euclidean distance. As a result, the average of similarities has reached the highest of 84 percent using Cosine similarity and at most 20 percent of a not satisfied movie recommended among 30 recommendation items. The experimental results on MovieLens-20M data that apply to our approach have achieved high similarity results among movies recommended, low frequency of not-satisfied movie ratings, and it consistently improved the performance in addressing the user cold-start problem.