MOVIE RECOMMENDATION SYSTEM USING WORD EMBEDDING

Abstract

As more movies are added online, it is becoming harder to decide what to watch. This project is designed to give movie recommendations, considering not only the keywords users type but also what those words truly mean. It employs Word2Vec to go over the metadata of films, including the names, genres and keywords and find interesting connections between them. It was found that Word2Vec generated far better and more appropriate suggestions when evaluated against using CountVectorizer. The Skip-Gram model in Word2Vec was able to spot similar meaning between movies even when they didn’t both contain the same words. The system learned from the TMDB dataset and its results were measured using Precision@10, Mean Reciprocal Rank (MRR) and NDCG. It was found that Word2Vec greatly improved the recommendations over strategies that depend on word frequency. In general, the study finds that adding context-aware ideas makes movie recommendations more personal and meaningful to viewers. Improvements for the future could involve mixing this technique with collaborative filtering to improve both the accuracy and user enjoyment.