Abstract
The goal of this project is to Implement a Personalized Movie Recommendation system using collaborative filtering/probabilistic relevance feedback techniques and an r nearest neighbor movie classification system, decision tree, random forest, n-ary SVM, LSH using database from IMDB-MovieLens with PCA, SVD, LDA and CPD models.
Phase 1
Implemented a program which considers,
- all the movies an actor played,
- a given genre,
- all movies watched by a user,
and creates and stores a weighted tag vector for each using TF, TF-IDF, TF-IDF-DIFF, P-DIFF1, P-DIFF2 models.
Phase 2
- Generating and identifying top latent semantics using PCA, SVD and LDA.
- CP decomposition of Tensors to identify non-overlapping groups based on their degree of memberships to the top 5 latent semantics.
- Implemented Personalized PageRank algorithm with RandomWalk and ReStarts to identify related movies/actors.
- Based on these techniques, built a minimal version of movie recommender that with given watched movies can identify related movies to recommend.
Pahse 3
Implemented a Personalized Movie Recommendation system using collaborative filtering/probabilistic relevance feedback techniques and an r nearest neighbor movie classification system, decision tree, random forest, n-ary SVM, LSH using database from IMDB-MovieLens with PCA, SVD, LDA and CPD models.
