Film2Vec – A Feature-based Film Distributed Representation for Rating Prediction

The st UTS-VNU Research School Advanced Technologies for IoT Applications Title: Film2Vec – A Feature-based Film Distributed Representation for Rating Prediction Author Names and Affiliations: Xanh Ho and Nhung T.H Nguyen – VNUHCM University of Science Abstract: Approaches for film recommendation systems usually exploit explicit descriptive features to compute ratings In this paper, we suggest a different approach – to rate films via their related neighbors computed via distributed representation of movies Specifically, we present Film2Vec, a distributed representation learning for films adapted from the distributed hypothesis from linguistics We implement our proposed idea using TensorFlow, a Google’s Deep Neural Networks software The experimental results on Movielens dataset show that Film2Vec can effectively reduce root mean square error (RMSE) in movie recommendation task, suggesting yet another beneficial application of deep learning Problem Statement Recommendation systems Recommend Many works use rating information Contributions Film2Vec – Representing Films as Vectors Few works use context of recommendation system Context of film: • Title • Actors - A • Tags - T • Genres - G • Directors - D Results 1.1 Pre-processing HetRec 2011 Film1 Film2 Film3 … Filmn A1 D120 G19 T18 A13 D14 G156 T17 A12 D23 G43 T65 A45 D2 G4 Film descriptions T1 Root Mean Square Error 1.05 0.95 0.9 0.85 0.8 0.75 0.7 F2V-TA F2V-TDGA Film vectors Film2Vec Worst F2V-TA CF CA Best F2V-TDGA ARR LLS IMBRF Previous approaches Future work References • Use other information of film such as countries, location and plot • Apply to other areas such as books, services, and papers [1] Baroni et al “Don’t count, predict! A systematic comparison of context-counting vs context-predicting semantic vectors”, ACL, 2014 [2] Bothos et al “Information market based recommender systems fusion”, HetRec, 2011 [3] Mikolov et al “Efficient Estimation of Word Representations in Vector Space”, ICLR, 2013