Research Article
Leveraging Hybrid Ensembling for Robust Movie Recommendations: A Comprehensive Approach
@INPROCEEDINGS{10.4108/eai.28-4-2025.2358038, author={Rithvika Sunkara and Gangalakshmi Jiyyani and Ajit Mahapatro and Joshna Priya Cheruku and Dasari Bhargav Krishna and Nollu Hari Satish Kumar}, title={Leveraging Hybrid Ensembling for Robust Movie Recommendations: A Comprehensive Approach}, proceedings={Proceedings of the 4th International Conference on Information Technology, Civil Innovation, Science, and Management, ICITSM 2025, 28-29 April 2025, Tiruchengode, Tamil Nadu, India, Part II}, publisher={EAI}, proceedings_a={ICITSM PART II}, year={2025}, month={10}, keywords={movie recommendation system hybrid approach neural collaborative filtering (ncf) deep learning content-based filtering}, doi={10.4108/eai.28-4-2025.2358038} }- Rithvika Sunkara
Gangalakshmi Jiyyani
Ajit Mahapatro
Joshna Priya Cheruku
Dasari Bhargav Krishna
Nollu Hari Satish Kumar
Year: 2025
Leveraging Hybrid Ensembling for Robust Movie Recommendations: A Comprehensive Approach
ICITSM PART II
EAI
DOI: 10.4108/eai.28-4-2025.2358038
Abstract
In this paper, an improved hybrid movie recommender system based on content-based filtering (CBF), collaborative filtering and deep learning is presented. The approach has three sections; collection of data, pre-processing of data, and feature engineering performing operations such as developing popularity metric, TF-IDF to get the keywords out and encoding the genres as multi-hot vectors respectively. The database is separated into the training and testing database for a fair evaluation. Key components include Neural Collaborative Filtering with embeddings, TF-IDF with cosine similarity, and transformer models (BERT) for semantic text analysis. The ensemble framework is made of these models, which are optimized with Gradient Boosting. Performance comparison through Precision@K, Recall@K, RMSE and diversity confirms the superiority of the proposed method in terms of accuracy and user satisfaction over the single-model methods.
