A SYNERGISTIC APPROACH TO PRECISION RATING PREDICTION IN RECOMMENDER SYSTEMS USING DEEP LEARNING AND COLLABORATIVE FILTERING
Umer Farooq, Shahid Ameer, Samreen Razzaq, Suleman Shahzad, Syed Sami Ahmad Bukhari, Anam Safdar Awan, Zainab Fatima, Amna Noor · 2025
In today’s digital world, users mainly rely on online platforms for various activities like shopping, social networking and streaming media. Therefore, recommender systems play crucial role in modern digital platforms as it guides the users by suggesting them content and items relevant to their interest. In this research study, we aim to develop an effective recommender system that accurately predicts unknown item ratings and make recommendations according to user’s interest more precisely by applying collaborating filtering (cf) based techniques on user-item interaction. We applied various cf based techniques including user k-nearest neighbors and item k-nearest neighbors utilizing cosine similarity and pearson correlation and matrix factorization-based methods which are alternating least squares and singular value decomposition. Machine learning methods such as slope-one, baseline and co-clustering as well as deep learning models like neural collaborative filtering, recurrent neural networks and long short-term memory (lstm) networks. For the performance evaluation of these methods and models the metrics used are mean absolute error (mae) and root mean square error (rmse) on two datasets which are book-crossing and recipe reviews. The results indicate that lstm consistently outperforms rest of the applied algorithms by achieving the lowest rmse and mae values 1.06 and 0.31 on book-crossing and 0.44 and 0.37 on recipe-reviews dataset. The empirical analysis shows capability of lstm model to capture complex patterns and long-term dependencies in users’ data. The findings highlight the effectiveness of deep learning models especially lstm in making accurate personalized recommendations greatly which can improve users’ satisfaction and engagement.