High-quality user experience improvements across e-commerce or content platforms are inherent in recommender systems. Traditional recommendation approaches, such as collaborative filtering (CF) and content-based filtering (CBF), are typically grounded only in user-item interaction data and often neglect the valuable semantic and sentimental knowledge embedded in user review data. Unfortunately, that limits how personalised and engaging it can be. In recent years, sentiment-aware recommendation has seen significant development. However, these models essentially combine sentiment information in shallow, static representations that fail to support adaptive personalisation and capture deep semantics. To overcome these limitations, in this paper, we propose a deep learning based sentiment-aware product recommendation system, namely DeepSentRec, that tightly integrates sentiment analysis, hybrid filtering and reinforcement learning. SentimentBERT is employed to obtain fine-grained sentiments from review texts, and HybridCF-SBERT combines collaborative and semantic content-based similarities. RLRanker-PPO, a reinforcement learning-based component, adaptively ranks recommended items based on user engagement behaviours. Such a three-step architecture enables strong, dynamic, and personalised product recommendations. The proposed system has been evaluated on four publicly available datasets: Amazon Reviews, Yelp Dataset, IMDB Sentiment Dataset, and Kaggle E-Commerce Reviews. Our method achieves 91.12% precision, 85.49% recall, 89.66% NDCG, and 24.96% CTR, significantly outperforming SOTA baselines. Experimental results demonstrate that DeepSentRec improves recommendation accuracy and increases user satisfaction and activity. Highly Modular & Adaptive in Nature — Diverse & high potential to accelerate and benefit from end-use across diverse domains, essentially a way of being a next-generation intelligent recommendation framework.
Kollu et al. (Mon,) studied this question.