A Novel Approach for Long Term Dynamic Recommendation System With Deep Reinforcement Learning Techniques

Abstract

Recommendation Systems offer to help consumers determine their interests by predicting their ratings and preferences for particular products. The Reinforcement Learning agent's capacity to learn from the environment as well as reward without training data makes it a perfect approach for such systems. Traditional works have examined Deep Reinforcement Learning (DRL) as a recommendation system because of its capacity. Existing studies experienced challenges such as scalability, overlapping values, information loss, and inappropriate model training, resulting in inaccurate proposals. As a result, the purpose of this study is to determine and tackle these existing issues. The proposed work shows a DRR (DRLbased Recommendation) system based on actor-critic learning. In an actor system, DWL-FA (Deep Weighted Likelihood-Factor Analysis) has been suggested to adapt an existing DNN (Deep Neural Network) to adjust for environmental shifts by removing undesirable regions from network outcomes. The attention mechanism provides the decoder with relevant information from the encoder's hidden states. This attention mechanism, together with the DWL-FA model, may focus on useful sequences and learn their connections. This helps the trained model learn better. In critical networks, HMP-WU (Hidden Markov ProbabilityWeight Updation) has been suggested to optimize user interactions with recommended items and the recommender system (agent). Weight Updation improves awareness of related sequences and reduces inaccurate predictions. The proposed techniques enhanced the system's outcomes by 5.74% in terms of the average p-value...