Adaptive Reinforcement Learning with Improved Artificial Bee Colony Optimization for Personalized and Enriched News Recommendation

Online news contents is increasing exponentially and news are collected from various sources. Personalized news recommendation system has been developed for supporting individual users and this approach will increase the user engagement and satisfaction. Traditional news recommendation system suffers with volatility of user preferences and feedback comment given news feeds. This paper proposes an adaptive reinforcement learning framework by designing with improved artificial bee colony optimization technique. This recommendation framework will enhance personalized news recommendations. The proposed news recommendation system uses reinforcement learning technique for creating an interactive user mode and adaptive recommendations based on continuous learning model. The traditional news recommendation system rely on static user item relationships, ARL optimization to participate in a long time through exploration and exploitation strategies. The efficiency and accuracy in a learning environment has been improved by applying IABC technique. The improved artificial bee colony optimization technique enhance learning rate, guided searching strategies, and efficient exploration. These improvements will enhance the results through fast convergence speed and solution quality. The news recommendations are performed concerning the personalized wish list and most common attributes in the research of news content. The Agent node will create a suggestion list with news feeds based on personally collected information from individuals. Based on the user news reading strategy, the environment will perform the rewarding mechanism. The proposed Agent node is designed using the IABCO algorithm, which makes a suggestion list with enriched news content by using adaptive threshold value selection-based probability of success. The performance evaluation has been conducted with following parameters, like precision, recall, F1 Score, Click Through Rate (CTR), Average Click Position, Diversity, and Coverage. A comparative analysis is carried out with existing news recommendation systems and this result shows that the proposed news recommendation system achieves 93.6 % precision, 92. % recall, and 92.9% F1 score values. This result shows that the superiority of proposed news recommendation approach and this would be able to provide highest accuracy value, which is nearly 97.5%.