Enhancing Diversity in Recommendation Systems using Likelihood-based Item Recommendation
Chetan J. Awati, Suresh Shirgave, Sandeep A. Thorat · 2022 International Conference on Augmented Intelligence and Sustainable Systems (ICAISS) · 2022
The popularity of items from the existing rating dataset has a significant impact on Recommendation Systems (RS). There is also a need to minimize bias of the recommendation model by addressing the diversity solution in which less popular items should also be included in the ranked items list during recommendation. The diversity aspect-based improvement of the RS filter may change the user's distribution ratio compared to traditional RS. This paper presents the probabilistic approach for recommending the users' interests during selective ranked items selection. In the proposed method list of 10 items is considered sequentially on a popularity basis. The first seven items are chosen for recommendation in descending popularity order. The popularity and likelihood of the item are estimated using Hidden Markov Model (HMM). The less likelihood list is also generated in which ascending set of items are selected as the remaining three items in the recommendation list. The missing filling entry, user interest pattern modelling, and HMM model-based likelihood estimation are the steps in the proposed approach. The comparative analysis using a histogram of user distribution is performed for EDUA, CML, and DPP parameters, which shows a better number of users for less likelihood items compared to state-of-the-art approaches.