ADC@θr: Adaptive divisional categorization of ratings under rating threshold θr for similarity computation in recommendation systems

• Published in: Applied intelligence (Dordrecht, Netherlands) Netherlands), 2022, Vol.52 (2), p.2134-2153
• Main Authors: Jain, Ankush, Nagar, Surendra, Singh, Pramod Kumar, Dhar, Joydip
• Format: Article
• Language: English
• Subjects: Artificial Intelligence; Computer Science; Machines; Manufacturing; Mechanical Engineering; Processes
• ISSN: 0924-669X; 1573-7497
• DOI: 10.1007/s10489-021-02428-w

In the recommendation systems (RSs), it is imperative to incorporate the hidden contextual meaning of users’ provided ratings in the similarity computation. To draw such contextual meanings, existing models use the fixed categorization of accessible ratings. However, due to the excessive variation in similarly co-rated item pairs, they produce ambiguous contextual meanings that yield inconsistent results for the user pairs similarities. Therefore, to deal with this problem, this paper proposes an adaptive divisional categorization (ADC)-based RS, namely ADC@ θ r , that obtains the optimal contextual divisions of accessible ratings under the rating threshold θ r . Here, accessible ratings are the numerical scores that RS users use to express their preferences on the underlying items. A set of adaptive divisions under rating threshold θ r is termed optimal if most of its rating divisions cover a large portion of rating records of a given dataset. For so, the proposed ADC@ θ r model keeps only those divisions of ratings whose significance values are high, i.e., cover a large portion of the rating records by the ratings of these divisions. Further, the contextual mean square deviation (CMSD) model is employed to compute user pairs’ similarity using the obtained adaptive divisions of accessible ratings. The experimental results obtained on the benchmark Movielens-100K and Movielens-1M datasets justify the proposed model’s superiority over the competitive models.