An Enhanced Item Recommendation Approach Using the Sigmoid Function and Jaccard Similarity Coefficient

Abstract

Based on prior interactions between users and service items, recommendation systems have developed into practical tools for filtering and obtaining vital data. These systems are often used in a range of commercial industries, including e-commerce, tourism, social networking, and academic research. Collaborative filtering, content-based filtering, and hybrid recommender systems are the three main categories of recommender systems. Collaborative filtering recommender systems, which presume that users would be interested in products that users similar to them have highly rated, take into consideration users' tastes (in terms of item preferences). Content-based filtering recommender systems base their recommendations on the textual content of a product, using the assumption that customers would prefer items that are similar to those they have previously enjoyed. In a hybrid recommender system, two techniques are combined. These systems struggle with scalability, data sparsity, and cold starts, which leads to low-accuracy prediction and coverage. In this study, we proposed a unique recommendation method and applied the sigmoid function to the Jaccard similarity index. In our proposed method, which included the rating significance of items, we used the sigmoid function on the Jaccard similarity index to evaluate the asymmetry relationship between users. The similarity between the target user and his or her neighbours is then assessed using the asymmetric association and rating significance. Several cuttingedge similarity metrics are evaluated using experiments on three real-world data sets. The results show that the new similarity model performs better than the baseline models in terms of diversity and prediction accuracy.