Article Preview
Top1. Introduction
To make correct decisions, people use websites for getting large amounts of information in form of opinions, feedback, reviews, ratings, and comments about an event, product, individuals, or service. Usually, users provide their views and comments regarding the items over the web. So the decision could be made by considering the ratings and useful information on the web provided by the users of the platform. This way of making decisions by employing the already available experience and feedback of the users over the web is analogous to real-life scenarios where humans make purchases based on the suggestions/recommendations provided by their kith and kin who have already consumed the product or have an idea about it. The same methodology when fed into a computerized system is termed a recommender system, The recommender systems deal with the problem of information overload by filtering relevant information from user’s interests or preferences. A recommender system is the information filtering system that has much ability to predict and recommend the relevant items to a particular user with the help of a user's profile (Behera et al.,2017; Folajimi et al.,2015; Barik et al.,2019). The requirements of a robust recommender system can be enhanced with the help of e-commerce sites and online businesses (Folajimi et al., 2015; Pradhan et al.,2020). Recommender systems are more crucial for better prediction and recommendation for both the users and service providers. These systems help to minimize the transaction costs of identifying better items in online e-commerce. The quality of decision-making can be improved with the help of a recommender system. In today’s e-commerce market, business holders generate more revenues by using a recommender system. Therefore, we require an efficient and accurate technology within the recommender engine so that it can provide better prediction and recommendation for the users with achieving higher privacy(Bobadilla et al.,2009; Barik et al.,2019).
According to the survey report of 1999, maximum numbers of people want to hide their information because of privacy concerns (Dinev et al., 2011). The main challenge is to utilize the users’ private information for collaborative filtering (CF) purposes. So, various privacy-preserving techniques are introduced in the domain of CF systems (Du et al.,2009). In the CF-based approach, preserving privacy and attaining accuracy are both statistically independent. Privacy is the main problem in CF. To solve this issue, two different approaches are used i.e. encryption method and the randomization method (Li et al., 2015). In this paper, different privacy-preserving-based machine learning methods are explained along with their advantages and disadvantages. Clustering is the one main part of machine learning. To provide better product recommendations, clustering plays the main role in the domain of recommender systems (Jain et al.,2020; Jain & Kumar,2020). Differential privacy protecting k-Means clustering algorithm is used which uses the contour coefficients to evaluate the clustering effect of each iteration and add different noise to different clusters (Liu et al., 2018). In this paper, the clustering technique is used in the entertainment-based recommender system where a different number of user clusters is created based on the item- similarity of the users. The basic k-Means along with the mode selection clustering algorithms are used to find the user clusters. To get better accuracy for creating user clusters and achieving a high level of privacy, a random perturbation-based fuzzified k-Mode clustering method is used in the recommender system.