A Framework for Recommender Systems Using Improved Collaborative Filtering

In the last few years, most of e-commerce, social media networks, and digital libraries websites require a recommender systems in their websites. These systems have the ability to suggest items to users that may meet their interests. Recommender systems rely on a collaborative filtering techniques (CF) are the most widely used and successful technique in this field. However, most of the recommender systems suffer from the missing value problem that leads to inaccurate prediction and leads to poor recommendation quality. To overcome this issue, this paper introduces an improved collaborative filtering framework. The proposed framework averages the distance between the active user and the selected reliable users based on the shared ratings given to some items in common. Then, makes a list of users who shared a specific number of items in common with the active user. After that, it predicts the unknown ratings based on novel mathematical formulas. Furthermore, it creates a list of nearest neighbors to the active user and then recommends a list of items. In this paper, the dataset proposed by literature was used to test the proposed framework. The experimental results show that the proposed work improved CF technique and outperform other literature in case of which user should be selected and how much is accurate the predicted rating. In this context, we adopted the Mean Absolute Error (MEA) metric that widely used to measure the accuracy of the improved CF.

Перевод пока недоступен