A Modified Similarity Measure for Improving Accuracy of User-Based Collaborative Filtering
Production sites suffer from idle in marketing of their products because of the lack in the efficient systems that analyze and track the evaluation of customers to products; therefore some products remain untargeted despite their good quality. This research aims to build a modest model intended to take two aspects into considerations. The first aspect is diagnosing dependable users on the site depending on the number of products evaluated and the user's positive impact on rating. The second aspect is diagnosing products with low weights (unknown) to be generated and recommended to users depending on logarithm equation and the number of co-rated users. Collaborative filtering is one of the most knowledge discovery techniques used positively in recommendation system. Similarity measures are the core operations in collaborative filtering; however, there is a certain deviance through using traditional similarity measures, which decreases the recommendation accuracy. Thus, the proposed model consists of a combination of measures: constraint Pearson correlation, jaccard distance measure and inverse user frequency (IUF).
The experimental results implemented on movielens data set using MATLAB show a comparison between the results of the proposed model and some of the traditional similarity measures. The outcome results of the comparison show that the proposed model can be used as a parameter in the prediction process to achieve accurate prediction results during recommendation process.