Vector-Based Similarity for Personalized Recommendation Systems

Personalized Recommendation Systems (RS systems) have been gaining popularity among researchers in data science as the systems help users search for products that satisfy their preferences. Likewise, the systems also provide benefits to companies who desire their products to be known to more future customers. Although a wide ranges of methods and techniques have been proposed to boost RS systems performances recently, the significant contributions of this kind of systems are always needed in this information age when more e-commerce platforms are available. In this article, two groups of methods including (1) basic prediction methods and (2) vector-based similarity methods combined with KNN are compared in this study. In particular, the basic prediction method employed Min, Max and Average of the user's ratings to predict the missing values. On the other hand, the vector-based methods employed the combination of KNN and vector-based similarity including Pearson and Cosine to predict the missing ratings. The experiments were carried out using the datasets MovieLens 100 K. The evaluation metrics includes Mean Absolute Error (MAE), Cohen's Kappa metric (KAPPA), and Root Mean Square Error (RMSE) to compared the performances of the models. The vector-based similarity using Pearson similarity combines with KNN provided better performance than the other models for the metrics including RMSE and MAE.