A New Similarity Measure Based on Adjusted Euclidean Distance for Memory-based Collaborative Filtering

Abstract

Memory-based collaborative filtering (CF) is applied to help users to find their favorite items in recommender systems. Up to now, this approach has been proven successful in recommender systems, such as e-commerce systems. The idea of this approach is that the interest of a particular user will be more consistent with those who share similar preference with him or her. Therefore, it is critical that an appropriate similarity measure should be selected for making recommendations. This paper proposes a new similarity measure named adjusted Euclidean distance (AED) method which unifies all Euclidean distances between vectors in different dimensional vector spaces. Our AED enjoy the advantages that it takes both the length of vectors and different dimension-numbers of vector spaces into consideration. Based on two datasets MovieLens and Book-Crossing, we conduct experiments comparing our AED with two notable existing methods. The experimental results demonstrate that our AED improves the accuracy of prediction and recommendation

An Improved Collaborative Filtering Algorithm Based on User Interest

“…Nathan and et al (2010) increased time factor in user similarity calculating process and achieved relatively good recommendation results [10]. Similar works have been down by PaPagelis M.et al (2005), Tieli Sun,et al (2009) and Sun H,et al(2011), who utilized item similarity to improve user similarity calculating process and achieved better results [11][12][13]. The other stream of research is by analyzing the characteristics of users and items comprehensively to enhance the performance of CF algorithm.…”

Data imputation for gas flow data in steel industry based on non-equal-length granules correlation coefficient

“…Up to now, there are many methods can be used to quantify the similarity or correlation of two sequences. An adjusted Euclidean distance based method was proposed in [35] , and a Mahalanobis distance measure termed as locally centred Mahalanobis distance was introduced in [38] , which can be used as a similarity measure. However, the distance measure cannot accurately express the correlation.…”

Collaborative Filtering Recommendation Model Based on Normalization Method

“…For an item which has not been scored by target user and needed to be predicted, evaluation value could be got by applying a calculation on neighbors' evaluation of thisitem [15]. In order to improve prediction accuracy, and considering different users' different rating scale, user's average score on all items is used in prediction.…”