A Novel Trust Computation Method Based on User Ratings to Improve the Recommendation

Abstract

Today, the trust has turned into one of the most beneficial solutions to improve recommender systems, especially in the collaborative filtering methods. However, trust statements suffer from a number of shortcomings, including the trust statements sparsity, users' inability to express explicit trust for other users in most of the existing applications. To overcome these problems, this work presents a method for computing implicit trust based on user ratings, in which four influential factors including Similarity, Confidence, Analogous Opinion, and Distance are utilized to achieve trust. For computing users' similarity, Person Correlation Coefficient measure was applied. Confidence was computed through users' common in rated items. To compute users' analogous opinions, their ratings were evaluated from three aspects of their satisfaction, dissatisfaction, and indifference about the items. Euclidean distance was employed on users ratings for computing the distance. Finally, the factors were combined to reach the implicit trust. Moreover, fuzzy c-means clustering was applied to initially partition similar users for enhancing the performance positively. Finally, two MovieLens datasets of 100K and 1M have been used to evaluate this approach, and results have shown that the approach significantly increases Accuracy, Precision and Recall, compared to some other existing methods.

A Compositional Adaptation-based Approach for Recommending Learning Resources in Software Development

“…These systems are able to intensify teaching and practicing based on a variety of reasoning methods [8]. There also exist some hybrid recommender systems, which are adaptive to learner's preferences and are able to generate recommendations by some hybrid approaches such as content-based filtering, collaborative selection and opinion mining as well [9][10][11]. Another kind of hybrid recommender system also exists, which makes use of ontology and sequential pattern mining (SPM) to evaluate the domain knowledge of the learner and learning resources, and determine the learners' consecutive learning patterns [12].…”

An Implicit Trust-Network construction approach and a Recommendation Methodology for recommender systems

“…Regarding the comparison and results presented in [5], each proposed criterion of similarity, confidence, and identical opinion decreases the error of predictions and enhances the coverage. The mentioned criteria have been compared with similar traditional metrics including PCC, Jaccard,Confidence Metric, and Analogous Opinion [6].…”

“…Additionally, the steps of constructing the implicit trustnetwork (ITN) can be reviewed in detail in [5]. The ITN is evaluated by measures of MAE, RMSE, coverage, and sensitivity to prove its efficiency compared to the state-of-art methods-such as CPD [7], BA-TRS [8], TrustANLF [9], ITCM [6], TrustSVD [10], SocialMF [11], and many others. Regarding the results presented in [5], ITN results in significantly improving the prediction error by decreasing around 13%-55% for MAE and around 34%-90% for RMSE.…”

Multi-criteria– Recommendations using Autoencoder and Deep Neural Networks with Weight Optimization using Firefly Algorithm

“…RS models can be built using content-based filtering (CBF) [2], collaborative filtering (CF) [3][4][5] and Hybrid (combination) strategies [6]. CF is further divided into user-based CF [3], item-based CF [4] and trust aware CF [7]. The pitfall of traditional RS are data sparsity and cold start issues [8].…”