خلاصة:
These days, due to growing the e-commerce sites, access to information about items is easier than past. But because of huge amount of information, we need new filtering techniques to find interested information faster and more accurate. Therefore Recommender Systems (RS) introduced for solving this problem. Although several recommender approaches have proposed, Collaborative Filtering (CF) approaches are the most successful ones. These approaches use historical behaviors of users for making recommendation. Next generation of CF, called Trust-based CF, use social relations and activities for measuring trust between users. One important step in these approaches is measuring the similarity between users, which affect recommendation results. Therefore variety methods for this reason have been proposed. In this paper, we will review and categorize the measurement methods. We will also analyze the methods to identify their characteristics, benefits and drawbacks.
ملخص الجهاز:
A Review on Similarity Measurement Methods in Trust-based Recommender Systems Morteza Ghorbani Moghaddam University Putra Malaysia, Malaysia morteza.
One important step in these approaches is measuring the similarity between users, which affect recommendation results.
Keywords: measurement methods, Trust-based approaches, recommender systems, Collaborative Filtering, E-commerce.
They served PCC method for measuring similarity between users' rates and used average weighted of previous rates for making prediction.
2013), used PCC measurement method for creating extended items based on trust relations between users.
PCC just uses the value rates, but it seems that number of common rates between two users is also important for measuring their similarity.
If number of common rates between two users is less than H, the measured similarity based on PCC decrease.
TrustWalker (Jamali & Ester, 2009) is a random walk model which has combined trust- based and item-based collaborative filtering approaches to improve accuracy of predictions and solve cold-start and sparsity problems.
, 2013) is a trust-based approach that uses extended items for measuring the similarity between users.
This new method, called Confidence-aware Pearson Correlation Coefficient (CoPCC), has been formulated as below: / (5) The difference between PCC and CoPCC is confidence factor that is cu p Related to rates of user u about item p.
In Merge, as discussed before, for measuring the importance weight of predicted rates in extended items, combination of trust, rating similarity and social similarity has been used.
/ (12) All of discussed methods used local rating information of user about items for making similarity.