Improving Recommendation Accuracy by Combining Trust Communities and Collaborative Filtering

Correlation clustering by contraction

“…This means, that upon meeting a key (which is a coordinate-pair), we need to check whether one of it is equal to j (to the key in the associative array s) or not. If it is, then we need to replace it with i (with the value according the key) [lines [6][7][8][9]. In the diagonal of D, only zero values are allowed, but we do not store them, so if something gets into this diagonal it is ignored [lines [10][11].…”

“…This kind of clustering has many applications: image segmentation [3], identification of biologically relevant groups of genes [4], examination of social coalitions [5], improvement of recommendation systems [6] reduction of energy consumption [7], modelling physical processes [8], (soft) classification [9], [10], etc. At correlation clustering, those cases where two dissimilar objects are in the same cluster, or two similar objects are in different clusters are treated as a conflict.…”

A novel collective matrix factorization model for recommendation with fine‐grained social trust prediction

“…It has been accepted that negative relations are also important as users are allowed to express both positive and negative opinions on others in real‐world websites like Epinions . Ma et al assumes that trust can be explained as “similarity” while distrust can be interpreted as “dissimilarity.” Ma et al employs trust matrix which contains both positive and negative connections as input of matrix factorization to generate user communities based on Extend Epinions dataset. Users in the same group trust each other while communities with negative connections are against each other.…”

“…TaRS can avoid such kind of attacks because users can add these people to their blacklists and explicitly express distrust on them . Guha et al have pointed out that it is unwise to ignore these kind of distrust relations, and now, it is widely accepted that distrust relations also play a significant role in the performance of recommender systems …”

“…1,10 Guha et al 11,12 have pointed out that it is unwise to ignore these kind of distrust relations, and now, it is widely accepted that distrust relations also play a significant role in the performance of recommender systems. [13][14][15] Epinions* is a website where users can express both positive and negative relations with each other, moreover users can express a five-scale rating on reviews that are written by other users. However, such kind of relations are currently defined in a binary fashion.…”