Trust-based local and social recommendation

Abstract

In this article, we propose an evolution of trust-based recommender systems that only relies on local information and can be deployed on top of existing social networks. Our approach takes into account friends' similarity and confidence on ratings, but limits data exchange to direct friends, in order to prevent ratings from being globally known. Therefore, calculations are limited to locally processed algorithms, privacy concerns can be taken into account and algorithms are suitable for decentralized or peer-to-peer architectures.We have implemented and evaluated our approach against five others, using the Epinions trust network. We show that local information with good default scoring strategies are sufficient to cover more users than classical collaborative filtering and trust-based recommender systems. Regarding accuracy, our approach performs better than most others, specially for cold start users, despite using less information.

Comparing the Predictive Capability of Social and Interest Affinity for Recommendations

“…On top of it, we implement the main building blocks of SR and CF as used for comparison in literature [3,20,21,35,25,23]. Specifically, we implement (a) item-and userbased CF variants as often used as reference point by previous work [20,35,38,25], and (b) a SR approach that aggregates the ratings similarly with CF, yet, instead of deriving users affinity based on how similar they rated items in the past, it does so based on their social ties. Next we describe each approach and motivate our choices.…”

“…The redline splitting the boxplot is the median, while the star is the average performance (also plotted above each boxplot). [20,22,35,39], and the averaged performance at user (resp. item) level.…”

“…In contrast, SR systems leverage users social ties to make predictions [46,33,37,16,35], assuming that these reflect common tastes or interests. SR systems deal better with cold start users, as they require users only to be connected to other users in the social network, and do not have to wait for users to grow a rating history to make predictions.…”

“…Alas, while these systems tend to achieve better coverage, they can also suffer due to sparse ratings and sparse trust relations. Thus, in order to consider the ratings of users that are not directly connected, various approaches propagate the trust among their users [46,16,33,35]. Yet, in these cases the recommender might end up considering ratings of weakly trusted users, thus affecting the precision [20].…”

“…A common practice in evaluating such approaches is to resort to (i) one [42,45,35,22,20,25,30], sometimes two [39,31] datasets and (ii) global averages for the metrics of choice. Alas, this has made it difficult to draw generalizable conclusions on the effectiveness of leveraging the social ties for recommendations compared with CF across datasets of different nature.…”