Database Reference
In-Depth Information
friends. It incorporates impacts from distant friends via an iterative classification.
We evaluated the performance of SNRS with several other methods on the Yelp
dataset through a tenfold cross-validation, and SNRS achieves the best result. In
terms of prediction accuracy, it yields a 14.3% improvement compared with that of
CF, while in terms of coverage, it yields a 31% improvement compared with CF. In
the sparsity test, SNRS returns consistently accurate predictions and high coverage
over a wide range of data sparsity. Even in a cold-start test, SNRS still performs
reasonably well. We also studied the role of distant friends in SNRS and found that
when the influences from distant friends are considered, the coverage of SNRS can
be significantly improved with only a slight reduction in the prediction accuracy.
To deal with heterogeneities in social networks, we further proposed an approach
for filtering social networks based on the semantics in fine-grained user ratings and
ratings of friends. Using this approach, relevant friends can be selected for inference
according to the type of target items. A specific class experiment was designed to
evaluate the effectiveness of semantic filtering in the social network that was formed
by a large group of graduate students. The experimental results reveal that SNRS
with semantic filtering can further improve the prediction accuracy by 11.6%.
Finally, we investigated two trust issues in SNRS. We showed that SNRS has the
capability of handling shilling attacks, as well as the problems caused by friends
with unreliable knowledge. Further research in these areas is desirable.
In our future work, we propose to study the performance of SNRS in other
datasets, such as categories other than restaurants on Yelp. We also want to investi-
gate how to apply SNRS to other Web 2.0 domains such as Facebook. For example,
Facebook recently started personalizing user contents such as news feeds. Intui-
tively, our framework may also be applicable to the recommendations of news feeds,
since the recommendation has to consider users' own preferences, the global
popularity of news itself (i.e., item likability), and users' social networks.
References
1. Mooney, R.J., Roy, L.: Content-based topic recommending using learning for text categori-
zation. In: Proceeding of ACM SIGIR'99 Workshop Recommender Systems: Algorithms and
Evaluation (1999)
2. Sarwar, B.M., Karypis, G., Konstan, J., Riedl, J.: Item-based collaborative filtering recom-
mendation algorithms. In: Proceedings of 10th International Conference on World Wide Web
(WWW'01) (2001)
3. Breese, J.S., Heckerman, D., Kadie, C.: Empirical analysis of predictive algorithms for
collaborative filtering. In: Proceedings of the 14th Conference on Uncertainty in Artificial
Intelligence, pp. 43-52 (1998)
4. Billsus, D., Pazzani, M.: Learning collaborative information filters. In: Proceedings of Inter-
national Conference on Machine Learning (1998)
5. Resnick, P., Iakovou, N., Sushak, M., Bergstrom, P., Riedl, J.: GroupLens: An open architec-
ture for collaborative filtering of Netnews. In: Proceedings of Computer Supported Coopera-
tive Work Conference (1994)
Search WWH ::
Custom Search