Personalized Resource Recommendations using Learning from Positive and Unlabeled Examples

Priyank Thakkar, Dr. K. Kotecha

Abstract


This paper proposes a novel approach for recommending social resources using learning from positive and unlabeled examples. Bookmarks submitted on social bookmarking system delicious1 and artists on online music system last.fm2 are considered as social resources. The foremost feature of this problem is that there are no labeled negative resources/examples available for learning a recommender/classifier. The memory based collaborative filtering has served as the most widely used algorithm for social resource recommendation. However, its predictions are based on some ad hoc heuristic rules and its success depends on the availability of a critical mass of users. This paper proposes model based two-step techniques to learn a classifier using positive and unlabeled examples to address personalized resource recommendations. In the first step of these techniques, naïve Bayes classifier is employed to identify reliable negative resources. In the second step, to generate effective resource recommender, classification and regression tree and least square support vector machine (LS-SVM) are exercised. A direct method based on LS-SVM is also put forward to realize the recommendation task. LS-SVM is customized for learning from positive and unlabeled data. Furthermore, the impact of feature selection on our proposed techniques is also studied. Memory based collaborative filtering as well as our proposed techniques exploit usage data to generate personalized recommendations. Experimental results show that the proposed techniques outperform existing method appreciably.

Keywords


Learning from positive and unlabeled data; memory based collaborative filtering; personalized resource recommender

Full Text:

PDF

Refbacks

  • There are currently no refbacks.


Copyright (c) 2016 Priyank Thakkar, Dr. K. Kotecha

Creative Commons License
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.