Logo image
Back
ProbitUCB: A Novel Method for Review Ranking
Conference proceeding   Peer reviewed

ProbitUCB: A Novel Method for Review Ranking

Wanying Ding, Yue Shang, Dae Hoon Park, Lifan Guo and Xiaohua Hu
TRENDS AND APPLICATIONS IN KNOWLEDGE DISCOVERY AND DATA MINING, PAKDD 2015, v 9441
01 Jan 2015
DOI: https://doi.org/10.1007/978-3-319-25660-3_1

Abstract

Computer Science Computer Science, Artificial Intelligence Science & Technology Technology
Online reviews play an important role in facilitating customers in making online purchase decisions. But with the dramatic increase in volume, it will cost customers hours going through all the reviews. This paper proposes a review ranking algorithm to present the most helpful reviews ahead, saving consumers' plenty of time in review hunting. Our ProbitUCB model implements a probabilistic kernel embedded UCB (Upper Confident Bound) ranking framework, and adopts a self-learning mechanism to distinguish out helpful reviews. Comparing to the current models, ProbitUCB's advantage is listing as follows: (1) it ranks under the exploit and explore mechanism, reducing the error brought from probability estimation inaccuracy; (2) it is training dataset free, saving users enormous amount of time in labeling data, which is required for most supervised methods; (3) it considers various potential features to rank, remedying the defect of only using word information in most unsupervised methods; (4) it adjusts the values of hyper parameters automatically, solving the intuitively value setting problem in many related work. Finally, we experiment our model on 6 datasets, and compare its performance with 10 other classical learn to rank algorithms, and the results show that our algorithm outperform all of them.

Metrics

1 Record Views

Details

InCites Highlights

Data related to this publication, from InCites Benchmarking & Analytics tool:

Collaboration types
Domestic collaboration
Web of Science research areas
Computer Science, Artificial Intelligence
Logo image