Volume 121, Number 6, March 2018
|Number of page(s)||7|
|Section||Interdisciplinary Physics and Related Areas of Science and Technology|
|Published online||17 May 2018|
Deviation-based spam-filtering method via stochastic approach
1 Department of Physics, Sungkyunkwan University - Suwon 16419, Republic of Korea
2 Department of Physics, Inha University - Incheon 22212, Republic of Korea
Received: 5 February 2018
Accepted: 26 April 2018
In the presence of a huge number of possible purchase choices, ranks or ratings of items by others often play very important roles for a buyer to make a final purchase decision. Perfectly objective rating is an impossible task to achieve, and we often use an average rating built on how previous buyers estimated the quality of the product. The problem of using a simple average rating is that it can easily be polluted by careless users whose evaluation of products cannot be trusted, and by malicious spammers who try to bias the rating result on purpose. In this letter we suggest how trustworthiness of individual users can be systematically and quantitatively reflected to build a more reliable rating system. We compute the suitably defined reliability of each user based on the user's rating pattern for all products she evaluated. We call our proposed method as the deviation-based ranking, since the statistical significance of each user's rating pattern with respect to the average rating pattern is the key ingredient. We find that our deviation-based ranking method outperforms existing methods in filtering out careless random evaluators as well as malicious spammers.
PACS: 89.75.-k – Complex systems / 89.65.-s – Social and economic systems / 89.20.Ff – Computer science and technology
© EPLA, 2018
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.