Abstract

The focus of this research is the N “most popular” (Top-N) news recommender systems (NRS), widely used by media sites (e.g. New York Times, BBC, Wall Street Journal all prominently use this). This common recommendation process is known to have major limitations in terms of creating artificial amplification in the counts of recommended articles and that it is easily susceptible to manipulation. To address these issues, probabilistic NRS has been introduced. One drawback of the probabilistic recommendations is that it potentially chooses articles to recommend that might not be in the current “best” list. However, the probabilistic selection of news articles is highly robust towards common manipulation strategies. This paper compares the two variants of NRS (Top-N and probabilistic) based on (1) accuracy loss (2) distortion in counts of articles due to NRS and (3) comparison of probabilistic NRS with an adapted "influence limiter" heuristic.

Download

Share

COinS
 

Analysis of Probabilistic News Recommender Systems

The focus of this research is the N “most popular” (Top-N) news recommender systems (NRS), widely used by media sites (e.g. New York Times, BBC, Wall Street Journal all prominently use this). This common recommendation process is known to have major limitations in terms of creating artificial amplification in the counts of recommended articles and that it is easily susceptible to manipulation. To address these issues, probabilistic NRS has been introduced. One drawback of the probabilistic recommendations is that it potentially chooses articles to recommend that might not be in the current “best” list. However, the probabilistic selection of news articles is highly robust towards common manipulation strategies. This paper compares the two variants of NRS (Top-N and probabilistic) based on (1) accuracy loss (2) distortion in counts of articles due to NRS and (3) comparison of probabilistic NRS with an adapted "influence limiter" heuristic.