Earticle

초록

영어

Popularity based top-N news recommender systems have often been criticized for limitations like monotonicity and popularity amplification. The probabilistic selection of news articles has been found to theoretically address some of the major limitations of top-N recommendation, but its potential impact on reader behavior has not been examined. Based on the variety-seeking nature of consumption, we hypothesize that the probabilistic system should outperform the top-N system across measures of web traffic, user satisfaction, and reader engagement selected using the Balanced Score Card. For evaluation, we conducted a couple of laboratory based experiments. In the first part, 68 participants were randomly assigned to two groups who were shown the same website. Based on the measures, we calculated the sample size for A/B testing. Web traffic and user satisfaction scores were higher for readers who were shown the probabilistic recommendations by 20% and 10%, respectively. Results regarding reader engagement were statistically insignificant. Based on our findings, we conclude that the probabilistic system can be used by media houses as an alternative to the top-N system.

목차

ABSTRACT
Ⅰ. Introduction
Ⅱ. Prior Theory and Research
2.1. Probabilistic Sampling for Recommen dation
2.2. Variety Seeking Behaviour and News Recommendation
2.3. Balanced Scorecard and News Recommendation
Ⅲ. Hypothesis Development
Ⅳ. Experimental Design
4.1. Metrics Operationalization
4.2. Data Collection
4.3. Pre-Test Analysis
4.4. Post-test Analysis
Ⅴ. Discussion
5.1. Theoretical Implications
5.2. Implications for Practice
5.3. Limitations
5.4. Conclusion

저자

• Somnath Bhattacharya [ Assistant Professor, CMS B School, Jain University, India ] Corresponding Author
• Shankar Prawesh [ Assistant Professor, Department of Management Sciences, IIT Kanpur, India ]

참고문헌