A methodology for learning, analyzing, and mitigating social influence bias in recommender systems

Krishnan, Sanjay; Patel, Jay; Franklin, Michael J.; Goldberg, Ken

doi:10.1145/2645710.2645740

Cited by 79 publications

(48 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Prior research also found that recommendations not only significantly affect consumers' preference ratings but also their economic behavior. Krishnan et al suggest introducing the 3-step rating system applied by their case-study, the California Report Card, and then use machine learning to estimate the social influence bias and to correct it before the review is posted onto the platform [28]. The proposed approach is different from the existing work due to the personalization applied in de-biasing the individual ratings according to the existing rating information.…”

Section: Related Workmentioning

confidence: 99%

DBT Recommender: Improved Trustworthiness of Ratings through De-Biasing Tendency of Users

Mukkamula¹

2018

IJIES

View full text Add to dashboard Cite

Recent advancements in business strategies marked the significance of e-commerce in marketing any service of the organization. Moreover, users are quiet dependent on the average ratings of the products showcased in the marketing interface in turn these average ratings made remarkable impact on sales phenomena of the product. The average rating of a product is the aggregation of individual users ratings biased with the tendency of the user towards publishing the opinion. The optimistic user tends to give a slight high rating than a neutral judgement and vice versa with a pessimistic user. However, these biased ratings produce an aggregate value that is degraded with its trustworthiness. This paper proposed a novel approach named DBT (De-biased Tendency) Recommender to analyze the bias in product rating which recalculates the average ratings of the products by making user tendencies as part of the process. The solution implemented on a big data environment on demand of high computation complexity involved in the process. Experimental results had shown a significant improvement in the trustworthiness of the product ratings with the proposed approach.

show abstract

Section: Related Workmentioning

confidence: 99%

DBT Recommender: Improved Trustworthiness of Ratings through De-Biasing Tendency of Users

Mukkamula¹

2018

IJIES

View full text Add to dashboard Cite

show abstract

“…Regardless of the topic, these research studies were conducted for ratings in a wide array of contexts: from political opinions (Krishnan et al, 2014) to entertainment products like movies (Schlosser, 2005;Lee et al, 2015) and music (Hu et al, 2009); from software (Duan et al, 2009) to physical products like printed books (Godes and Mayzlin, 2004), to services like restaurants (Cai et al, 2009) and accommodation (Yacouel and Fleischer, 2012).…”

Section: Literature Review 21 a General Assessmentmentioning

confidence: 99%

“…Empirical data have also been used to study rating behavior patterns on the Internet (Krishnan et al, 2014). In the remaining of this section, we review the literature by research area, recalling the differences in sectors and methodologies when appropriate.…”

Section: Literature Review 21 a General Assessmentmentioning

confidence: 99%

“…Using this methodology, Muchnik et al (2013) found asymmetric herding: whereas positive social influence accumulates, creating a tendency toward rating bubbles (the likelihood of positive ratings increased by 32% as a consequence of a manipulated initial positive rating, leading to an increase in the final rating by 25% on average), negative social influence inspired users to correct manipulated negative ratings. Krishnan et al (2014) studied individual behavior in a rating system evaluating California state policies (the California Report Card), where the individual rating builds in three phases: in the first one, users rate the policy; then they observe the median rating; finally they are allowed to change their own evaluation (although they are not explicitly told to be allowed to do so) before submitting the final score. Social influence bias is found to be relevant, since 35% of users changed their rating after the observation of the median and, for this group, final ratings were significantly more concentrated around the median than initial ratings, the effect being symmetric for those with initial ratings above and below the median.…”

Section: Social Influence Bias In Online Ratingsmentioning

confidence: 99%

See 1 more Smart Citation

Social Influence Bias in Online Ratings: A Field Experiment

2016

View full text Add to dashboard Cite

The aim of this paper is to study the empirical phenomenon of rating bubbles, i.e. clustering on extremely positive values in e-commerce platforms and rating web sites. By means of a field experiment that exogenously manipulates prior ratings for a hotel in an important Italian tourism destination, we investigate whether consumers are influenced by prior ratings when evaluating their stay (i.e., social influence bias). Results show that positive social influence exists, and that herd behavior is asymmetric: information on prior positive ratings has a stronger influence on consumers' rating attitude than information on prior mediocre ratings. Furthermore, we are able to exclude any brag-or-moan effect: the behavior of frequent reviewers, on average, is not statistically different from the behavior of consumers who have never posted ratings online. Yet, non-reviewers exhibit a higher influence to excellent prior ratings, thus lending support to the social influence bias interpretation. Finally, also repeat customers are affected by prior ratings, although to a lesser extent with respect to new customers.

show abstract

“…In the OpinionSpace project [17], our results suggested that 2D visual interfaces for presenting ideas increased engagement. We have further explored best practices for online civic engagement platforms [52], bias mitigation due to social influences [53], and applying similar ideas to large college course evaluations [54]. The development setting poses a number of new challenges including: (1) Internet connectivity, (2) language and cultural differences, and (3) rapid field analysis.…”

Section: Related Workmentioning

confidence: 99%

DevCAFE 1.0: A participatory platform for assessing development initiatives in the field

Nonnecke

Krishnan

Patel

et al. 2015

2015 IEEE Global Humanitarian Technology Conference (GHTC)

Self Cite

View full text Add to dashboard Cite

Abstract-The design and assessment of development initiatives is increasingly participatory, where decision makers consider feedback from affected populations. While digital data collection facilitates faster and more reliable analysis, existing data collection tools are not optimized for unstructured qualitative (textual) data and peer-topeer participant collaboration. In this paper, we propose a system called the Development Collaborative Assessment and Feedback Engine version 1.0 (DevCAFE), a customizable participatory assessment platform that collects and integrates quantitative assessment, qualitative feedback and peer-to-peer collaborative filtering. DevCAFE incorporates a library of statistical analyses for researchers to quickly identify quantitative and qualitative trends while collecting field data. DevCAFE can run on any mobile device with a web-browser and can work with or without Internet connectivity. We present results from two pilot projects: (1) 137 participants evaluating family planning education trainings at three Nutrition Education Centers in rural Uganda, and (2) 4,518 participants evaluating policy priorities for elected leaders in the June 2015 Mexico mid-term elections. DevCAFE collected over 19,000 peer-to-peer ratings of 336 submitted ideas. Feedback gathered through DevCAFE enabled targeted reforms to the family planning efforts in Uganda and the need for increased government attention to public safety in Mexico. Case studies and interactive demos are available at: http://opinion.berkeley.edu/devcafe/

show abstract

A methodology for learning, analyzing, and mitigating social influence bias in recommender systems

Cited by 79 publications

References 35 publications

DBT Recommender: Improved Trustworthiness of Ratings through De-Biasing Tendency of Users

DBT Recommender: Improved Trustworthiness of Ratings through De-Biasing Tendency of Users

Social Influence Bias in Online Ratings: A Field Experiment

DevCAFE 1.0: A participatory platform for assessing development initiatives in the field

Contact Info

Product

Resources

About