Temporal mechanisms of polarization in online reviews

Matakos, Antonis; Tsaparas, Panayiotis

doi:10.1109/asonam.2016.7752286

Cited by 8 publications

(6 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Polarized ratings can indicate controversial items; a recommender that mistakenly estimates a high rating for what a user would perceive as a low rating (or vice versa) can be a serious error particularly in domains like politics [9,16]. Following previous work [28], we adopt a variance threshold, V AR(R i ) >= 3, to identify items with polarized ratings. We also ensure that the items have at least a minimum number of ratings, |R i |>= 5, leading to the smaller dataset in Table 2.…”

Section: Data-driven Studymentioning

confidence: 99%

Improving the Estimation of Tail Ratings in Recommender System with Multi-Latent Representations

Zhao

Zhu

Zhang

et al. 2020

Proceedings of the 13th International Conference on Web Search and Data Mining

View full text Add to dashboard Cite

The importance of the distribution of ratings on recommender systems (RS) is well-recognized. And yet, recommendation approaches based on latent factor models and recently introduced neural variants (e.g., NCF) optimize for the head of these distributions, potentially leading to large estimation errors for tail ratings. These errors in tail ratings that are far from the mean predicted rating fall out of a uni-modal assumption underlying these popular models, as we show in this paper. We propose to improve the estimation of tail ratings by extending traditional single latent representations (e.g., an item is represented by a single latent vector) with new multi-latent representations for better modeling these tail ratings. We show how to incorporate these multi-latent representations in an end-to-end neural prediction model that is designed to better reflect the underlying ratings distributions of items. Through experiments over six datasets, we find the proposed model leads to a significant improvement in RMSE versus a suite of benchmark methods. We also find that the predictions for the most polarized items are improved by more than 15%.

show abstract

Section: Data-driven Studymentioning

confidence: 99%

Improving the Estimation of Tail Ratings in Recommender System with Multi-Latent Representations

Zhao

Zhu

Zhang

et al. 2020

Proceedings of the 13th International Conference on Web Search and Data Mining

View full text Add to dashboard Cite

show abstract

“…In the absence of polarization, the distribution of opinions is either J-shaped or bell shaped. However, as polarization emerges, the resulting distribution shifts to a U-shaped distribution with two peaks emerging around the two dominant and confronted opinions at the extreme sides of the rating scale [33,44]. Different examples of such distributions are shown in figure 1.2.…”

Section: Detecting Polarizationmentioning

confidence: 99%

“…In another research, [44] studied the temporal evolution of text-based reviews. In particular, the authors investigated the self-selection bias, where only users that strongly disagree with the item's current average rating will make an effort to provide explicit feedback.…”

Section: Rated and Recommended Polarization: Polarization In Recommenmentioning

confidence: 99%

“…We extract another set of features, Feature Set Two, inspired by the literature [33,36,44,104]. We start by estimating the best number of Gaussian distributions that can be fitted to the item's rating histogram.…”

Section: Feature Extractionmentioning

confidence: 99%

“…We first build a rating histogram of each item in the datasets. We compare our Polarization Detection Classifier (PDC) to several recent models [33,36,44,104]. All of these methods use metrics that measure the polarization score for each item based on various criteria.…”

Section: Book-crossings [99]mentioning

confidence: 99%

See 2 more Smart Citations

Peeking into the other half of the glass : handling polarization in recommender systems.

Badami¹

View full text Add to dashboard Cite

This dissertation is about filtering and discovering information online while using recommender systems. In the first part of our research, we study the phenomenon of polarization and its impact on filtering and discovering information. Polarization is a social phenomenon, with serious consequences, in real-life, particularly on social media. Thus it is important to understand how machine learning algorithms, especially recommender systems, behave in polarized environments. We study polarization within the context of the users' interactions with a space of items and how this affects recommender systems. We first formalize the concept of polarization based on item ratings and then relate it to the item reviews, when available. We then propose a domain independent data science pipeline to automatically detect polarization using the ratings rather than the properties, typically used to detect polarization, such as item's content or social network topology. We perform an extensive comparison of polarization measures on several benchmark data sets and show that our polarization detection framework can detect different degrees of polarization and outperforms existing measures in capturing an intuitive notion of polarization. We also investigate and uncover certain peculiar patterns that are characteristic of environments where polarization emerges: A machine learning algorithm finds it easier to learn discriminating models in polarized environments: The models will quickly learn to v keep each user in the safety of their preferred viewpoint, essentially, giving rise to filter bubbles and making them easier to learn. After quantifying the extent of polarization in current recommender system benchmark data, we propose new counter-polarization approaches for existing collaborative filtering recommender systems, focusing particularly on the state of the art models based on Matrix Factorization. Our work represents an essential step toward the new research area concerned with quantifying, detecting and counteracting polarization in human-generated data and machine learning algorithms.We also make a theoretical analysis of how polarization affects learning latent factor models, and how counter-polarization affects these models. In the second part of our dissertation, we investigate the problem of discovering related information by recommendation of tags on social media micro-blogging platforms. Real-time micro-blogging services such as Twitter have recently witnessed exponential growth, with millions of active web users who generate billions of micro-posts to share information, opinions and personal viewpoints, daily. However, these posts are inherently noisy and unstructured because they could be in any format, hence making them difficult to organize for the purpose of retrieval of relevant information. One way to solve this problem is using hashtags, which are quickly becoming the standard approach for annotation of various information on social media, such that varied posts about the same or related topic are annotated wi...

show abstract

Comprehensive examination of online reviews divergence over time and platform types

Kirilenko,

Stepchenkova,

Gromoll

et al. 2024

International Journal of Hospitality Management

View full text Add to dashboard Cite

Temporal mechanisms of polarization in online reviews

Cited by 8 publications

References 10 publications

Improving the Estimation of Tail Ratings in Recommender System with Multi-Latent Representations

Improving the Estimation of Tail Ratings in Recommender System with Multi-Latent Representations

Peeking into the other half of the glass : handling polarization in recommender systems.

Comprehensive examination of online reviews divergence over time and platform types

Contact Info

Product

Resources

About