Zhen Tan, Dawei Li, Song Wang, Alimohammad Beigi, Bohan Jiang, Amrita Bhattacharjee, Mansooreh Karami, Jundong Li, Lu Cheng, Huan Liu. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

An Echo Chamber on social media refers to the environment where like-minded people hear the echo of each others' voices, opinions, or beliefs, which reinforce their own. Echo Chambers can turn social media platforms into collaborative venues that polarize and radicalize users rather than broadening their exposure to diverse information. Having a quantified metric for measuring the Echo Chamber effect can aid moderators and policymakers in tracking and mitigating online polarization and radicalization. Existing methods for Echo Chamber detection are either one-dimensional, only considering the network behavior of users while ignoring their semantic behavior, or require demanding supervised labeling, which is both expensive and less generalizable. This paper proposes a new metric to quantify the Echo Chamber effect using Transformer models for context-sensitive processing of natural language (NLP). Our metric quantifies (1) the effect of an Echo Chamber through the inverse effect of user diversity , and (2) polarization by means of user separability between two Echo Chambers in a topic. Leveraging this metric, we further propose an NLP-based embedding that represents the users' activity. Our model is simultaneously effective, computationally cheap, and unsupervised. As our method is unsupervised, it makes existing collaborative moderation efforts to thwart Echo Chamber effects more efficient by addressing the problem of identifying narrow information bases for algorithmic biases and misinformation detection. We run our analysis on three recent highly controversial political topics and a non-controversial topic: Russo-Ukrainian War, Abortion, Gun-Control, and SXSW music festival. Our results offer data-driven findings such as a higher Echo Chamber effect among Republicans over Democrats and diverse explicit support for Ukraine, especially among Democrats. We also observe a direct relationship between the Echo Chamber effect and polarization while observing that the low Echo Chamber effect for the Russo-Ukraine war is accompanied by a low polarization; and vice versa for Gun-Control.

Contrastive self-supervised learning has become a prominent technique in representation learning. The main step in these methods is to contrast semantically similar and dissimilar pairs of samples. However, in the domain of Natural Language Processing (NLP), the augmentation methods used in creating similar pairs with regard to contrastive learning (CL) assumptions are challenging. This is because, even simply modifying a word in the input might change the semantic meaning of the sentence, and hence, would violate the distributional hypothesis. In this review paper, we formalize the contrastive learning framework, emphasize the considerations that need to be addressed in the data transformation step, and review the state-of-the-art methods and evaluations for contrastive representation learning in NLP. Finally, we describe some challenges and potential directions for learning better text representations using contrastive methods.

With social media being a major force in information consumption, accelerated propagation of fake news has presented new challenges for platforms to distinguish between legitimate and fake news. Effective fake news detection is a non-trivial task due to the diverse nature of news domains and expensive annotation costs. In this work, we address the limitations of existing automated fake news detection models by incorporating auxiliary information (e.g., user comments and user-news interactions) into a novel reinforcement learning-based model called REinforced Adaptive Learning Fake News Detection (REAL-FND). REAL-FND exploits cross-domain and within-domain knowledge that makes it robust in a target domain, despite being trained in a different source domain. Extensive experiments on real-world datasets illustrate the effectiveness of the proposed model, especially when limited labeled data is available in the target domain.