Evaluating Cognitive-Behavioral Fixation via Multimodal User Viewing Patterns on Social Media

Yujie Wang, Yunwei Zhao, Jing Yang, Han Han, Shiguang Shan, Jie Zhang · 2025

Read Evaluating Cognitive-Behavioral Fixation via Multimodal User Viewing Patterns on Social Media by Yujie Wang, Yunwei Zhao, Jing Yang, Han Han, Shiguang Shan, Jie Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-04823.txt.

Introduction

Staying locked onto a narrow slice of content is a pattern social media may encourage. Yujie Wang, lead author with five co-authors, propose a framework for measuring that cognitive-behavioural fixation in a 2025 paper. Psychology has studied such fixation widely, but detecting it from user behaviour by computer remains underexplored.

Their framework combines a multimodal topic extraction module with a fixation scoring module. The aim is to score that fixation from what users view.. Together they give adaptive, layered, and interpretable assessments from viewing patterns across content types. Tests on existing benchmarks and a new mixed-type dataset show the approach works. The authors release their code for research use.

Catalog record

CREATORYujie Wang, Yunwei Zhao, Jing Yang, Han Han, Shiguang Shan, Jie Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-04823

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Evaluating Cognitive-Behavioral Fixation via Multimodal User Viewing Patterns on Social Media Yujie Wang, Yunwei Zhao, Jing Yang, Han Han, Shiguang Shan, Jie Zhang Abstract Digital social media platforms frequently contribute to cognitive-behavioral fixation, a phenomenon in which users exhibit sustained and repetitive engagement with narrow content domains. While cognitive-behavioral fixation has been extensively studied in psychology, methods for computationally detecting and evaluating such fixation remain underexplored. To address this gap, we propose a novel framework for assessing cognitive-behavioral fixation by analyzing users’ multimodal social media engagement patterns. Specifically, we introduce a multimodal topic extraction module and a cognitive-behavioral fixation quantification module that collaboratively enable adaptive, hierarchical, and interpretable assessment of user behavior. Experiments on existing benchmarks and a newly curated multimodal dataset demonstrate the effectiveness of our approach, laying the groundwork for scalable computational analysis of cognitive fixation. All code in this project is publicly available for research purposes at https://github.com/Liskie/cognitive-fixation-evaluation 1 Introduction Digital media and online platforms significantly shape user behavior through personalized algorithms and constant connectivity, which curate the information users encounter and structure their engagement patterns. This pervasive mediation often amplifies existing preferences and habits, potentially leading to self-reinforcing cognitive and behavioral feedback loops Vicario et al. 2016 ; Cinelli et al. 2021 . In other words, rather than merely reflecting individual preferences, social media and recommendation systems may actively shape and constrain perceptions, preferences, and decision-making processes. While such tailored content enhances user experience, it also raises concerns about cognitive-behavioral fixation phenomena, which is defined as obsessive preoccupations with specific ideas or activities, impairing balanced information processing and flexible thinking Dielenberg 2024 . Common manifestations include misinformation loops Vicario et al. 2016 , echo chambers Cinelli et al. 2021 , and compulsive engagement patterns Markett and Montag 2023 . For example, misinformation loops emerge when recommendation algorithms continually amplify users’ preexisting misconceptions with ideologically aligned but factually inaccurate content Vicario et al. 2016 . Similarly, echo chambers restrict users to homogeneous viewpoints, exacerbating confirmation bias and social polarization Cinelli et al. 2021 . Meanwhile, compulsive engagement driven by persuasive design (e.g., endless feeds, personalized notifications) fosters addictive and repetitive usage behaviors Markett and Montag 2023 . Collectively, these fixation effects distort perceptions of reality, entrench false beliefs, deepen societal divisions, and negatively impact mental health, with consequences including anxiety, diminished well-being, and reduced attention spans Markett and Montag 2023 . Addressing cognitive-behavioral fixation is crucial not only for safeguarding individual mental health but also for maintaining societal stability, as fixation-driven behaviors (e.g., widespread misinformation acceptance, extreme polarization) can undermine public discourse and trust in information ecosystems. Despite these risks, a significant research gap remains: no existing computational framework can automatically detect and quantify cognitive-behavioral fixation within online environments . Existing studies often treat echo chambers, misinformation spread, and compulsive behaviors separately, lacking a unified model to quantify fixation holistically. Psychological studies, while insightful, primarily rely on qualitative methods such as case studies, surveys, or controlled experiments Meloy and Rahman 2020 , making them difficult to scale for automated, real-time digital behavior analysis. In response, we formalize the novel task of computational cognitive-behavioral fixation evaluation for social media users. We propose a multimodal analytical framework designed to automatically detect and quantify fixation based on users’ digital interactions, specifically focusing on text posts and video content. Our framework supports adaptive, hierarchical, and interpretable evaluation of cognitive-behavioral fixation in real-world social online environments. Figure 1: Overview of our proposed framework for cognitive-behavioral fixation evaluation. The key contributions of this work are: 1. We introduce the first formalization of cognitive-behavioral fixation evaluation as a computational task for social media behavior analysis. 2. We propose an adaptive , hierarchical , and interpretable framework that generalizes across modalities and dataset scales, extracts multi-level user interests, and quantifies fixation. 2 Related Work 2.1 Cognitive-Behavioral Fixation in Psychology Cognitive-behavioral fixation describes obsessive preoccupation with specific ideas or behaviors that impairs flexible thinking and decision-making Dielenberg 2024 . Psychology traditionally studies fixation through clinical observation and case studies, linking it to confirmation bias, belief perseverance, and cognitive rigidity Lord et al. 1979 ; Meloy and Rahman 2020 . Such mechanisms clearly manifest in online contexts like echo chambers and misinformation loops, where repetitive exposure reinforces narrow beliefs and resistance to counter-evidence, exacerbating polarization and impairing public discourse Zollo et al. 2017 ; Cinelli et al. 2021 . While psychology provides rich theoretical foundations, it lacks automated, scalable methods for detecting fixation in large-scale digital behavior data—a gap this work addresses. 2.2 Topic Extraction Classical topic modeling approaches, such as Latent Dirichlet Allocation (LDA) Blei et al. 2001 and Non-negative Matrix Factorization (NMF) Lee and Seung 1999 , enable unsupervised extraction of themes but struggle with short or noisy texts. Recent neural topic models, including Embedded Topic Model (ETM) and Contextualized Topic Models (CTM), leverage embedding-based representations to enhance coherence and scalability Bianchi et al. 2021a ; Dieng et al. 2020 . DeTiME further improved coherence through diffusion-enhanced large language model embeddings Xu et al. 2023 . Multimodal topic extraction methods, such as Multimodal LDA Zhu et al. 2010 and CLIP-based models Luo et al. 2022 , integrate visual and textual inputs into a unified semantic space. Our work uniquely applies multimodal topic extraction for cognitive-level behavioral analysis. 2.3 Behavioral Analysis and Metrics Quantitative metrics such as Shannon entropy, KL divergence, and burstiness have widely analyzed user engagement to characterize topical diversity, concentration, and temporal patterns. Shannon entropy measures topical diversity, where lower entropy reflects narrower user focus Weng et al. 2012 ; Sonoda et al. 2022 ; Song et al. 2014 . Topic concentration is typically quantified using the Herfindahl–Hirschman Index (HHI) Zhu et al. 2015 , and KL divergence captures shifts in sequential user interests Sritrakool and Maneeroj 2021 . Burstiness metrics based on inter-event timing highlight clustered activity patterns indicating compulsive or reactionary behavior Goh and Barabási 2008 ; Karsai et al. 2018 . Entropy-based methods have also quantified ideological fixation and selective exposure Muñoz et al. 2024 ; Pratelli et al. 2024 . These existing metrics form the foundation for our multimodal fixation evaluation framework. 3 Methodology 3.1 Problem Formulation We focus on analyzing users’ multimodal browsing behaviors on social media platforms to quantitatively evaluate cognitive-behavioral fixation. Formally, for each user u , we define their viewing history as an ordered sequence: \mathcal{H}^{(u)}=\{h_{1},h_{2},\dots,h_{T}\} (1) where each interaction h_{t} corresponds to a viewed content item (e.g., a post or video) at timestamp t . Each h_{t} contains both textual and visual information. From each content h_{t} , we extract a set of fine-grained topic tags: T(h_{t})=\{\text{topic}_{t,1},\dots,\text{topic}_{t,m}\} (2) representing the semantic essence of the content. These topics form the basis for later analyses. Given the temporal sequence of extracted topics across all viewed content, our goal is to assess the extent to which the user’s attention is narrowly concentrated, repetitive, and persistent over time, i.e., key signatures of cognitive-behavioral fixation. 3.2 Overall Framework Our framework enables adaptive , hierarchical , and interpretable analysis of cognitive-behavioral fixation in real-world social media environments. Multimodal Hierarchical Topic Extraction This module processes user interactions with textual and visual content to extract two levels of topic representations. The first-level consists of fine-grained topic phrases that summarize individual content items.

Read the complete text