Revealing Subtle Marketing: A Benchmark and Framework for Social Media Covert Advertisements Detection
Abstract: Currently, posting covert advertisements on social media is an increasingly common marketing strategy.
This practice will mislead users, which may influence their decisions and cause unfair competition, highlighting the urgent need for effective detection methods.
However, research on this topic remains limited.
In this study, we formalize the covert advertisement detection task and present the first social media covert advertisement benchmark.
The benchmark includes Chinese and English posts collected from two representative social media platforms (Rednote and Instagram) with manually annotated labels.
We evaluate several multimodal methods and find that, as covert advertisements can appear within a single modality or through cross-modal interplay, these methods struggle with effective detection and fail to adequately balance single-modal and fused features.
To address this challenge, we propose SCAN (Social-media Covert Advertisement Detection using Multi-view Network), a framework that leverages cooperative training to better balance and utilize both single-modal and fused features.
Our results show that SCAN can further advance covert advertisement detection performance.
We believe our benchmark and method will contribute to future research in social media covert advertisement detection.
Paper Type: Long
Research Area: Computational Social Science and Cultural Analytics
Research Area Keywords: NLP tools for social analysis
Languages Studied: Chinese,English
Submission Number: 1011
Loading