Brief context
Publication timing, weekly edition context, and source links for this brief.
Original paper
The executive brief below is grounded in the source paper and linked back to the arXiv abstract.
Current benchmarks for evaluating large language models (LLMs) in social media moderation completely overlook a serious threat: covert advertisements, which disguise themselves as regular posts to deceive and mislead consumers into making purchases, leading to significant ethical and legal concerns. In this paper, we present the CHASM, a first-of-its-kind dataset designed to evaluate the capability of Multimodal Large Language Models (MLLMs) in detecting covert advertisements on social media. CHASM is a high-quality, anonymized, manually curated dataset consisting of 4,992 instances, based on real-world scenarios from the Chinese social media platform Rednote. The dataset was collected and annotated under strict privacy protection and quality control protocols. It includes many product experience sharing posts that closely resemble covert advertisements, making the dataset particularly challenging.The results show that under both zero-shot and in-context learning settings, none of the current MLLMs are sufficiently reliable for detecting covert advertisements.Our further experiments revealed that fine-tuning open-source MLLMs on our dataset yielded noticeable performance gains. However, significant challenges persist, such as detecting subtle cues in comments and differences in visual and textual structures.We provide in-depth error analysis and outline future research directions. We hope our study can serve as a call for the research community and platform moderators to develop more precise defenses against this emerging threat.
Executive brief
A short business-reader brief that explains why the paper matters now and what to watch or do next.
Why this is worth your attention
Covert advertising is becoming a moderation and compliance problem that looks less like spam detection and more like fraud review: the evidence is scattered across captions, images, comments, and creator behavior. This paper shows that generic multimodal models are not yet dependable for that job, but targeted fine-tuning on a curated dataset can move performance meaningfully. If the result generalizes, the advantage shifts toward platforms and vendors with proprietary moderation data and workflows that can keep humans in the loop for ambiguous cases.
- Do not assume a frontier multimodal model can be dropped into ad-disclosure enforcement and trusted. The paper’s best prompted baselines are around 0.59 F1, which is useful for triage but risky for user penalties, creator demonetization, or compliance decisions without human review.
- The practical lesson is not “buy a bigger model”; it is that domain-specific training data changes the economics. A fine-tuned open-source Qwen2.5-7B reaches 0.756 F1, suggesting platforms and moderation vendors may get better performance from targeted datasets than from expensive generic reasoning models.
- A serious detector has to inspect images, post text, comments, and comment evolution over time—not just the caption at upload. Ask vendors whether they ingest comment threads and image-embedded cues, because missed comment evidence remains a major error source even after fine-tuning.
- The hard cases are not obvious spam; they are product-sharing posts that may or may not contain paid intent. Because the dataset includes only 12.3% covert ads and annotator agreement is moderate, any deployment needs clear policy thresholds, appeal paths, and reviewer calibration—not just a model score.
- The dataset is valuable, but it is built from RedNote and Chinese-language social commerce behavior. The adoption signal that would matter is comparable performance on other platforms, languages, and creator-commerce formats where disclosure norms and evasion tactics differ.
Evidence ledger
The strongest claims in the brief, along with the confidence and citation depth behind them.
CHASM is a 4,992-instance multimodal dataset from RedNote, including 612 covert-ad positives and 1,127 product-sharing challenging negatives.
Prompted frontier and open-source multimodal models are not reliable enough for covert-ad detection out of the box.
Fine-tuning open-source multimodal models on CHASM materially improves performance, with Qwen2.5-7B reaching 0.756 F1 and 0.852 AUC.
Effective detection needs multimodal and temporal signals, especially images and comments, rather than static text-only review.
The task remains ambiguous and not ready for fully automated punitive action without reviewer oversight.
Related briefs
More plain-English summaries from the archive with nearby topics or operator relevance.
cs.LG
AgentOpt v0.1 Technical Report: Client-Side Optimization for LLM-Based Agent
Wenyue Hua et al.
cs.LG
Gym-Anything: Turn any Software into an Agent Environment
Pranjal Aggarwal, Graham Neubig, Sean Welleck
cs.CR
The Salami Slicing Threat: Exploiting Cumulative Risks in LLM Systems
Yihao Zhang et al.
cs.LG
Bridging MARL to SARL: An Order-Independent Multi-Agent Transformer via Latent Consensus
Zijian Zhao, Jing Gao, Sen Li