Pith. sign in

REVIEW 4 major objections 2 minor 41 references

Maximizing Mutual Information Between Prompt and Response Improves LLM Performance With No Additional Data

T0 review · 4 major / 2 minor · reviewed 2026-07-14 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Maximizing mutual information between prompt and response improves LLM performance with no extra data or external supervision.

desk verdict We only have the MIPO abstract; the full text is a different paper (LLM-MRD), so the PMI claim and all gains are uncheckable. read the letter →

arxiv 2603.19294 v5 pith:HVONQ4JD submitted 2026-03-10 cs.LG cs.AIcs.CL

Hyunji Nam , Haoran Li , Natasha Jaques This is my paper · ORCID
classification cs.LGcs.AIcs.CL
keywords mutualinformationpreferenceoptimizationself-improvementlargelanguagemodelscontrastivepairsDirectpersonalizationnoadditionaldata
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Post-training for large language models usually depends on human labels or external verifiers, which are scarce, expensive, and poorly suited to non-verifiable goals such as personalization. This paper introduces Mutual Information Preference Optimization (MIPO), a contrastive data-augmentation method that needs no new corpus: for each prompt it generates a positive response conditioned on that prompt and a negative response conditioned on a random, unrelated prompt. Applying Direct Preference Optimization to these pairs is shown to maximize the pointwise mutual information between prompts and responses under the base model itself. On 1–7B Llama and Qwen instruct models the procedure yields 3–16 percent gains (and 51 percent on one small model) on personalization relative to prompting baselines, and 1–20 percent gains on math and multiple-choice question answering, all without additional data or external supervision. The result suggests a practical self-improvement route that relies only on intrinsic contrastive signals and applies to both verifiable and non-verifiable domains.

What carries the argument

Mutual Information Preference Optimization (MIPO): a contrastive pairing procedure that forms DPO training pairs from correct-prompt positives and random-prompt negatives so the resulting update maximizes base-model pointwise mutual information between prompt and response.

What would settle it

Run the identical DPO pipeline but replace the random-prompt negatives with responses sampled from the correct prompt (or with pure noise); if the reported gains on personalization and math vanish, the claim that this specific contrastive construction maximizes useful PMI is falsified.

Watch

Extended reading notes

Core claim

Constructing preference pairs by taking a response generated under the correct prompt as preferred and a response generated under a random unrelated prompt as dispreferred, then training with Direct Preference Optimization, maximizes pointwise mutual information under the base LLM between prompts and model responses; this single intrinsic objective produces consistent accuracy gains on personalization, math, and multiple-choice tasks without any extra labeled data or external verifiers.

Load-bearing premise

The method assumes that a response produced under a random, unrelated prompt is a sufficiently informative negative, so that preference learning on these pairs truly raises useful mutual information rather than merely teaching the model to reject obvious mismatches.

Editorial extensions

If this is right

  • Self-improvement loops can run without human preference labels or task-specific verifiers by using only random-prompt contrastive pairs.
  • Personalization and other non-verifiable domains become trainable under the same intrinsic objective used for math and multiple-choice answering.
  • Existing instruct models can be further improved post hoc with no new corpus collection.
  • Pointwise mutual information under the base model is a usable stand-alone training signal for preference optimization.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same random-prompt negative construction could transfer to other preference algorithms beyond DPO, such as IPO or KTO.
  • Gains on verifiable tasks without answer checking imply that raising prompt-response mutual information can act as a soft regularizer against generic or off-topic completions.
  • If random negatives prove too easy, harder negatives drawn from near-miss prompts could further strengthen the PMI objective.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 2 minor

Summary. The submission is labeled arXiv:2603.19294 and titled “Maximizing Mutual Information Between Prompt and Response Improves LLM Performance With No Additional Data.” Its abstract proposes Mutual Information Preference Optimization (MIPO): construct DPO preference pairs by sampling a positive response under the true prompt and a negative under a random unrelated prompt, claim that DPO on these pairs maximizes pointwise mutual information under the base LLM, and report 3–16% (up to 51%) personalization gains and 1–20% math/MCQA gains with no extra data or external supervision. The body of the supplied manuscript, however, is an entirely different paper (LLM-MRD: LLM-Guided Multi-View Reasoning Distillation for Fake News Detection), with its own teacher–student architecture, Calibration Distillation, and experiments on Weibo/GossipCop. No MIPO derivation, algorithm, or experiment appears in the full text.

Significance. If the abstract’s claims were supported by a matching manuscript—i.e., a clean PMI–DPO identity under the base model and reproducible gains on both non-verifiable (personalization) and verifiable (math/MCQA) tasks without labels or verifiers—the work would be a meaningful contribution to self-improvement and intrinsic-signal post-training. That significance cannot be assessed here: the provided full text does not contain MIPO, so neither the theoretical identity nor the reported percentage gains can be checked against proofs, ablations, baselines, or error bars.

major comments (4)
  1. Title/abstract vs. body mismatch: the abstract and paper_id describe MIPO (contrastive prompt–response pairs + DPO maximizing base-LLM PMI), but §§1–4 and all figures/tables of the supplied full text describe LLM-MRD (BERT/MAE/CLIP student, LLM teacher multi-view reasoning, Calibration Distillation on Weibo/Weibo21/GossipCop). There is no MIPO method section, no PMI derivation, and no personalization/math/MCQA results. The central claims of 2603.19294 are therefore unevaluable from the materials provided.
  2. Abstract claim that “using Direct Preference Optimization to learn from this paired data maximizes pointwise mutual information under the base LLM”: no equation, proof sketch, or formal statement appears in the manuscript body. Without that derivation it is impossible to verify whether the identity is a genuine result or an artifact of sampling both sides from the same model and pairing procedure (the load-bearing theoretical claim of the abstract).
  3. Empirical claims (3–16%/51% personalization; 1–20% math/MCQA; Llama/Qwen 1–7B) have no supporting tables, ablations, baselines, or error bars in the supplied text. The only experimental content present is for multimodal fake-news detection (ACC/F1 on Weibo etc.), which does not address MIPO. The reported gains cannot be audited.
  4. Core construction assumption (negative = response conditioned on a random unrelated prompt is a valid, non-degenerate DPO negative for maximizing base PMI) is stated only in the abstract and is never justified, stress-tested, or ablated in the body. This is the weakest load-bearing assumption of the claimed method and remains unchecked.
minor comments (2)
  1. Abstract typo: “Experiments with with 1-7B parameter” (duplicated “with”).
  2. Even within the mismatched LLM-MRD body, related-work and figure captions are incomplete or cut off (e.g., §2.1 ends mid-sentence; Fig. 4 discussion is truncated), which would need cleanup if that paper were under review separately.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity can be established: the supplied full manuscript is a different paper (LLM-MRD), so MIPO’s PMI–DPO derivation cannot be inspected.

full rationale

The load-bearing claim of the target paper (MIPO / arXiv:2603.19294) is that DPO on preference pairs formed by a positive response conditioned on the correct prompt and a negative response conditioned on a random unrelated prompt maximizes pointwise mutual information under the base LLM. Only the abstract of that paper is present. The CACHEABLE full-text body is instead the complete manuscript of a different work (LLM-MRD, multimodal fake-news distillation). Without MIPO’s equations, sampling definitions, or proof that the DPO objective on those pairs equals base-model PMI, no self-definitional reduction, fitted-input-as-prediction, self-citation chain, uniqueness import, ansatz smuggling, or renaming can be quoted and exhibited. Hard rule 1 forbids manufacturing circularity from the abstract’s wording alone. Consequently the honest finding is no significant circularity (score 0), with empty steps; the PMI equivalence and reported gains remain uncheckable rather than shown circular.

Assumptions & free parameters 0 free parameters · 3 assumptions · 1 invented entities

Abstract-only review of MIPO. No free parameters, formal axioms, or invented physical entities are specified beyond the methodological construction. Load-bearing modeling choices are listed as domain assumptions.

assumptions (3)
  • ad hoc to paper A response sampled under a random unrelated prompt is a valid negative preference relative to a response sampled under the true prompt for DPO.
    Core data-construction choice stated in the abstract; validity is not independently justified in the available text.
  • domain assumption DPO on these pairs maximizes pointwise mutual information between prompt and response under the base LLM.
    Theoretical claim in the abstract; derivation not available in the provided material.
  • domain assumption Gains from this objective transfer to personalization, math, and multiple-choice QA without external supervision.
    Empirical premise of the abstract’s results section claims.
invented entities (1)
  • Mutual Information Preference Optimization (MIPO)
    purpose: Name for the contrastive pair construction plus DPO training that is claimed to maximize base-model PMI.
    Methodological construct introduced by the paper; not an external physical entity. Independent evidence would be public code and replications.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Maximizing Mutual Information Between Prompt and Response Improves LLM Performance With No Additional Data." pith.science (2026). https://pith.science/paper/HVONQ4JD

@misc{pith2026260319294,
  author       = {Pith},
  title        = {Pith review of: Maximizing Mutual Information Between Prompt and Response Improves LLM Performance With No Additional Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HVONQ4JD}},
  note         = {Machine review of arXiv:2603.19294}
}
read the original abstract

While post-training has successfully improved large language models (LLMs) across a variety of domains, these gains heavily rely on human-labeled data or external verifiers. Existing data has already been exploited, and new data is expensive to collect. Moreover, true intelligence goes far beyond verifiable tasks. Therefore, we need self-improvement frameworks that are less dependent on external signals and more broadly applicable to both verifiable and non-verifiable domains. We propose **Mutual Information Preference Optimization (MIPO)**, a contrastive data augmentation method that constructs preference pairs by generating a positive response conditioning on the correct prompt, and a negative response by conditioning on a random, unrelated prompt. We show that using Direct Preference Optimization to learn from this paired data maximizes pointwise mutual information *under the base LLM* between prompts and model responses. Experiments with with 1-7B parameter Llama and Qwen instruct models show that MIPO achieves 3-16% gains (and 51% increase for Qwen2.5-1.5B-Instruct) on personalization compared to prompting baselines. Surprisingly, MIPO can also be useful in verifiable domains, such as math and multiple-choice question answering, yielding 1-20% gains *without any additional data or external supervision*. These results suggest a promising direction for self-improvement using intrinsic signals derived from contrastive data pairs.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 11 linked inside Pith

  1. [1]

    arXiv preprint arXiv:2201.07472 (2022)

    Chakraborty, R., Bhavsar, M., Dandapat, S.K., Chandra, J.: Detecting stance in tweets: A signed network based approach. arXiv preprint arXiv:2201.07472 (2022)

    arXiv 2022

  2. [2]

    IEEE Access9, 106907–106917 (2021)

    Ni, S., Li, J., Kao, H.Y.: MVAN: Multi-view attention networks for fake news detection on social media. IEEE Access9, 106907–106917 (2021)

    2021

  3. [3]

    In: 2019 IEEE Fifth International Conference on Multimedia Big Data (BigMM), pp

    Singhal, S., Shah, R.R., Chakraborty, T., Kumaraguru, P., Satoh, S.: Spotfake: A multi-modal framework for fake news detection. In: 2019 IEEE Fifth International Conference on Multimedia Big Data (BigMM), pp. 39–47. IEEE (2019)

    2019

  4. [4]

    In: Proceedings of the ACM Web Conference 2022, pp

    Chen, Y., Li, D., Zhang, P., Sui, J., Lv, Q., Tun, L., Shang, L.: Cross-modal ambiguity learning for multimodal fake news detection. In: Proceedings of the ACM Web Conference 2022, pp. 2897–2905. ACM (2022)

    2022

  5. [5]

    arXiv preprint arXiv:2409.01787 (2024)

    Wang, Y., Gu, Z., Zhang, S., Zheng, S., Wang, T., Li, T., Feng, H., Xiao, Y.: LLM- GAN: construct generative adversarial network through large language models for explainable fake news detection. arXiv preprint arXiv:2409.01787 (2024)

    arXiv 2024

  6. [6]

    arXiv preprint arXiv:2506.21557 (2025)

    Yan, K., Liu, M., Liu, Y., Fu, R., Wen, Z., Tao, J., Liu, X.: Debunk and In- fer: Multimodal Fake News Detection via Diffusion-Generated Evidence and LLM Reasoning. arXiv preprint arXiv:2506.21557 (2025)

    arXiv 2025

  7. [7]

    CoRR (2024)

    Liu, X., Li, P., Huang, H., Li, Z., Cui, X., Liang, J., Qin, L., Deng, W., He, Z.: Fakenewsgpt4: advancing multimodal fake news detection through knowledge- augmented LVLMs. CoRR (2024)

    2024

  8. [8]

    Engineering Applications of Artificial Intelligence142, 109931 (2025)

    Jiang, Y., Wang, T., Xu, X., Wang, Y., Song, X., Maynard, D.: Cross-modal aug- mentation for few-shot multimodal fake news detection. Engineering Applications of Artificial Intelligence142, 109931 (2025)

    2025

Show all 41 references
  1. [9]

    arXiv preprint arXiv:2510.05839 (2025) LLM-MRD 15

    Zhou, H., Wei, Y., Yang, J., Zhang, Z.: Towards Robust and Realible Multimodal Fake News Detection with Incomplete Modality. arXiv preprint arXiv:2510.05839 (2025) LLM-MRD 15

    2025

  2. [10]

    International Journal of Web Information Systems21(2), 139–157 (2025)

    Qiu, Y., Ma, K., Zhang, W., Pan, R., Chen, Z.: DSEN-EK: Dual-layer Semantic In- formation Extraction Network with External Knowledge for Fake News Detection. International Journal of Web Information Systems21(2), 139–157 (2025)

    2025

  3. [11]

    Social Network Analysis and Mining 13(1), 101 (2023)

    Comito, C., Caroprese, L., Zumpano, E.: Multimodal fake news detection on social media: a survey of deep learning techniques. Social Network Analysis and Mining 13(1), 101 (2023)

    2023

  4. [12]

    In: 2024 7th In- ternational Conference on Data Science and Information Technology (DSIT), pp

    Ren, S., Liu, Y., Zhu, Y., Bing, W., Ma, H., Wang, W.: MMSFD: Multi-grained and Multi-modal Fusion for Short Video Fake News Detection. In: 2024 7th In- ternational Conference on Data Science and Information Technology (DSIT), pp. 1–11. IEEE (2024)

    2024

  5. [13]

    Advances in Neural Information Processing Systems35, 24824–24837 (2022)

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q.V., Zhou, D., et al.: Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems35, 24824–24837 (2022)

    2022

  6. [14]

    IEEE Transactions on Circuits and Systems for Video Technology 35(7), 6413–6423 (2025)

    Wei, Y., Zhou, H., Yuan, S., Chen, M., Shi, H., Jia, Z., Wang, L., He, X.: DeepMSD: Advancing Multimodal Sarcasm Detection Through Knowledge-Augmented Graph Reasoning. IEEE Transactions on Circuits and Systems for Video Technology 35(7), 6413–6423 (2025)

    2025

  7. [15]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Zhang, C., Feng, Z., Zhang, Z., Qiang, J., Xu, G., Li, Y.: Is LLMs Hallucination Usable? LLM-based Negative Reasoning for Fake News Detection. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39(1), pp. 1031–1039. AAAI Press (2025)

    2025

  8. [16]

    arXiv preprint arXiv:2503.10200 (2025)

    Chen, B., Yue, Z., Chen, S., Wang, Z., Liu, Y., Li, P., Wang, Y.: Lvagent: Long video understanding by multi-round dynamical collaboration of mllm agents. arXiv preprint arXiv:2503.10200 (2025)

    2025

  9. [17]

    High-Confidence Computing4(2), 100211 (2024)

    Yao, Y., Duan, J., Xu, K., Cai, Y., Sun, Z., Zhang, Y.: A survey on large language model (llm) security and privacy: The good, the bad, and the ugly. High-Confidence Computing4(2), 100211 (2024)

    2024

  10. [18]

    arXiv preprint arXiv:2410.19064 (2024)

    Liu, Y., Song, Z., Zhang, J., Zhang, X., Chen, X., Yan, R.: The Stepwise Deception: Simulating the Evolution from True News to Fake News with LLM Agents. arXiv preprint arXiv:2410.19064 (2024)

    2024 arXiv

  11. [19]

    In: 2024 20th IEEE International Colloquium on Signal Processing & Its Applications (CSPA), pp

    Teo, T.W., Chua, H.N., Jasser, M.B., Wong, R.T.: Integrating large language mod- els and machine learning for fake news detection. In: 2024 20th IEEE International Colloquium on Signal Processing & Its Applications (CSPA), pp. 102–107. IEEE (2024)

    2024

  12. [20]

    In: Proceedings of the IEEE/ACM 46th Interna- tional Conference on Software Engineering, pp

    Nam, D., Macvean, A., Hellendoorn, V., Vasilescu, B., Myers, B.: Using an llm to help with code understanding. In: Proceedings of the IEEE/ACM 46th Interna- tional Conference on Software Engineering, pp. 1–13. ACM (2024)

    2024

  13. [21]

    In: MILCOM 2024-2024 IEEE Military Communications Conference (MILCOM), pp

    Nathanson, S., Yoo, Y., Na, D., Cao, Y., Watkins, L.: A Step Towards Modern Disinformation Detection: Novel Methods for Detecting LLM-Generated Text. In: MILCOM 2024-2024 IEEE Military Communications Conference (MILCOM), pp. 615–620. IEEE (2024)

    2024

  14. [22]

    Future Gener- ation Computer Systems, 107877 (2025)

    Huang, H, Sun, N., Tani, M., Zhang, Y., Jiang, J., Jha, S.: Can LLM-generated misinformation be detected: A study on Cyber Threat Intelligence. Future Gener- ation Computer Systems, 107877 (2025)

    2025

  15. [23]

    Social Network Analysis and Mining15(1), 1–16 (2025)

    Chalehchaleh, R., Farahbakhsh, R., Crespi, N.: Addressing data scarcity in multi- lingual fake news detection: an LLM-based dataset augmentation approach. Social Network Analysis and Mining15(1), 1–16 (2025)

    2025

  16. [24]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Yu, X., Sheng, Z., Lu, W., Luo, X., Zhou, J.: Racmc: Residual-aware compensation network with multi-granularity constraints for fake news detection. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39(1), pp. 986–994. AAAI Press (2025) 16 W. Zhou et al

    2025

  17. [25]

    arXiv preprint arXiv:2003.04981 (2020)

    Zhou, X., Wu, J., Zafarani, R.: SAFE: similarity-aware multi-modal fake news detection. arXiv preprint arXiv:2003.04981 (2020)

    2003 arXiv

  18. [26]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Ying, Q., Hu, X., Zhou, Y., Qian, Z., Zeng, D., Ge, S.: Bootstrapping multi-view representations for fake news detection. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37(4), pp. 5384–5392. AAAI Press (2023)

    2023

  19. [27]

    arXiv preprint arXiv:2507.13415 (2025)

    Zhu, P., Jing, Y., Cheng, L., Chen, B., Cui, X., Wu, L., Tang, K.: SEER: Seman- tic Enhancement and Emotional Reasoning Network for Multimodal Fake News Detection. arXiv preprint arXiv:2507.13415 (2025)

    2025 arXiv

  20. [28]

    In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp

    Wang, Y., Ma, F., Jin, Z., Yuan, Y., Xun, G., Jha, K., Su, L., Gao, J.: Eann: Event adversarial neural networks for multi-modal fake news detection. In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 849–857. ACM (2018)

    2018

  21. [29]

    In: 2023 IEEE International Conference on Multimedia and Expo (ICME), pp

    Zhou, Y., Yang, Y., Ying, Q., Qian, Z., Zhang, X.: Multimodal fake news detection via clip-guided learning. In: 2023 IEEE International Conference on Multimedia and Expo (ICME), pp. 2825–2830. IEEE (2023)

    2023

  22. [30]

    In: Proceedings of the ACM on Web Conference 2025, pp

    Liu, Y., Liu, Y., Li, Z., Yao, R., Zhang, Y., Wang, D.: Modality interactive mixture- of-experts for fake news detection. In: Proceedings of the ACM on Web Conference 2025, pp. 5139–5150. ACM (2025)

    2025

  23. [31]

    arXiv preprint arXiv:2507.09647 (2025)

    Zhu, P., Jing, Y., Cheng, L., Tang, K., Guo, Y.: Ken: Knowledge augmentation and emotion guidance network for multimodal fake news detection. arXiv preprint arXiv:2507.09647 (2025)

    2025 arXiv

  24. [32]

    arXiv preprint arXiv:2506.00488 (2025)

    Hu, S., Hu, J., Zhang, H.: Synergizing LLMs with Global Label Propagation for Multimodal Fake News Detection. arXiv preprint arXiv:2506.00488 (2025)

    2025 arXiv

  25. [33]

    arXiv preprint arXiv:2405.01593 (2024)

    Li, X., Zhang, Y., Malthouse, E.C.: Large language model agent for fake news detection. arXiv preprint arXiv:2405.01593 (2024)

    2024 arXiv

  26. [34]

    In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pp

    He, K., Chen, X., Xie, S., Li, Y., Doll´ ar, P., Girshick, R.: Masked autoencoders are scalable vision learners. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pp. 16000–16009. IEEE (2022)

    2022

  27. [35]

    In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, vol

    Devlin, J., Chang, M.W., Lee, K., Toutanova, K.: Bert: Pre-training of deep bidi- rectional transformers for language understanding. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologi...

    2019

  28. [36]

    In: International Conference on Machine Learning, pp

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning, pp. 8748–8763. PMLR (2021)

    2021

  29. [37]

    arXiv preprint arXiv:2211.01335 (2022)

    Yang, A., Pan, J., Lin, J., Men, R., Zhang, Y., Zhou, J., Zhou, C.: Chinese clip: Con- trastive vision-language pretraining in chinese. arXiv preprint arXiv:2211.01335 (2022)

    2022 arXiv

  30. [38]

    arXiv preprint arXiv:1908.10084 (2019)

    Reimers, N., Gurevych, I.: Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. arXiv preprint arXiv:1908.10084 (2019)

    1908 arXiv

  31. [39]

    In: Proceedings of the AAAI Conference on Artificial Intel- ligence, vol

    Lu, W., Tong, Y., Ye, Z.: DAMMFND: Domain-Aware Multimodal Multi-view Fake News Detection. In: Proceedings of the AAAI Conference on Artificial Intel- ligence, vol. 39(1), pp. 559–567. AAAI Press (2025)

    2025

  32. [40]

    In: Proceedings of the 33rd ACM International Conference on Multimedia, pp

    Tong, Y., Lu, W., Cui, X., Mao, Y., Zhao, Z.: Dapt: Domain-aware prompt-tuning for multimodal fake news detection. In: Proceedings of the 33rd ACM International Conference on Multimedia, pp. 7902–7911. ACM (2025)

    2025

  33. [41]

    In: Proceedings of the 32nd ACM International Conference on Multimedia, pp

    Tong, Y., Lu, W., Zhao, Z., Lai, S., Shi, T.: MMDFND: Multi-modal multi-domain fake news detection. In: Proceedings of the 32nd ACM International Conference on Multimedia, pp. 1178–1186. ACM (2024)

    2024

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.