Pith. sign in

REVIEW 4 major objections 5 minor 57 references

Enhancing Hyperbole and Metaphor Detection with Their Bidirectional Dynamic Interaction and Emotion Knowledge

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that explicitly modeling emotion and the mutual influence between hyperbole and metaphor yields state-of-the-art detection on four standard datasets, with F1 gains of 28.1 points on TroFi and 23.1 points on HYPO-L over…

desk verdict Reasonable LLM-prompting pipeline for hyperbole/metaphor detection, but the headline SoTA gains are computed against a BERT-scale baseline and lack significance testing; still deserves referee time. read the letter →

arxiv 2506.15504 v1 pith:UAC5DHWR submitted 2025-06-18 cs.CL cs.LG

Li Zheng , Sihang Wang , Hao Fei , Zuquan Peng , Fei Li , Jianming Fu , Chong Teng , Donghong Ji This is my paper · ORCID
classification cs.CLcs.LG
keywords hyperboledetectionmetaphoremotion-guidedreasoningbidirectionaldynamicinteractiondomainmappinglargelanguagemodelsrhetoricaldevicesfigurative
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

This paper tries to establish that hyperbole and metaphor should not be detected in isolation: knowing one helps detect the other, and the emotion carried by a sentence is the key that unlocks both. It proposes EmoBi, a large-language-model pipeline that analyzes a sentence's emotion, maps the source and target domains of its figurative language, and then runs two interacting detection passes with a verification step. The reported result is that EmoBi beats every baseline on the HYPO, HYPO-L, LCC, and TroFi datasets, including a 28.1-point F1 gain on TroFi hyperbole detection and a 23.1-point gain on HYPO-L metaphor detection over the previous state of the art. A sympathetic reader would care because, if true, this means emotion and cross-device interaction are not frills but core signals for recognizing figurative language.

What carries the argument

The central mechanism is the bidirectional dynamic interaction module: two sequential LLM prompts in which the metaphor analysis (together with emotion and domain-mapping output) is passed into the hyperbole judgment, and the hyperbole analysis is passed into the metaphor judgment, so each device's detection supplies context for the other. The supporting parts are the emotion analysis prompt, the emotion-based domain mapping prompt that names source and target domains, and the verification mechanism that re-evaluates a detected device and keeps or revises it. What carries the argument is the prompt chain itself: no parameters are trained; the model is the sequence of prompts plus the LLM backend.

What would settle it

Run EmoBi on TroFi with the verification mechanism removed and the two directional passes combined by a pre-specified deterministic rule; if the F1 advantage over MTL-F-RoBERTa falls well below the reported 28.1-point gain, then the claimed bidirectional interaction plus verification is not what produces the result. A complementary check is to swap the order of the two passes or vary the LLM temperature; if F1 swings by more than a few points, the reported numbers are not stable.

Watch

Extended reading notes

Core claim

The paper's central claim is that hyperbole and metaphor detection are mutually supportive tasks and that emotion is the connective tissue that makes the support explicit. The authors propose EmoBi, an LLM-based pipeline that first extracts the emotion of a sentence, then prompts the LLM to identify the source and target domains of any figurative expression along with their emotional connection, and then runs two cross-guided detection passes: metaphor knowledge feeds the hyperbole judgment, and hyperbole knowledge feeds the metaphor judgment, followed by a verification step that asks the model to re-check each detected device. The experiments report higher precision, recall, and F1 than all baselines on all four datasets, with the largest margins being 28.1 F1 points on TroFi hyperbole detection and 23.1 F1 points on HYPO-L metaphor detection compared with the MTL-F-RoBERTa state of the art.

Load-bearing premise

The load-bearing premise is that the two directional judgments (metaphor-guided hyperbole and hyperbole-guided metaphor) can be merged into one correct final label and that the verification step reliably improves accuracy, but the paper does not specify the merge rule or the revision criterion, so if those choices are arbitrary the claimed benefit of bidirectional interaction is not demonstrated.

Editorial extensions

If this is right

  • If the results are correct, emotion analysis becomes a necessary first stage for rhetorical-device detection, and models that skip it will underperform on figurative language.
  • The reported gains imply that the previous multi-task state of the art left substantial signal unused; future work should compare against emotion-guided prompting rather than fine-tuned BERT-family models alone.
  • The bidirectional interaction principle suggests that joint detection frameworks should be built as two ordered passes rather than as a single shared-feature classifier, because each direction supplies different context.
  • The verification mechanism, once specified, could be reused as a post-hoc accuracy booster for other LLM classification tasks that produce two related labels.
  • The framework's reliance on an LLM means its performance is bounded by the backend model's emotional and conceptual reasoning; the paper's scale experiments indicate larger LLMs widen the gap over prompt-only baselines.

Reading between the lines

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

  • Beyond the paper: if the gains are reproducible, the likely explanation is not the bidirectional interaction per se but the explicit emotion and domain supervision; a testable prediction is that adding emotion-label supervision to the MTL-F fine-tuning objective would close much of the reported gap.
  • Beyond the paper: because the paper does not define how the two directional outputs are merged or when verification revises a label, an independent check should fix a deterministic merge rule and rerun; unless F1 stays within a few points, the bidirectional benefit is not stable.
  • Beyond the paper: the same prompt chain could be inverted into an explanation generator, producing emotion, source/target domains, and a rationale for each predicted rhetorical device, which would give downstream systems human-readable justifications.
  • Beyond the paper: a controlled study varying only the order of the two passes (hyperbole-first vs metaphor-first) would reveal whether the interaction is genuinely symmetric or whether one direction carries most of the benefit.
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 / 5 minor

Summary. The paper proposes EmoBi, an LLM-prompting framework for joint hyperbole and metaphor detection. The pipeline consists of three stages: (1) an LLM-based emotion analysis module that extracts emotion cues from the input sentence; (2) an emotion-based domain mapping module that prompts the LLM to identify source and target domains and their emotional connection; and (3) a bidirectional dynamic interaction module in which metaphor knowledge is used to guide hyperbole detection and hyperbole knowledge is used to guide metaphor detection, followed by a verification mechanism that re-evaluates the outputs. The system is evaluated on four datasets (HYPO, HYPO-L, LCC, TroFi) and compared with fine-tuned multi-task BERT-family baselines and with two LLM prompting baselines (standard prompting and vanilla chain-of-thought). The authors report large F1 improvements over MTL-F-RoBERTa, which they call the current SoTA, and smaller but consistent improvements over the chain-of-thought baseline, with ablations showing contributions from each module.

Significance. If the empirical claims hold, EmoBi would be a strong LLM-based approach to figurative language detection, and the idea of explicitly using emotion knowledge and bidirectional metaphor-hyperbole interaction is interesting and well motivated by linguistic observations. The paper also provides a useful case study showing how emotion and domain analysis can change predictions. However, the central claim of state-of-the-art performance is currently supported only by an inappropriate comparison class (fine-tuned BERT-scale models) while the comparison against the paper's own LLM baselines shows much smaller gains, and no statistical significance evidence is provided. The lack of a precise specification of the verification mechanism and of the bidirectional interaction aggregation further weakens the reproducibility of the results. The work is a reasonable incremental contribution to LLM prompting for rhetorical device detection, but the evidence as presented does not yet justify the headline SoTA claim.

major comments (4)
  1. [Section 4.2, Table 1, Abstract] The headline improvements of 28.1% (TroFi hyperbole) and 23.1% (HYPO-L metaphor) are absolute F1-point differences between EmoBi and MTL-F-RoBERTa, a fine-tuned BERT-scale model. Because EmoBi is an LLM-prompting framework, the appropriate comparison class is other LLM-based detectors. Against the paper's own CoT baseline, the gains are 5.7-7.7 F1 points, and against the prompt-based baseline they are similar in magnitude. The abstract and Section 4.3 therefore overstate the advantage by selecting a comparison model from a different paradigm. The authors should report the improvement over the best LLM baseline as the primary SoTA comparison, and should rephrase the abstract accordingly.
  2. [Section 3.5, Section 4.4] The verification mechanism is never specified. The text states only that 'If an error is detected in the identified hyperboles or metaphors, the model re-evaluate and adjust the results,' with no definition of how errors are detected, what prompt is used, how the revised judgment is produced, or how the revision is combined with the original prediction. Since the ablation in Table 2 attributes 1.3-1.9 F1 points to this component, the mechanism must be fully described (including the exact prompt template and the rule for accepting or rejecting a revision) for the ablation to be interpretable and for the method to be reproducible.
  3. [Section 3.5, Eq. (4)] The two passes of the 'bidirectional dynamic interaction' are described asymmetrically. Equation (4) gives the metaphor-guided hyperbole detection, but the reverse pass (hyperbole-guided metaphor detection) has no equation and no prompt template. It is also not stated whether both labels are produced in a single LLM call, whether the two passes are executed sequentially in one prompt or as separate calls, or how the outputs are aggregated. Without this information, the claimed benefit of bidirectional interaction is not precisely defined, and the comparison against the 'Together' condition in Figure 5 cannot be assessed.
  4. [Table 1, Table 2, Section 4.2] No standard deviations, number of runs, significance tests, or decoding parameters (e.g., temperature, top-p, number of samples) are reported for any experiment. The reported differences are as small as 1.2-1.9 F1 points in the ablations, which are likely within sampling noise for LLM prompting, especially for datasets of the sizes used here. The authors should report multiple runs with variance and apply an appropriate significance test (e.g., bootstrap or paired test) to support the claim that EmoBi 'significantly outperforms' the baselines.
minor comments (5)
  1. [Abstract, Section 4.3] The phrase 'increased by 28.1%' refers to an absolute difference in F1 points (84.2 vs. 56.1), not a relative percentage improvement. The text should say '28.1 percentage points' to avoid misleading readers.
  2. [Table 1] The bracketed values are absolute F1-point differences rather than relative improvements; the caption should state this explicitly.
  3. [Section 4.5, Figures 3-5] Figures 3-5 have garbled or missing axis labels and legends (the text appears as broken Unicode sequences). They need to be regenerated with readable fonts and clear legends before the paper can be reviewed or published.
  4. [Section 4.2, Related Work] The paper does not compare with recent LLM-based metaphor or hyperbole detection methods cited in Section 2 (e.g., Xu et al. 2024 on chain-of-thought for multimodal metaphor detection, and Yang et al. 2024 on ChatGPT for verb metaphor detection). Adding these or similar LLM baselines would strengthen the SoTA claim.
  5. [Section 5] There are several typographical errors, including 'we proposes' in the Conclusion and 'a impregnable' in the Figure 1 caption; the paper should be carefully proofread.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the proposed framework is an empirical LLM-prompting pipeline whose claims rest on direct comparisons and ablations, not on fitted parameters, self-referential definitions, or load-bearing self-citations.

full rationale

The paper's derivation chain is self-contained in the sense that no claimed result reduces by construction to an input or to a self-citation. The framework (EmoBi) is a sequence of LLM prompts: emotion analysis (Eq. 1), emotion-based domain mapping (Eq. 2), and bidirectional dynamic interaction (Eqs. 3-4) produce intermediate knowledge and final hyperbole/metaphor labels from the sentence plus previously generated knowledge. None of these equations defines a target label in terms of itself; each label is an additional LLM output conditioned on the sentence and on intermediate reasoning. The headline improvements (28.1% on TroFi hyperbole, 23.1% on HYPO-L metaphor) are empirical F1 differences reported in Table 1 against MTL-F-RoBERTa, a BERT-scale baseline; the paper's own LLM baselines (CoT-based) show smaller but still positive gaps. Whether MTL-F-RoBERTa is the appropriate 'current SoTA' comparison class is a fairness/correctness issue, not a circularity issue. The ablation study in Table 2 measures component contributions by removal; no parameter is fitted to test labels, so no prediction is statistically forced. Self-citations appear (e.g., Zheng et al. 2023b, 2024a, 2025) but only as background or application references; none is load-bearing for the central claim. The verification mechanism does use the LLM to re-evaluate its own outputs, but this is an empirical design choice whose effect is measured (removing it drops F1 by 1.2-1.9 points), not an algebraic reduction of the output to the input. No step meets the standard of exhibiting a specific circular reduction, so the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper introduces no mathematical free parameters or invented entities. Its assumptions are empirical hypotheses about language, emotion, and LLM behavior, each tested only on the four selected datasets. The prompt templates themselves are hand-designed but not numeric free parameters.

assumptions (4)
  • domain assumption Emotion analysis is a key factor in understanding rhetorical effects and improves detection accuracy.
    Section 3.3 asserts emotion is 'a key factor in facilitating the understanding of rhetorical effects'; this is a hypothesis tested only via ablation on four datasets.
  • domain assumption Identifying source and target domains from an emotion perspective helps detect hyperboles and metaphors.
    Section 3.4 assumes domain mapping 'constructs a crucial bridge' for understanding; no external evidence is provided beyond the ablated F1 deltas.
  • ad hoc to paper Hyperbole and metaphor mutually reinforce each other in detection.
    Section 3.5 postulates the bidirectional interaction; the paper provides no linguistic theory beyond two illustrative examples and an ablation.
  • domain assumption The LLM can reliably produce accurate emotion, domain, and verification outputs when prompted.
    The pipeline depends on LLM-generated intermediates; Section 6 admits error propagation, so this assumption can fail without a fallback.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Hyperbole and Metaphor Detection with Their Bidirectional Dynamic Interaction and Emotion Knowledge." pith.science (2026). https://pith.science/paper/UAC5DHWR

@misc{pith2026250615504,
  author       = {Pith},
  title        = {Pith review of: Enhancing Hyperbole and Metaphor Detection with Their Bidirectional Dynamic Interaction and Emotion Knowledge},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UAC5DHWR}},
  note         = {Machine review of arXiv:2506.15504}
}
read the original abstract

Text-based hyperbole and metaphor detection are of great significance for natural language processing (NLP) tasks. However, due to their semantic obscurity and expressive diversity, it is rather challenging to identify them. Existing methods mostly focus on superficial text features, ignoring the associations of hyperbole and metaphor as well as the effect of implicit emotion on perceiving these rhetorical devices. To implement these hypotheses, we propose an emotion-guided hyperbole and metaphor detection framework based on bidirectional dynamic interaction (EmoBi). Firstly, the emotion analysis module deeply mines the emotion connotations behind hyperbole and metaphor. Next, the emotion-based domain mapping module identifies the target and source domains to gain a deeper understanding of the implicit meanings of hyperbole and metaphor. Finally, the bidirectional dynamic interaction module enables the mutual promotion between hyperbole and metaphor. Meanwhile, a verification mechanism is designed to ensure detection accuracy and reliability. Experiments show that EmoBi outperforms all baseline methods on four datasets. Specifically, compared to the current SoTA, the F1 score increased by 28.1% for hyperbole detection on the TroFi dataset and 23.1% for metaphor detection on the HYPO-L dataset. These results, underpinned by in-depth analyses, underscore the effectiveness and potential of our approach for advancing hyperbole and metaphor detection.

Figures

Figures reproduced from arXiv: 2506.15504 by the authors.

Figure 1
Figure 1. Examples of Hyperbole and Metaphor Detec [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The overall architecture of our model. detection. The domain mapping module utilizes semantic and emotion information to identify target and source domains, aiding in understanding im￾plicit meanings. Finally, the bidirectional dynamic interaction mechanism enables mutual promotion in detection via knowledge transfer. 3.3 Emotion Analysis The first stage involves a comprehensive sentiment analysis of the given sente… view at source ↗
Figure 3
Figure 3. Comparison results of different LLM scales. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Comparison results of different LLMs. pared with the prompt-based method, our method exhibits a more significant enhancement in perfor￾mance when the model scale is enlarged. This indi￾cates that our method can deeply mine the semantic and emotional information within …
Figure 6
Figure 6. Figure 6: Case study to compare with prompt-based and [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 39 canonical work pages

  1. [1]

    URL: " 'urlintro :=

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year eprint doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRINGS urlintro eprinturl eprintpr...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Md Shad Akhtar, Ayush Kumar, Asif Ekbal, and Pushpak Bhattacharyya. 2016. A hybrid deep learning architecture for sentiment analysis. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 482--493

    work page 2016

  4. [4]

    Naveen Badathala, Abisek Rajakumar Kalarani, Tejpalsingh Siledar, and Pushpak Bhattacharyya. 2023. A match made in heaven: A multi-task framework for hyperbole and metaphor detection. arXiv preprint arXiv:2305.17480

    work page Pith review arXiv 2023

  5. [5]

    Maciej Besta, Nils Blach, Ales Kubicek, Robert Gerstenberger, Michal Podstawski, Lukas Gianinazzi, Joanna Gajda, Tomasz Lehmann, Hubert Niewiadomski, Piotr Nyczyk, et al. 2024. Graph of thoughts: Solving elaborate problems with large language models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 17682--17690

    2024

  6. [6]

    Julia Birke and Anoop Sarkar. 2006. A clustering approach for nearly unsupervised recognition of nonliteral language. In 11th Conference of the European chapter of the association for computational linguistics, pages 329--336

    2006

  7. [7]

    Christian Burgers, Elly A Konijn, and Gerard J Steen. 2016. Figurative framing: Shaping public discourse through metaphor, hyperbole, and irony. Communication theory, 26(4):410--430

    work page 2016

  8. [8]

    Robyn Carston and Catherine Wearing. 2011. Metaphor, hyperbole and simile: A pragmatic approach

    work page 2011

Show all 57 references
  1. [9]

    Xin Chen, Suge Wang, Xiaoli Li, Zhen Hai, Yang Li, Deyu Li, and Jianghui Cai. 2023. Identifying implicit emotions via hierarchical structure and rhetorical correlation. International Journal of Machine Learning and Cybernetics, 14(11):3753--3764

    2023

  2. [10]

    Verna Dankers, Marek Rei, Martha Lewis, and Ekaterina Shutova. 2019. Modelling the interplay of metaphor and emotion through multitask learning. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference...

    2019

  3. [11]

    Jacob Devlin. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805

    2018 arXiv

  4. [12]

    Yuzhe Ding, Kang He, Bobo Li, Li Zheng, Haijun He, Fei Li, Chong Teng, and Donghong Ji. 2025. Zero-shot conversational stance detection: Dataset and approaches. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL'25)

    2025

  5. [13]

    Vesna G Djokic, Ekaterina Shutova, and Rebecca Fiebrink. 2021. Metavr: Understanding metaphors in the mind and relation to emotion through immersive, spatial interaction. In Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems, pages 1--4

    2021

  6. [14]

    Mohamad Elzohbi and Richard Zhao. 2023. Contrastwsd: Enhancing metaphor detection with word sense disambiguation following the metaphor identification procedure. arXiv preprint arXiv:2309.03103

    2023 arXiv

  7. [15]

    Vithyatheri Govindan and Vimala Balakrishnan. 2022. A machine learning approach in analysing the effect of hyperboles using negative sentiment tweets for sarcasm detection. Journal of King Saud University-Computer and Information Sciences, 34(8):5110--5120

    2022

  8. [16]

    Kang He, Yuzhe Ding, Haining Wang, Fei Li, Chong Teng, and Donghong Ji. 2025. Dalr: Dual-level alignment learning for multimodal sentence representation learning. In Findings of the Association for Computational Linguistics: ACL 2025

    2025

  9. [17]

    Abisek Rajakumar Kalarani, Pushpak Bhattacharyya, and Sumit Shekhar. 2024. Unveiling the invisible: Captioning videos with metaphors. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 6306--6320

    2024

  10. [18]

    Z Lan. 2019. Albert: A lite bert for self-supervised learning of language representations. arXiv preprint arXiv:1909.11942

    2019 arXiv

  11. [19]

    Ain Lee, Juhyun Lee, Sooyeon Ahn, and Youngik Lee. 2023. Mindterior: A mental healthcare game with metaphoric gamespace and effective activities for mitigating mild emotional difficulties. In Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems, ...

    2023

  12. [20]

    Xiaonan Li, Kai Lv, Hang Yan, Tianyang Lin, Wei Zhu, Yuan Ni, Guotong Xie, Xiaoling Wang, and Xipeng Qiu. 2023. Unified demonstration retriever for in-context learning. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long P...

    2023

  13. [21]

    Xingwei Liang, Geng Tu, Jiachen Du, and Ruifeng Xu. 2024. Multi-modal attentive prompt learning for few-shot emotion recognition in conversations. Journal of Artificial Intelligence Research, 79:825--863

    2024

  14. [22]

    Bing Liu. 2020. Sentiment analysis: Mining opinions, sentiments, and emotions. Cambridge university press

    2020

  15. [23]

    Yinhan Liu. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692, 364

    2019 arXiv

  16. [24]

    Zhiyuan Ma, Zhihuan Yu, Jianjun Li, and Guohui Li. 2023. Hybridprompt: bridging language models and human priors in prompt tuning for visual question answering. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 13371--13379

    2023

  17. [25]

    Ben Mann, N Ryder, M Subbiah, J Kaplan, P Dhariwal, A Neelakantan, P Shyam, G Sastry, A Askell, S Agarwal, et al. 2020. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 1

    2020 arXiv

  18. [26]

    there's millions of them

    Michael McCarthy and Ronald Carter. 2004. “there's millions of them”: hyperbole in everyday conversation. Journal of pragmatics, 36(2):149--184

    2004

  19. [27]

    Saif Mohammad, Ekaterina Shutova, and Peter Turney. 2016. Metaphor as a medium for emotion: An empirical study. In Proceedings of the fifth joint conference on lexical and computational semantics, pages 23--33

    2016

  20. [28]

    Michael Mohler, Mary Brunson, Bryan Rink, and Marc Tomlinson. 2016. Introducing the lcc metaphor datasets. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pages 4221--4227

    2016

  21. [29]

    Yair Neuman, Dan Assaf, Yohai Cohen, Mark Last, Shlomo Argamon, Newton Howard, and Ophir Frieder. 2013. Metaphor identification in large texts corpora. PloS one, 8(4):e62343

    2013

  22. [30]

    Bo Pang, Lillian Lee, and Shivakumar Vaithyanathan. 2002. Thumbs up? sentiment classification using machine learning techniques. arXiv preprint cs/0205070

    2002 arXiv

  23. [31]

    Wenbo Qiao, Peng Zhang, and ZengLai Ma. 2024. A quantum-inspired matching network with linguistic theories for metaphor detection. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), page...

    2024

  24. [32]

    A Rakhlin. 2016. Convolutional neural networks for sentence classification. GitHub, 6:25

    2016

  25. [33]

    Azlaan Mustafa Samad, Kshitij Mishra, Mauajama Firdaus, and Asif Ekbal. 2022. Empathetic persuasion: reinforcing empathy and persuasiveness in dialogue systems. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 844--856

    2022

  26. [34]

    Nina Schneidermann, Daniel Hershcovich, and Bolette Sandford Pedersen. 2023. Probing for hyperbole in pre-trained language models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop), pages 200--211

    2023

  27. [35]

    Yuan Tian, Ruike Zhang, Nan Xu, and Wenji Mao. 2024. Bridging word-pair and token-level metaphor detection with explainable domain mining. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13311--13325

    2024

  28. [36]

    Yufei Tian, Nanyun Peng, et al. 2021. Hypogen: Hyperbole generation with commonsense and counterfactual knowledge. arXiv preprint arXiv:2109.05097

    2021 arXiv

  29. [37]

    Enrica Troiano, Carlo Strapparava, G \"o zde \"O zbal, and Serra Sinem Tekiro g lu. 2018. A computational exploration of exaggeration. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 3296--3304

    2018

  30. [38]

    Peter D Turney. 2002. Thumbs up or thumbs down? semantic orientation applied to unsupervised classification of reviews. arXiv preprint cs/0212032

    2002 arXiv

  31. [39]

    Yusong Wang, Dongyuan Li, Kotaro Funakoshi, and Manabu Okumura. 2023. Emp: Emotion-guided multi-modal fusion and contrastive learning for personality traits recognition. In Proceedings of the 2023 ACM International Conference on Multimedia Retrieval, pages 243--252

    2023

  32. [40]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824--24837

    2022

  33. [41]

    Shengqiong Wu, Hao Fei, Leigang Qu, Wei Ji, and Tat-Seng Chua. 2023. Next-gpt: Any-to-any multimodal llm. arXiv preprint arXiv:2309.05519

    2023 arXiv

  34. [42]

    Chenxing Xie, Yanding Wang, and Yang Cheng. 2024. Does artificial intelligence satisfy you? a meta-analysis of user gratification and user satisfaction with ai-powered chatbots. International Journal of Human--Computer Interaction, 40(3):613--623

    2024

  35. [43]

    Yanzhi Xu, Yueying Hua, Shichen Li, and Zhongqing Wang. 2024. Exploring chain-of-thought for multi-modal metaphor detection. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 91--101

    2024

  36. [44]

    Cheng Yang, Puli Chen, and Qingbao Huang. 2024. Can chatgpt’s performance be improved on verb metaphor detection tasks? bootstrapping and combining tacit knowledge. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Paper...

    2024

  37. [45]

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. 2024. Tree of thoughts: Deliberate problem solving with large language models. Advances in Neural Information Processing Systems, 36

    2024

  38. [46]

    Huixuan Zhang and Xiaojun Wan. 2023. Image matters: A new dataset and empirical study for multimodal hyperbole detection. arXiv preprint arXiv:2307.00209

    2023 arXiv

  39. [47]

    Huixuan Zhang and Xiaojun Wan. 2024. Image matters: A new dataset and empirical study for multimodal hyperbole detection. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 8652--8661

    2024

  40. [48]

    Linhao Zhang, Li Jin, Guangluan Xu, Xiaoyu Li, Cai Xu, Kaiwen Wei, Nayu Liu, and Haonan Liu. 2024 a . Camel: Capturing metaphorical alignment with context disentangling for multimodal emotion recognition. In Proceedings of the AAAI Conference on Artificial Intelligence, volume...

    2024

  41. [49]

    Linhao Zhang, Jintao Liu, Li Jin, Hao Wang, Kaiwen Wei, and Guangluan Xu. 2024 b . Gome: Grounding-based metaphor binding with conceptual elaboration for figurative language illustration. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing...

    2024

  42. [50]

    Yunxiang Zhang and Xiaojun Wan. 2021. Mover: Mask, over-generate and rank for hyperbole generation. arXiv preprint arXiv:2109.07726

    2021 arXiv

  43. [51]

    Li Zheng, Boyu Chen, Hao Fei, Fei Li, Shengqiong Wu, Lizi Liao, Donghong Ji, and Chong Teng. 2024 a . Self-adaptive fine-grained multi-modal data augmentation for semi-supervised muti-modal coreference resolution. In Proceedings of the 32nd ACM International Conference on Mult...

    2024

  44. [52]

    Li Zheng, Hao Fei, Ting Dai, Zuquan Peng, Fei Li, Huisheng Ma, Chong Teng, and Donghong Ji. 2025. Multi-granular multimodal clue fusion for meme understanding. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 26057--26065

    2025

  45. [53]

    Li Zheng, Hao Fei, Fei Li, Bobo Li, Lizi Liao, Donghong Ji, and Chong Teng. 2024 b . Reverse multi-choice dialogue commonsense inference with graph-of-thought. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19688--19696

    2024

  46. [54]

    Li Zheng, Donghong Ji, Fei Li, Hao Fei, Shengqiong Wu, Jingye Li, Bobo Li, and Chong Teng. 2023 a . Ecqed: emotion-cause quadruple extraction in dialogs. arXiv preprint arXiv:2306.03969

    2023 arXiv

  47. [55]

    Li Zheng, Fei Li, Yuyang Chai, Chong Teng, and Donghong Ji. 2023 b . A bi-directional multi-hop inference model for joint dialog sentiment classification and act recognition. In CCF International Conference on Natural Language Processing and Chinese Computing, pages 235--248

    2023

  48. [56]

    Qianrui Zhou, Hua Xu, Hao Li, Hanlei Zhang, Xiaohan Zhang, Yifan Wang, and Kai Gao. 2024. Token-level contrastive learning with modality-aware prompting for multimodal intent recognition. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 17114--17122

    2024

  49. [57]

    Kaijie Zhu, Qinlin Zhao, Hao Chen, Jindong Wang, and Xing Xie. 2024. Promptbench: A unified library for evaluation of large language models. Journal of Machine Learning Research, 25(254):1--22

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.