Pith. sign in

REVIEW 5 major objections 5 minor 35 references

Detecting and Understanding Hateful Contents in Memes Through Captioning and Visual Question-Answering

T0 review · 5 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A modular pipeline of OCR, neutral captioning, sub-label retrieval, and multi-turn visual question answering outperforms every compared baseline on the public Hateful Memes benchmark, hitting 73.50 percent accuracy and 78.35 percent AUROC.

desk verdict A plausible modular pipeline for hateful-meme detection reports a modest gain over GPT-4o mini on FHM, but the evidence is provisional because sub-label provenance and the curated expansion are unspecified, there are no error bars, and decontamination is unaddressed. read the letter →

arxiv 2504.16723 v1 pith:5FFMZGN7 submitted 2025-04-23 cs.CV cs.AI

Ali Anaissi , Junaid Akram , Kunal Chaturvedi , Ali Braytee This is my paper · ORCID
classification cs.CVcs.AI
keywords hatefulmemedetectionmultimodalclassificationvisualquestionansweringretrieval-augmentedgenerationopticalcharacterrecognitionimagecaptioningsub-labelhatespeech
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

Memes hide hate in the interaction between image and text, so the paper tries to detect that hate by sequencing several specialized steps rather than training one fused model. The steps are: extract the overlaid text with OCR, describe the image neutrally, classify the likely kind of hate, retrieve similar labeled examples, and ask the image targeted questions through visual question-answering. On the public Hateful Memes benchmark this full pipeline reaches 73.50 percent accuracy and 78.35 percent AUROC, better than every unimodal and multimodal baseline it is compared with, though still about eleven points below human annotators. The authors identify fine-grained sub-label retrieval and multi-turn visual question-answering as the ingredients that carry the gain, since removing either one drops performance substantially.

What carries the argument

The load-bearing mechanism is the paper's RAG (sub_label + VQA) pipeline. OCR and captioning first convert the meme into text and a neutral visual description; a sub-label classifier then partitions hateful content into categories such as race, religion, and other; a vector retrieval step pulls similar labeled examples for the matched sub-label; and a multi-turn visual question-answering loop supplies extra context to the final classifier. The sub-label partition is what makes retrieval precise, and the question-answering dialogue is what surfaces implicit cues that OCR and captioning alone miss. The paper's ablation comparison isolates these two components as the source of the performance gain.

What would settle it

Create a new set of memes that has never appeared online, label it under the same protocol, and run the full pipeline; if the accuracy gap over the strongest baseline shrinks from about four points to near zero, the reported gain is mostly memory of the public benchmark rather than the OCR-caption-retrieval-questioning mechanism.

Watch

Extended reading notes

Core claim

The central claim, stated on the paper's own terms, is that hateful memes are best detected by combining OCR-extracted text, a neutral caption of the image, retrieval of example memes matched to a fine-grained hate sub-label (such as race, religion, or other), and a multi-turn visual question-answering dialogue that probes symbols and context. On the public Hateful Memes dataset, this full pipeline--called RAG (sub_label + VQA)--achieves 73.50% accuracy and 78.35% AUROC, surpassing all compared methods. The sub-label-only variant reaches 72.00% accuracy and 76.52% AUROC, while replacing sub-labels with generic explanation retrieval drops the result to 59.20% accuracy and 63.01% AUROC, about the level of text-only classification. The paper attributes these results to the synergy between precisely targeted retrieval and VQA-derived context, and treats the remaining gap to 84.70% human accuracy as evidence of how much cultural and contextual nuance remains uncaptured.

Load-bearing premise

The reported edge depends on the commercial vision-language model used for visual question-answering having not memorized the public test memes during pretraining, and on the added curated examples not overlapping the test set.

Editorial extensions

If this is right

  • Separating text extraction, visual description, and reasoning can beat end-to-end fusion models on hateful-meme detection.
  • Retrieving example memes by fine-grained hate sub-labels adds accuracy, while retrieving by free-text explanations does not.
  • Multi-turn visual question answering supplies context that a single pass over OCR text and a caption misses.
  • Deployed at scale, the pipeline would need faster sub-label lookups and cheaper question-answering to meet latency budgets.

Reading between the lines

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

  • In my reading, the same sub-label-then-retrieve-then-question recipe could transfer to detecting sarcasm or conspiracy content, where hate is implicit rather than lexical.
  • A direct test of the paper's mechanism would be to swap the commercial vision-language model for an open-weight model of comparable size; if the four-point gain persists, the structure, not the specific model, is what matters.
  • The near-tie between explanation-based retrieval and text-only classification suggests that retrieval corpora for moderation should be organized by target category rather than by rationale text, a rule that may generalize beyond this benchmark.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes a modular pipeline for hateful meme detection that combines PaddleOCR for text extraction, neutral image captioning, sub-label classification (race, religion, other), retrieval-augmented generation (RAG) with a vector database, and a multi-turn visual question answering (VQA) module driven by GPT-4.0/GPT-3.5. The system is evaluated on the Facebook Hateful Memes (FHM) benchmark, where the best configuration (RAG with sub-labels plus VQA) is reported to achieve 73.50% accuracy and 78.35% AUROC, exceeding all compared baselines including GPT-4o mini (69.50% accuracy). The paper also includes qualitative examples of system outputs and a discussion of the contributions of each module.

Significance. If the reported results are reproducible and free of leakage, the framework would provide a useful demonstration that fine-grained retrieval and iterative VQA can push automated hateful-meme detection closer to human performance, while the modular design offers interpretability benefits. The paper is well motivated, and the ablation across RAG variants (explanation, sub-label, sub-label+VQA) is a reasonable way to isolate the contribution of each component. However, the central empirical claim is currently under-supported: there are no error bars or significance tests, the provenance of the 'carefully curated examples' and of the sub-labels is unspecified, no decontamination check is reported for the public benchmark and API-based models, and no code or data splits are provided. The reported gain over the strongest baseline is modest (about 4 accuracy points), so without these controls the claim that the method 'significantly outperforms all baselines' cannot be accepted.

major comments (5)
  1. [4.1] The sentence 'we expand the dataset with carefully curated examples reflecting these nuances' leaves the source, size, and selection criteria of these examples unspecified. Because the RAG vector database is populated from this expanded set, if any curated example overlaps with the test split, the retrieval step would leak target information and inflate the Table 1 results. Please disclose the provenance and number of curated examples and report an explicit overlap audit between the curated set, the RAG corpus, and the test split.
  2. [3.3 and Table 1] The sub-label categories (race, religion, other) and the procedure for assigning them to memes are never described. The FHM dataset provides only binary hateful/non-hateful labels, so sub-labeling is an added signal whose source is unclear. The paper must state whether sub-labels were manually annotated, predicted by a supervised model, or generated by an LLM, and whether the sub-label predictor for test memes used any test-set information. Without this, the RAG (sub_label) configuration is not reproducible and its reported 72.00% accuracy cannot be independently verified.
  3. [4.3, Table 1] The statement 'Our proposed method significantly outperforms all baselines' is unsupported by any statistical evidence. Table 1 reports point estimates only, with no standard deviations, confidence intervals, significance tests, or number of independent runs. This is particularly important because the VQA module involves stochastic LLM sampling. Please report means and variances over at least five runs for all method rows, and perform a paired significance test (e.g., bootstrap or McNemar) against the strongest baselines, including GPT-4o mini.
  4. [3.2 and 4.1] The evaluation uses GPT-4.0 and GPT-3.5 as the VQA reasoner on the public FHM test set, which is a well-known benchmark. The paper does not address the risk that these models may have memorized FHM examples during pretraining, which would make the reported gains reflect contamination rather than the proposed pipeline. Please report a decontamination check (e.g., exact and near-duplicate search of test memes against training data) or evaluate on a non-public or temporally separated test set.
  5. [4.2] The text says 'We conducted five rounds of scoring to mitigate model variability' but the five rounds are described only for VQAScore. It is unclear whether the classification results in Table 1 are obtained from a single run or from multiple runs, and how sampling temperature, seeds, and VQA output reuse were handled. Clarify the number of runs used for the accuracy and AUROC numbers and how variability across runs was aggregated.
minor comments (5)
  1. [References] References [24] and [25] both cite the same ViLBERT paper (Lu et al., 2019); please deduplicate and cite the appropriate version consistently.
  2. [3.3] There is a typo in the sentence ending 'content explanation..' with a double period; please fix.
  3. [Figure 1] Figure 1 is referenced with labels A, B, and C in Sections 3.1–3.3, but the figure panel labels are not explained in the caption; please add a clear caption identifying each panel.
  4. [3.3] The implementation details of the RAG system (embedding model, vector database, chunk size, number of retrieved chunks, prompt templates) are omitted; please provide these in an appendix or supplementary material to enable reproduction.
  5. [Figure 2] The qualitative example is helpful, but only two cases are shown; adding failure cases and an error analysis would strengthen the discussion of limitations.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the pipeline's reported gains are empirical benchmark comparisons, not derivations from fitted or self-cited inputs.

full rationale

The paper's central claim is an empirical one: the RAG (sub_label + VQA) pipeline achieves 73.50% accuracy and 78.35% AUROC on the Facebook Hateful Memes dataset, outperforming unimodal and multimodal baselines. None of the method's components is defined in terms of the target labels. OCR, captioning, sub-label retrieval, and VQA all produce intermediate representations that are then fed to a classifier; the final numbers are measured against held-out ground-truth labels rather than derived algebraically from any fitted constant. The RAG component retrieves labeled examples as context, which is standard in-context learning, not circular, unless the retrieved database contains the test items themselves, and the paper provides no evidence of that. The 'carefully curated examples' and the source of sub-labels are under-specified, raising a potential data-provenance and leakage concern, but leakage is an evaluation-validity issue, not a definitional circularity. There are self-citations (e.g., Refs. [4], [23], [30]) but none is load-bearing for the reported performance; the comparison baselines and dataset are external. No equation in the paper reduces to an input, and no fitted parameter is renamed as a prediction. Therefore, no circular step can be exhibited with the quoted reduction required by the rubric.

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

The pipeline rests on benchmark label quality, on the fidelity of OCR and captioning, and on the assumption that commercial LLMs have not memorized the public test set. It also relies on an unstated set of human curation choices for the data expansion and sub-label partitions. No theoretical axioms are introduced; the vulnerabilities are empirical and data-related.

free parameters (4)
  • Sub-label categories = race, religion, others (no annotation protocol)
    The set of sub-labels is chosen by hand in Section 3.3 to partition hateful content; the paper gives no criteria or validation for this partition.
  • Number of curated expansion examples = not reported
    Section 4.1 says the dataset is expanded with "carefully curated examples" but does not state how many, how selected, or whether they touch test splits.
  • Number of retrieved RAG chunks = not reported
    Section 3.3 describes a vector database query and ranking but never states the retrieval count or similarity threshold used in experiments.
  • VQA dialogue rounds = multi-turn, exact count not reported
    Section 3.2 says the system refines with follow-up questions until analysis is complete; the stopping criterion and number of turns are unspecified.
assumptions (4)
  • domain assumption FHM ground-truth labels are correct and consistent.
    The evaluation in Section 4.2 treats the Facebook Hateful Memes labels as ground truth; no relabeling or label-noise analysis is performed.
  • domain assumption GPT models used in VQA have not seen FHM test labels during pretraining.
    Section 3.2 uses GPT-4.0 and GPT-3.5; Section 4.1 uses a public benchmark. The paper never checks decontamination, and if the models memorized FHM, the performance comparison would be inflated.
  • domain assumption OCR and captioning capture all task-relevant meme content.
    Section 3.1 feeds only OCR text and an LLM-generated caption into RAG and VQA; images that rely on non-captioned visual detail are only covered by VQA, and caption quality is not evaluated.
  • ad hoc to paper Curated expansion examples generalize to the test distribution.
    Section 4.1 introduces "carefully curated examples" for sub-labels without describing selection rules; if these examples are close to test memes, the retrieval gain is partly a form of test-set leakage.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Detecting and Understanding Hateful Contents in Memes Through Captioning and Visual Question-Answering." pith.science (2026). https://pith.science/paper/5FFMZGN7

@misc{pith2026250416723,
  author       = {Pith},
  title        = {Pith review of: Detecting and Understanding Hateful Contents in Memes Through Captioning and Visual Question-Answering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5FFMZGN7}},
  note         = {Machine review of arXiv:2504.16723}
}
read the original abstract

Memes are widely used for humor and cultural commentary, but they are increasingly exploited to spread hateful content. Due to their multimodal nature, hateful memes often evade traditional text-only or image-only detection systems, particularly when they employ subtle or coded references. To address these challenges, we propose a multimodal hate detection framework that integrates key components: OCR to extract embedded text, captioning to describe visual content neutrally, sub-label classification for granular categorization of hateful content, RAG for contextually relevant retrieval, and VQA for iterative analysis of symbolic and contextual cues. This enables the framework to uncover latent signals that simpler pipelines fail to detect. Experimental results on the Facebook Hateful Memes dataset reveal that the proposed framework exceeds the performance of unimodal and conventional multimodal models in both accuracy and AUC-ROC.

Figures

Figures reproduced from arXiv: 2504.16723 by the authors.

Figure 1
Figure 1. The overall framework, RAG (sub_label + VQA) for detecting hateful [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Outputs for a hateful example (left) and a non-hateful example (right). [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 26 canonical work pages

  1. [1]

    IEEE Transactions on Consumer Electronics (June 2024)

    Aamir, M., Raut, R., Jhaveri, R.H., Akram, A.: Ai-generated content-as-a-service in iomt-based smart homes: Personalizing patient care with human digital twins. IEEE Transactions on Consumer Electronics (June 2024)

    work page 2024

  2. [2]

    arXiv preprint arXiv:2405.11215 (2024)

    Agarwal, S., Sharma, S., Nakov, P., Chakraborty, T.: MemeMQA: Multimodal Question Answering for Memes via Rationale-Based Inferencing. arXiv preprint arXiv:2405.11215 (2024)

    arXiv 2024

  3. [3]

    In: 2018 International Conference on Frontiers of Information Technology (FIT)

    Akram, J., Tahir, A.: Lexicon and heuristics based approach for identification of emotion in text. In: 2018 International Conference on Frontiers of Information Technology (FIT). pp. 293–297. IEEE (December 2018)

    work page 2018

  4. [4]

    In: 2024 IEEE International Conference on Data Mining Work- shops (ICDMW)

    Anaissi, A., Braytee, A., Akram, J.: Fine-tuning llms for reliable medical question- answering services. In: 2024 IEEE International Conference on Data Mining Work- shops (ICDMW). IEEE (December 2024)

    2024

  5. [5]

    In: Proceedings of the 26th International Conference on World Wide Web Companion (WWW ’17 Companion) (2017)

    Badjatiya, P., Gupta, S., Gupta, M., Varma, V.: Deep Learning for Hate Speech Detection in Tweets. In: Proceedings of the 26th International Conference on World Wide Web Companion (WWW ’17 Companion) (2017)

    work page 2017

  6. [6]

    arXiv preprint arXiv:2109.10649 (2021)

    Blaier, E., Malkiel, I., Wolf, L.: Caption Enriched Samples for Improving Hateful Memes Detection. arXiv preprint arXiv:2109.10649 (2021)

    arXiv 2021

  7. [7]

    In: Proceedings of the International AAAI Conference on Web and Social Media

    Davidson, T., Warmsley, D., Macy, M., Weber, I.: Automated Hate Speech Detec- tion and the Problem of Offensive Language. In: Proceedings of the International AAAI Conference on Web and Social Media. vol. 11 (2017)

    work page 2017

  8. [8]

    Devlin, J., Chang, M.W., Lee, K., Toutanova, K.: Bert: Pre-training of deep bidi- rectional transformers for language understanding (2019)

    2019

Show all 35 references
  1. [9]

    Du, Y., Li, C., Guo, R., Yin, X., Liu, W., Zhou, J., Bai, Y., Yu, Z., Yang, Y., Dang, Q., Wang, H.: Pp-ocr: A practical ultra lightweight ocr system (2020)

    2020

  2. [10]

    arXiv preprint arXiv:2005.04982 (2020) 12 A

    Gómez, R., Gibert, J., Gómez, L., Karatzas, D.: Exploring Hate Speech Detection in Multimodal Publications. arXiv preprint arXiv:2005.04982 (2020) 12 A. Anaissi et al

    2020 arXiv

  3. [11]

    ACM Transactions on Asian and Low-Resource Language Information Processing (2023)

    Hamza, A., Javed, A.R., Iqbal, F., Yasin, A., Srivastava, G., Połap, D., ..., Jalil, Z.: Multimodal Religiously Hateful Social Media Memes Classification based on Textual and Image Data. ACM Transactions on Asian and Low-Resource Language Information Processing (2023)

    2023

  4. [12]

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition (2015)

    2015

  5. [13]

    In: Proceedings of the ACM Web Conference 2022

    Hee, M.S., Lee, R.K.W., Chong, W.H.: On explaining multimodal hateful meme detection models. In: Proceedings of the ACM Web Conference 2022. p. 3651–3655. WWW ’22, Association for Computing Machinery, New York, NY, USA (2022)

    2022

  6. [14]

    In: IEEE International Conference on Computer Vi- sion (ICCV)

    Howard, A., Sandler, M., Chen, B., Wang, W., Chen, L.C., Tan, M., ..., Le, Q.: Searching for MobileNetV3. In: IEEE International Conference on Computer Vi- sion (ICCV). pp. 1314–1324 (2019)

    2019

  7. [15]

    Information Fusion74, 50–64 (October 2021)

    Khan, M.T.R., Saad, M.M., Tariq, M.A., Kim, D.: Spice-it: Smart covid-19 pan- demic controlled eradication over ndn-iot. Information Fusion74, 50–64 (October 2021)

    2021

  8. [16]

    arXiv preprint arXiv:2005.04790 (2021)

    Kiela, D., Firooz, H., Mohan, A., Goswami, V., Singh, A., Ringshia, P., Testuggine, D.: The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes. arXiv preprint arXiv:2005.04790 (2021)

    2021 arXiv

  9. [17]

    arXiv preprint arXiv:2107.04313 (2021)

    Kirk, H.R., Jun, Y., Rauba, P., Wachtel, G., Li, R., Bai, X., ..., Asano, Y.M.: Memes in the Wild: Assessing the Generalizability of the Hateful Memes Challenge Dataset. arXiv preprint arXiv:2107.04313 (2021)

    2021 arXiv

  10. [18]

    In: Proceedings of the 2021 Workshop on Online Abuse and Harms (WOAH 2021) (2021)

    Kougia, V., Pavlopoulos, J.: Multimodal or Text? Retrieval or BERT? Benchmark- ing Classifiers for the Shared Task on Hateful Memes. In: Proceedings of the 2021 Workshop on Online Abuse and Harms (WOAH 2021) (2021)

    2021

  11. [19]

    SN Computer Science 2(2), 95 (Apr 2021)

    Kovács, G., Alonso, P., Saini, R.: Challenges of Hate Speech Detection in Social Media: Data Scarcity, and Leveraging External Resources. SN Computer Science 2(2), 95 (Apr 2021)

    2021

  12. [20]

    Neural Information Processing Systems (NeurIPS) paper (2020)

    Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., ..., Kiela, D.: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Neural Information Processing Systems (NeurIPS) paper (2020)

    2020

  13. [21]

    Li, L.H., Yatskar, M., Yin, D., Hsieh, C.J., Chang, K.W.: Visualbert: A simple and performant baseline for vision and language (2019)

    2019

  14. [22]

    arXiv preprint arXiv:2404.01291 (2024)

    Lin, Z., Pathak, D., Li, B., Li, J., Xia, X., Neubig, G., ..., Ramanan, D.: Eval- uating Text-to-Visual Generation with Image-to-Text Generation. arXiv preprint arXiv:2404.01291 (2024)

    2024 arXiv

  15. [23]

    In: Com- panion Proceedings of the ACM Web Conference 2024

    Liu, Z., Braytee, A., Anaissi, A., Zhang, G., Qin, L.: Ensemble pretrained models for multimodal sentiment analysis using textual and video data fusion. In: Com- panion Proceedings of the ACM Web Conference 2024. pp. 1841–1848 (2024)

    2024

  16. [24]

    Neural Information Processing Systems (NeurIPS) paper (2019)

    Lu, J., Batra, D., Parikh, D., Lee, S.: ViLBERT: Pretraining Task-Agnostic Vi- siolinguistic Representations for Vision-and-Language Tasks. Neural Information Processing Systems (NeurIPS) paper (2019)

    2019

  17. [25]

    Lu, J., Batra, D., Parikh, D., Lee, S.: Vilbert: Pretraining task-agnostic visiolin- guistic representations for vision-and-language tasks (2019)

    2019

  18. [26]

    Meta AI Research: Hateful Memes Challenge and dataset.https://ai.meta.com/ blog/hateful-memes-challenge-and-data-set/ , accessed: 2024-11-23

    2024

  19. [27]

    In: Proceedings of ConIT 2024

    Munzni, S., Dixit, S., Bhat, A.: Classification of Hateful Memes by Multimodal Analysis using CLIP. In: Proceedings of ConIT 2024. pp. 1–5 (2024)

    2024

  20. [28]

    https://openai

    OpenAI: Gpt-4o: Openai’s most advanced multimodal model. https://openai. com/index/gpt-4o (2024), accessed: 2024-10-03 Detecting and Understanding Hateful Contents in Memes 13

    2024

  21. [29]

    In: Zong, C., Xia, F., Li, W., Navigli, R

    Pramanick, S., Dimitrov, D., Mukherjee, R., Sharma, S., Akhtar, M.S., Nakov, P., Chakraborty, T.: Detecting harmful memes and their targets. In: Zong, C., Xia, F., Li, W., Navigli, R. (eds.) Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021. pp. 2783–2...

    2021

  22. [30]

    In: 2024 IEEE International Conference on Data Mining Workshops (ICDMW)

    Qian, C., Shi, X., Yao, S., Liu, Y., Zhou, F., Zhang, Z.: Optimized biomedical question-answering services with llm and multi-bert integration. In: 2024 IEEE International Conference on Data Mining Workshops (ICDMW). IEEE (December 2024)

    2024

  23. [31]

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., ..., Sutskever, I.: Learning Transferable Visual Models From Natural Language Su- pervision (2021)

    2021

  24. [32]

    IEEE Transactions on Consumer Electronics (April 2024)

    Rathore, R.S., Jhaveri, R.H., Akram, A.: Galtrust: Generative adversarial learning- based framework for trust management in spatial crowdsourcing drone services. IEEE Transactions on Consumer Electronics (April 2024)

    2024

  25. [33]

    In: 2018 24th International Conference on Au- tomation and Computing (ICAC)

    Rehman, A.U., Rehman, Z., Ali, W., Shah, M.A., Salman, M.: Statistical topic modeling for urdu text articles. In: 2018 24th International Conference on Au- tomation and Computing (ICAC). pp. 1–6. IEEE (September 2018)

    2018

  26. [34]

    Information Processing and Management 60(5), 103474–103474 (2023)

    Rizzi, G., Gasparini, F., Saibene, A., Rosso, P., Fersini, E.: Recognizing misog- ynous memes: Biased models and tricky archetypes. Information Processing and Management 60(5), 103474–103474 (2023)

    2023

  27. [35]

    In: Gangemi, A., Navigli, R., Vidal, M.E., Hitzler, P., Troncy, R., Hollink, L., Tordai, A., Alam, M

    Zhang, Z., Robinson, D., Tepper, J.: Detecting hate speech on twitter using a convolution-gru based deep neural network. In: Gangemi, A., Navigli, R., Vidal, M.E., Hitzler, P., Troncy, R., Hollink, L., Tordai, A., Alam, M. (eds.) The Semantic Web. pp. 745–760. Springer Interna...

    2018

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.