Pith. sign in

REVIEW 1 major objections 6 minor 21 references

Impromptu Cybercrime Euphemism Detection

T0 review · 1 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A context-based detector can pick out drug euphemisms that never appeared in its training data, reporting a 76-fold precision improvement over the prior best masked-language-model detector.

desk verdict New dataset and task framing, but the 'impromptu' claim is untested—held-out DEA slang is not a novel-euphemism test, and the evaluation metrics are too ambiguous to back the headline numbers. read the letter →

arxiv 2412.01413 v2 pith:GGTTV5SP submitted 2024-12-02 cs.CL

Xiang Li , Yucheng Zhou , Laiping Zhao , Jing Li , Fangming Liu This is my paper · ORCID
classification cs.CL
keywords impromptueuphemismdetectioncybercrimeeuphemismsdrugslangmaskedlanguagemodelingcontextaugmentationmulti-rounditerativetrainingICEDdatasetcontentmoderation
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 'impromptu' cybercrime euphemisms—drug slang terms that are newly coined, undocumented, and rare—can be detected in a large unlabeled corpus by context, even when the model never sees the specific term in training. To test this, the authors build the ICED dataset from 440 manually verified sentences containing 22 drug euphemisms from the US Drug Enforcement Administration's slang list, drawn from Reddit forums, then hide those euphemisms from training. Their CAMIT pipeline first uses a coarse classifier to discard most harmless sentences, then a fine-grained masked-language-model detector that learns to predict a masked token's drug-related meaning using context augmentation and multi-round iterative training. On 90 million tokens with only 440 positives, they report Top@20 precision of 4.61 per thousand versus 0.06 per thousand for the best previous MLM detector, a 76-fold improvement. If the claim holds, content moderation could flag novel drug slang without waiting for it to appear on official lists.

What carries the argument

The load-bearing mechanism is masked-token prediction with a context augmentation head. Candidate words are first retrieved by Word2Vec cosine similarity to seed drug words; sentences containing those candidates are masked and fed to a binary classifier that gates them as euphemism-related or not. The surviving sentences go to a fine-grained model that jointly optimizes two losses: a standard masked-language-modeling loss on a single masked token, and a context-augmentation loss in which 50% of tokens are masked and a two-layer augmentation head reconstructs the sentence before the MLM head reads the refined representations. Multi-round iterative training re-filters the training corpus after each round, using the model itself to remove noisy sentences. The design's purpose is to force the model to infer semantics from context rather than from any remembered vocabulary, which is what the authors identify as the key difference between detecting common versus impromptu euphemisms.

What would settle it

A direct test would coin a set of genuinely new drug euphemisms (for example, 'blueberry kus' variants) that are absent from any public list and have near-zero corpus frequency, inject them into a forum-like corpus, and check whether CAMIT ranks them at the top; if it does not, the 76-fold result may reflect the model's familiarity with DEA-listed terms rather than an ability to detect impromptu coinages.

Watch

Extended reading notes

Core claim

The paper's central claim is that a masked language model can identify held-out drug euphemisms in raw forum text if it is trained to reconstruct masked tokens from heavily masked context. The authors argue that previous euphemism detectors fail on impromptu euphemisms because they match words against known slang lists or rely on distributional differences between black and white corpora, both of which presuppose prior exposure. CAMIT avoids that by treating euphemism detection as a two-stage ranking problem: a Word2Vec seed-expansion identifies candidate drug-related words, a coarse Transformer classifier removes most harmless sentences, and a fine-grained BERT-style detector assigns each masked candidate a probability of being a euphemism while a context augmentation head encourages the model to use surrounding context rather than surface form. The reported result is that the 22 held-out euphemisms rank near the top of the model's predictions, lifting precision from near zero to 4.61 per thousand at Top@20 and recall from 0.10 to 0.53 percent. The paper's interpretation is that contextual semantic inference, not memorization, is what makes impromptu euphemism detection possible.

Load-bearing premise

The dataset treats 22 known drug slang terms from the DEA list as 'impromptu' simply by hiding them from training, and the paper assumes that these held-out known euphemisms behave like newly coined, undocumented ones; if that equivalence fails, the reported gains measure zero-shot recognition of familiar slang rather than detection of novel euphemisms.

Editorial extensions

If this is right

  • A moderation pipeline built on CAMIT could flag newly coined drug slang on public forums without maintaining an up-to-date list of slang terms.
  • The ICED benchmark turns impromptu euphemism detection into a measurable task, giving future systems a common ground for comparison.
  • The 76-fold precision gain is specifically at Top@20 on word-level euphemisms; phrase-level detection and non-drug cybercrime categories remain untested by the paper.
  • Backbone choice matters: BERT-base outperforms T5-small at every threshold, suggesting bidirectional context is important for this task.
  • The ChatGPT-generated development set offers a way to stop training without using the ICED labels, which keeps the held-out setting intact.

Reading between the lines

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

  • If held-out known euphemisms truly behave like impromptu ones, the same recipe may transfer to other low-frequency slang domains, such as fraud or weapons terminology, wherever a small seed list is available.
  • A stricter validation would coin genuinely new euphemisms—variants like 'strawberry kush' that are absent from every public list—and check whether the model ranks them; the paper does not run this test.
  • At Top@20 precision of 4.61 per thousand, most flagged tokens are still false positives, so a practical deployment would need a separate threshold or a human-review stage before any moderation action.
  • Because all positive sentences come from five Reddit subforums, transfer to other platforms and registers is open; a test on Telegram or TikTok text would show whether the context signal generalizes.
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

1 major / 6 minor

Summary. The paper introduces the Impromptu Cybercrime Euphemism Detection (ICED) dataset and a detection framework called CAMIT, which combines coarse-grained filtering, fine-grained masked-language-model scoring with context augmentation, and multi-round iterative training. The authors report a 76-fold precision improvement over a prior euphemism detector and claim the first exploration of impromptu euphemism detection. The evaluation holds out 22 DEA-listed drug euphemisms from the training corpus and measures precision and recall at different Top@k thresholds.

Significance. If the central claim were supported, the paper would address an important and under-studied problem: detecting newly coined, low-frequency euphemisms in social media text. The proposed two-stage framework and the use of context augmentation are reasonable design ideas, and the paper reports comparisons against several baseline methods. The authors also include a Limitations section that acknowledges single-forum and drug-only scope. However, the significance is currently not established because the dataset does not actually contain impromptu euphemisms as defined in the paper, the evaluation metrics are ambiguous, and there is a potential leakage path in the candidate-selection procedure.

major comments (1)
  1. [Section 5.1] The use of a ChatGPT-generated development set as the training-stopping criterion is not validated. The paper introduces a separate dev set of 132 samples produced by ChatGPT, but it does not show that performance on this generated set correlates with performance on the ICED target set. If the stopping criterion is poorly calibrated, the reported benefits of multi-round iterative training (Figure 8) may reflect overfitting to the ChatGPT-generated distribution rather than to the task of detecting euphemisms. At minimum, the paper should report the correlation between the ChatGPT dev-set metric and the ICED evaluation metric, or justify the criterion with a separate validation study.
minor comments (6)
  1. [Abstract] The abstract contains a typo: 'the mode's capability' should be 'the model's capability.'
  2. [Introduction] The introduction says 'Mult-round Iterative Training' in the CAMIT acronym definition; this should be 'Multi-round.'
  3. [Conclusion] The conclusion begins 'We mask a first exploration'; this should be 'We make a first exploration.'
  4. [Table 1] The table caption uses 'Precion (‰)' and 'Recall' but the 'Precion' is a misspelling of 'Precision.'
  5. [Figure 6] The boxplot in Figure 6 lacks axis labels and a legend explaining what 'Rank' refers to, and the caption does not describe how the ranks are computed; this makes the figure difficult to interpret.
  6. [References] Table 3 is presented as 'Euphemisms confirmed during manual checks,' but its contents are full sentences rather than a table of euphemisms; the formatting and caption should be revised to make the samples and their relation to the target euphemisms clear.

Circularity Check

2 steps flagged · score 3.0 of 10

No equation-level circularity, but the 'impromptu' claim is partly definitional: the 22 test targets are pre-selected DEA-listed slang identified as 'impromptu euphemisms' by dataset construction, and the detector's Word2Vec candidate pool is built with the same corpus and similarity criterion used to select those targets.

  1. self definitional [Section 3.2 (ICED Construction Pipeline); Section 5.1 (Evaluation Metrics, Eqs. 9-10).]
    "From the DEA inverted index, we select twenty-two drug categories, each yielding one cybercrime euphemism. This choice includes fifteen single-word euphemisms and seven two-word combinations. ... These 440 sentences, with each sentence corresponding to a target token, collectively constitute “Target Corpus”, which serves as the repository for impromptu euphemisms in our dataset."

    The introduction defines impromptu euphemisms as 'previously undocumented, highly time-sensitive' terms, but Section 3.2 operationalizes them as 22 items from the DEA's 2018 slang document (documented, static). Section 5.1 then defines the evaluation target as nimp = 'the number of impromptu cybercrime euphemisms in ICED' — the 440 sentences built from those 22 pre-selected terms. The property 'impromptu' and the property 'test target' are therefore identified by construction: a token scores only if it is one of the 22 pre-selected DEA terms, and the paper's central claims ('first exploration of impromptu euphemism detection'; detection of impromptu euphemisms) read that property off the dataset definition.

  2. fitted input called prediction [Section 3.2 (dataset target selection) versus Sections 4.2-4.3 (detector candidate generation).]
    "The most frequently used drug name is selected as the seed word to query the 50 words with the closest word vector cosine similarity ... Then, we use the DEA vocabulary to find the intersection with this vocabulary ... To train a fine-grained classification model, we employed the ICED dataset to train a Word2Vec model. This allowed us to identify the top 1000 words with the highest cosine similarity to the mean word vector of the seed words."

    The same Word2Vec similarity trained on the same corpus drives both target selection and detector input. In Section 3.2 the 22 target euphemisms are pre-selected as the intersection of the DEA vocabulary with the Word2Vec neighbourhood of drug-name seed words. In Sections 4.2-4.3 the detector's candidate pool (top 100, then top 1000 words most similar to the seed vectors) is generated by the identical recipe on the identical corpus, and the model can only report tokens passing this Word2Vec-derived filter. The targets were thus chosen by exactly the similarity signal the pipeline uses to propose candidates and to build its training corpus.

full rationale

The held-out protocol is genuine: Section 3.2 removes all sentences containing the 22 target euphemisms from the training corpus, and Section 4.1 states the labels 'are only utilized for evaluation metrics rather than provide supervision signals during model training.' Evaluation therefore measures zero-shot ranking of 440 held-out sentences; no test label enters training, so there is no fitted-input-called-prediction reduction in the usual sense. The only self-citation (Li and Zhou 2023, Section 2.2) is a passing reference in 'Unlike feature augmentation (Li and Zhou, 2023)' and is not load-bearing. What keeps the score above 2: first, the central construct is defined by the dataset — 'impromptu euphemisms' are, by construction, the 22 DEA-listed terms held out, while the introduction defines impromptu euphemisms as previously undocumented and highly time-sensitive; the identity is asserted (Section 3.1 'mimicking impromptu euphemism usage') and never tested, and the Limitations section flags single-forum and drug-only scope but not this gap. Second, the Word2Vec model that pre-filters the target set during dataset construction is re-fitted to the same corpus to generate the detector's candidate pool, so the detection pipeline is aligned with target selection by construction, though the alignment is not by itself decisive (embedding-based Cant Reader scores 0.00). Third, the headline '76-fold improvement' (Section 5.2) is a ratio over a near-zero baseline — MLM precision 0.06 per mille to CAMIT 4.61 per mille at Top@20 — so the relative gain is inflated while absolute precision remains below 0.5%. These are overclaim/construct-validity concerns as much as circularity; the empirical comparison itself is self-contained and non-trivial, warranting a moderate score of 3 rather than 6+.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

The method is empirical and does not derive parameters from first principles. Many design choices are set by hand (masking rate, list sizes, iteration count), and the central evaluation relies on assumptions about semantic similarity and transfer. No new physical or conceptual entities are introduced.

free parameters (6)
  • number of selected euphemisms per drug category = 22
    Authors select 22 euphemisms (15 words, 7 phrases) from the DEA list to define the target corpus; this choice determines the test label set.
  • top-k for coarse candidate words = 100
    Section 4.2: the top 100 words most similar to each seed word are used to build positive samples for the coarse classifier.
  • top-k for fine-grained candidate words = 1000
    Section 4.3: the top 1000 words most similar to the mean seed vector are used to build the fine-grained training corpus.
  • CAM masking rate = 50%
    Section 4.3: context augmentation masks 50% of tokens, an ad hoc choice.
  • number of iterative training rounds = 2
    Section 4.4 and Figure 8: authors stop at two iterations based on observed diminishing returns; this is chosen post hoc.
  • seed word = most frequent drug name (not named)
    Section 3.2: the most frequently used drug name is used as seed to query Word2Vec neighbors, but the specific word is not reported.
assumptions (3)
  • domain assumption Word2Vec cosine similarity is a valid proxy for semantic relatedness of euphemisms
    Used in Sections 3.2 and 4.2 to select candidate words and sentences; no validation of this proxy for the euphemism detection task.
  • domain assumption Masked language modeling on drug-related sentences transfers to held-out euphemisms
    The fine-grained detector assumes BERT fine-tuned on masked sentences with similar words will generalize to unseen euphemism terms.
  • ad hoc to paper ChatGPT-generated dev set is a valid stopping criterion
    Section 4.5: the development set is generated by ChatGPT rather than human annotation, and its distribution may not match the test corpus.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Impromptu Cybercrime Euphemism Detection." pith.science (2026). https://pith.science/paper/GGTTV5SP

@misc{pith2026241201413,
  author       = {Pith},
  title        = {Pith review of: Impromptu Cybercrime Euphemism Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GGTTV5SP}},
  note         = {Machine review of arXiv:2412.01413}
}
read the original abstract

Detecting euphemisms is essential for content security on various social media platforms, but existing methods designed for detecting euphemisms are ineffective in impromptu euphemisms. In this work, we make a first attempt to an exploration of impromptu euphemism detection and introduce the Impromptu Cybercrime Euphemisms Detection (ICED) dataset. Moreover, we propose a detection framework tailored to this problem, which employs context augmentation modeling and multi-round iterative training. Our detection framework mainly consists of a coarse-grained and a fine-grained classification model. The coarse-grained classification model removes most of the harmless content in the corpus to be detected. The fine-grained model, impromptu euphemisms detector, integrates context augmentation and multi-round iterations training to better predicts the actual meaning of a masked token. In addition, we leverage ChatGPT to evaluate the mode's capability. Experimental results demonstrate that our approach achieves a remarkable 76-fold improvement compared to the previous state-of-the-art euphemism detector.

Figures

Figures reproduced from arXiv: 2412.01413 by the authors.

Figure 1
Figure 1. The Construction Pipeline for ICED dataset. social media, e.g., “coke” refers to actual cola. 3.2 ICED Construction Pipeline The design of our dataset comprises three distinct components: “Target Corpus”, containing exclu￾sively manually verified impromptu cybercrime euphemisms; “Deduplication Corpus”, which in￾cludes common cybercrime euphemisms and cy￾bercrime euphemisms with normal meaning; and “White Corpus”, en… view at source ↗
Figure 2
Figure 2. The training and inference pipeline of our method. Training flow is represented by the blue line. Inference flow is denoted by the orange line. a word list. Then, we use the DEA vocabulary to find the intersection with this vocabulary to narrow the scope and facilitate the search for cybercrime euphemisms. After obtaining such a simplified DEA vocabulary, construct an inverted index and obtain the index of the sente… view at source ↗
Figure 3
Figure 3. Training for fine-grained classification entails two components: mask language modeling (MLM), represented by the gray line, and context augmentation modeling (CAM), as denoted by the blue line. of euphemism in sentence. The loss function is defined as follows: Lcg = − X i yi log(p) (3) where yi is the sentence label. Finally, based on the classification outcomes predicted by the model Mclf , we filter the dataset D… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Prompt of generating euphemisms based on the seeds. Drug name: amphetamine Drug sentence: way to fucking much man dont_ask for help when its clear the reason_why man you_should have_known that wasnt good for you mdma has enough amphetamine in it why_would you_wanna tak…
Figure 5
Figure 5. Figure 5: Prompt of creating samples, both with and without euphemisms. evaluation, comprising 132 samples, was created with the assistance of ChatGPT. It encompasses 66 positive samples, each containing a euphemism, and 66 negative samples, each incorporating a word or sentence…
Figure 6
Figure 6. Figure 6: Boxplot of prediction ranks w/o CAM w/ CAM [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Precision and recall variation caused by removing context augmentation modeling 5.2 Main Results The detection results of the existing methods and CAMIT for the word impromptu cybercrime eu￾phemisms in the ICED dataset are shown in Ta￾ble 1. Form the table, the existin…
Figure 8
Figure 8. Figure 8: Precision and recall variation caused by different rounds of iterative training [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 20 canonical work pages

  1. [2]

    In Proceedings of the 26th International Conference on World Wide Web (WWW), WWW ’17, page 657–666, Republic and Canton of Geneva, CHE

    Tools for automated analysis of cybercriminal markets. In Proceedings of the 26th International Conference on World Wide Web (WWW), WWW ’17, page 657–666, Republic and Canton of Geneva, CHE. International World Wide Web Conferences Steering Committee. Jherez Taylor, Melvyn Peignon, and Yi-Shin Chen

    work page

  2. [3]

    ArXiv, abs/1711.10093

    Surfacing contextual hate speech words within social media. ArXiv, abs/1711.10093. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Pro- cessing Systems 30: Annual Conference on Neural Information Processing Syste...

    arXiv 2017

  3. [6]

    Great Hormones at Bedtime 7. Goop 8. Gamma-OH 9. G-Juice 10. G-Riffick 11. Gook 12. G-Spot 13. GHB 14. Gammagoblin 15. G-Force

    work page

  4. [7]

    School buses 9

    Upjohns 8. School buses 9. Handlebars 10. Planks 11. Bricks 12. Alpraz 13. Footballs 14. White boys 15. White girls 16. Bicycle parts

    work page

  5. [8]

    XTC 15. Adam 16. Eve 17. Scooby snacks 18. Thizz 19. Eskimo 20. White dove GHB 1. G 2. Liquid E 3. Grievous Bodily Harm 4. Georgia Home Boy 5. Gina

    work page

  6. [10]

    Chiva 15. Boy 16. Girl 17. Mud 18. Poppy 19. Diesel 20. Snow Hydrocodone 1. Vikes 2. Norcos 3. Hydros 4. Watsons 5. Lorris 6. Tabs 7. 357s 8. Bananas 9. Fluff 10. Scratch 11. Juice 12. Hydro 13. Vicos 14. Whiteys

    work page

  7. [12]

    Huachuma 9

    San Pedro 8. Huachuma 9. Moon 10. God’s flesh 11. Divine cactus 12. Sacred cactus 13. Visionary cactus 14. Shamanic medicine

    work page

  8. [13]

    Mind-expanding cactus 17

    Psychedelic cactus 16. Mind-expanding cactus 17. Spirit plant 18. Teacher plant 19. Wisdom plant 20. Hallucinogenic cactus Methamphetamine 1. Meth 2. Crystal 3. Ice 4. Tina 5. Crank 6. Glass 7. Speed 8. Go fast 9. Rocket fuel 10. Rocket candy 11. Rocket 12. Chalk 13. White cross 14. Zip 15. Zoom 16. Yaba 17. Batu 18. Shards 19. Hanyak 20. Hiropon Methylph...

    work page

Show all 21 references
  1. [14]

    Wake ups 15. Zoom 16. Amps 17. Pep pills 18. Study buddies 19. Smarties 20. Beans Cocaine 1. Coke 2. Blow 3. Snow 4. White 5. Powder 6. Yayo 7. Nose candy 8. Charlie 9. C 10. Dust 11. Toot 12. Line 13. Rail 14. Bump 15. Sniff 16. Skiing 17. Blizzard 18. Avalanche 19. Rockstar ...

  2. [15]

    Watson 387 16. 10s 17. 5s 18. 7.5s 19. 10/325s 20. 5/325s Ketamine 1. K 2. Special K 3. Vitamin K 4. Kit Kat 5. Horse tranquilizer 6. Cat Valium 7. Super acid 8. K-hole 9. K-land 10. K-lean 11. K-powder 12. K-rave 13. K-rock 14. K-blast 15. K-bomb 16. K-juice 17. K-wax 18. K-d...

  3. [16]

    G-Flower 18

    G-Bliss 17. G-Flower 18. G-Whiz 19. G-Thing 20. G-Phoria Heroin 1. H 2. Smack 3. Dope 4. Junk 5. Horse 6. Black tar 7. China white 8. Brown sugar 9. Dragon 10. White lady 11. Big H 12. Skag 13. Thunder

  4. [17]

    Yellow boys 19

    French fries 18. Yellow boys 19. Green monsters 20. Hulk bars Amphetamine 1. Speed 2. Uppers 3. Addys 4. Dexies 5. Bennies 6. Black beauties 7. Cross tops 8. Crank 9. Crystal 10. Glass 11. Ice 12. Meth 13. Go fast

  5. [18]

    West Coast 9

    Diet Coke 8. West Coast 9. Pineapple 10. Kiddie Coke 11. R-ball

  6. [19]

    Rids 13. Ritz 14. Riddlin 15. Ritalina 16. Ritaline 17. Ritalyn 18. Ritalina LA 19. Ritalin SR 20. Ritalin LA Opium 1. O 2. Op 3. Poppy 4. Poppy seeds 5. Poppy straw 6. Poppy tea 7. Thebaine 8. Laudanum 9. Dreamer 10. Big O 11. Hop 12. Hophead

  7. [20]

    Aunti Em 15

    Auntie 14. Aunti Em 15. Aunti Emma 16. Black stuff 17. Block 18. Brown sugar 19. Chinese molasses 20. Dover’s powder Oxycodone 1. Oxy 2. OC 3. Hillbilly heroin 4. Percs 5. Blues 6. 30s 7. Roxies 8. Oxycontin 9. Oxy 80s 10. Oxy 40s 11. Oxy 20s 12. Oxy 10s 13. Oxy 5s 14. Oxy 15s...

  8. [21]

    Zoom 13. TAC 14. Rocket smoke 15. Crystal joint 16. Elephant tranquilizer 17. Happy sticks 18. Tic tac 19. Water 20. Peace pill Percocet 1. Percs 2. Paulas 3. Roxies 4. Blueberries 5. 512s 6. 30s 7. Oxy 8. Oxycontin 9. Oxycodone 10. Hillbilly heroin 11. Vikes 12. Painkillers

  9. [22]

    Killers 15

    Happy pills 14. Killers 15. OCs 16. Oxy 80s 17. Oxy 40s 18. Oxy 20s 19. Oxy 10s 20. Oxy 5s Promethazine 1. Lean 2. Sizzurp 3. Purple drank 4. Dirty Sprite 5. Texas Tea 6. Barre 7. Purple jelly 8. Tsikuni 9. Drank 10. Syrup 11. Purple stuff 12. Leanin’ 13. Purple rain 14. Purpl...

  10. [23]

    Gold caps 8

    Liberty caps 7. Gold caps 8. Philosopher’s stones 9. Mushies

  11. [24]

    Zoomers 12

    Funguys 11. Zoomers 12. God’s flesh 13. Sacred mushrooms 14. Teonanácatl 15. Psilocybe 16. Psilocin 17. Psilo 18. Little smoke 19. Silly putty 20. Alice in Wonderland Steroids 1. Roids 2. Juice 3. Gear 4. Sauce 5. Pumpers 6. Stackers 7. Hype 8. A-bombs 9. D-bol 10. Winnie 11. ...

  12. [25]

    White Dove 17

    Stardust 16. White Dove 17. White Rush 18. White Sands 19. Zoom

  13. [2017]

    In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (EMNLP), Copenhagen, Denmark, pages 2598–2607

    Identifying products in ol cybercrime market- places: A dataset for fine-grained domain adaptation. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (EMNLP), Copenhagen, Denmark, pages 2598–2607. Associa- tion for Computational Linguist...

    2017

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.