Codewords Detection in Microblogs Focusing on Differences in Word Use Between Two Corpora

Hada, Takuro; Sei, Yuichi; Tahara, Yasuyuki; Ohsuga, Akihiko

doi:10.1109/iccece49321.2020.9231109

Cited by 13 publications

(3 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Euphemism detection and its related work has recently received increased attention from the natural language processing and security and privacy communities Magu and Luo, 2018;Pei et al, 2019;Felt and Riloff, 2020;Zhao et al, 2016;Yang et al, 2017;Zhu et al, 2020;Yuan et al, 2018;Hada et al, 2020;Zhu et al, 2021). Existing euphemism detection work have established a number of models by supervised (Pei et al, 2019), semi-supervised and unsupervised learning schemes (Zhao et al, 2016;Magu and Luo, 2018), on diverse categories and platforms (Yang et al, 2017;Hada et al, 2020), with and without distantsupervision Felt and Riloff, 2020).…”

Section: Related Workmentioning

confidence: 99%

“…Research on automatic euphemism detection has recently received increased attention in the natural language processing communities Magu and Luo, 2018;Pei et al, 2019;Felt and Riloff, 2020), and the security and privacy communities (Zhao et al, 2016;Yang et al, 2017;Yuan et al, 2018;Hada et al, 2020;Zhu et al, 2021). However, existing approaches can only detect single-word euphemisms (e.g., "popcorn", "coke"), and fail to detect multi-word euphemisms (e.g., "black tar", "cbd oil") automatically.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Euphemistic Phrase Detection by Masked Language Model

Zhu

Bhat

2021

Preprint

View full text Add to dashboard Cite

It is a well-known approach for fringe groups and organizations to use euphemismsordinary-sounding and innocent-looking words with a secret meaning-to conceal what they are discussing. For instance, drug dealers often use "pot" for marijuana and "avocado" for heroin. From a social media content moderation perspective, though recent advances in NLP have enabled the automatic detection of such single-word euphemisms, no existing work is capable of automatically detecting multi-word euphemisms, such as "blue dream" (marijuana) and "black tar" (heroin). Our paper tackles the problem of euphemistic phrase detection without human effort for the first time, as far as we are aware. We first perform phrase mining on a raw text corpus (e.g., social media posts) to extract quality phrases. Then, we utilize word embedding similarities to select a set of euphemistic phrase candidates. Finally, we rank those candidates by a masked language model-SpanBERT. Compared to strong baselines, we report 20-50% higher detection accuracies using our algorithm for detecting euphemistic phrases.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Euphemistic Phrase Detection by Masked Language Model

Zhu

Bhat

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…These words were not limited to use in crime-related contexts, and it is conceivable that crime-related codewords function with other methods to conceal a given message. As an effort to detect crimerelated codewords in Japanese, Hada et al [19] Focused on the difference in similar words between two corpora, and are working on codeword detection. However, there is room for improvement such as improvement of accuracy and expansion of corpus scale.…”

Section: Related Work On Codeword Detectionmentioning

confidence: 99%

Codeword Detection, Focusing on Differences in Similar Words Between Two Corpora of Microblogs

Hada¹,

Sei²,

Tahara³

et al. 2021

AETiC

Self Cite

View full text Add to dashboard Cite

Recently, the use of microblogs in drug trafficking has surged and become a social problem. A common method applied by cyber patrols to repress crimes, such as drug trafficking, involves searching for crime-related keywords. However, criminals who post crime-inducing messages maximally exploit “codewords” rather than keywords, such as enjo kosai, marijuana, and methamphetamine, to camouflage their criminal intentions. Research suggests that these codewords change once they gain popularity; thus, effective codeword detection requires significant effort to keep track of the latest codewords. In this study, we focused on the appearance of codewords and those likely to be included in incriminating posts to detect codewords with a high likelihood of inclusion in incriminating posts. We proposed new methods for detecting codewords based on differences in word usage and conducted experiments on concealed-word detection to evaluate the effectiveness of the method. The results showed that the proposed method could detect concealed words other than those in the initial list and to a better degree than the baseline methods. These findings demonstrated the ability of the proposed method to rapidly and automatically detect codewords that change over time and blog posts that instigate crimes, thereby potentially reducing the burden of continuous codeword surveillance.

show abstract