An insight analysis and detection of drug-abuse risk behavior on Twitter with self-taught deep learning

Han, Hu; Phan, Nhat Hai; Chun, Soon Ae; Geller, James; Vo, Huy T.; Ye, Xinyue; Jin, Ruoming; Ding, Kele; Kenne, Deric R.; Dou, Dejing

doi:10.1186/s40649-019-0071-4

Cited by 29 publications

(17 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our own prior studies used both supervised machine learning classifiers and an unsupervised topic model to detect internet pharmacies selling opioids (including fentanyl) [6,7,21,26]. Others have primarily focused on analyzing Twitter messages for opioid and substance abuse behavior with manually annotated data, examining social circles of users, measuring user sentiment, using natural language processing, and using deep learning [22,23,27,28]. Other studies have used deep learning models to detect and describe adverse drug reactions via Twitter [29,30].…”

Section: Introductionmentioning

confidence: 99%

A Machine Learning Approach for the Detection and Characterization of Illicit Drug Dealers on Instagram: Model Evaluation Study

Li¹,

Xu²,

Shah³

et al. 2019

J Med Internet Res

View full text Add to dashboard Cite

Background Social media use is now ubiquitous, but the growth in social media communications has also made it a convenient digital platform for drug dealers selling controlled substances, opioids, and other illicit drugs. Previous studies and news investigations have reported the use of popular social media platforms as conduits for opioid sales. This study uses deep learning to detect illicit drug dealing on the image and video sharing platform Instagram. Objective The aim of this study was to develop and evaluate a machine learning approach to detect Instagram posts related to illegal internet drug dealing. Methods In this paper, we describe an approach to detect drug dealers by using a deep learning model on Instagram. We collected Instagram posts using a Web scraper between July 2018 and October 2018 and then compared our deep learning model against 3 different machine learning models (eg, random forest, decision tree, and support vector machine) to assess the performance and accuracy of the model. For our deep learning model, we used the long short-term memory unit in the recurrent neural network to learn the pattern of the text of drug dealing posts. We also manually annotated all posts collected to evaluate our model performance and to characterize drug selling conversations. Results From the 12,857 posts we collected, we detected 1228 drug dealer posts comprising 267 unique users. We used cross-validation to evaluate the 4 models, with our deep learning model reaching 95% on F1 score and performing better than the other 3 models. We also found that by removing the hashtags in the text, the model had better performance. Detected posts contained hashtags related to several drugs, including the controlled substance Xanax (1078/1228, 87.78%), oxycodone/OxyContin (321/1228, 26.14%), and illicit drugs lysergic acid diethylamide (213/1228, 17.34%) and 3,4-methylenedioxy-methamphetamine (94/1228, 7.65%). We also observed the use of communication applications for suspected drug trading through user comments. Conclusions Our approach using a combination of Web scraping and deep learning was able to detect illegal online drug sellers on Instagram, with high accuracy. Despite increased scrutiny by regulators and policymakers, the Instagram platform continues to host posts from drug dealers, in violation of federal law. Further action needs to be taken to ensure the safety of social media communities and help put an end to this illicit digital channel of sourcing.

show abstract

Section: Introductionmentioning

confidence: 99%

A Machine Learning Approach for the Detection and Characterization of Illicit Drug Dealers on Instagram: Model Evaluation Study

Li¹,

Xu²,

Shah³

et al. 2019

J Med Internet Res

View full text Add to dashboard Cite

show abstract

“…In our past work, 16 we aimed to investigate the opportunity of using social media as a resource for the automatic monitoring of prescription drug abuse by developing an automatic classification system that can classify possible abuse versus no-abuse posts. In some studies, 53 54 , deep learning models were developed to detect drug abuse risk behavior using two datasets. The first dataset was manually annotated and a deep learning model trained on the first dataset was applied to annotate the second dataset automatically.…”

Section: Discussion and Post-classification Analysesmentioning

confidence: 99%

Text Classification Models for the Automatic Detection of Nonmedical Prescription Medication Use from Social Media

Al-Garadi

Dan

Ruan

et al. 2020

Preprint

View full text Add to dashboard Cite

Prescription medication (PM) misuse/abuse has emerged as a national crisis in the United States, and social media has been suggested as a potential resource for performing active monitoring. However, automating a social media-based monitoring system is challenging--requiring advanced natural language processing (NLP) and machine learning methods. In this paper, we describe the development and evaluation of automatic text classification models for detecting self-reports of PM abuse from Twitter. We experimented with state-of-the-art bi-directional transformer-based language models, which utilize tweet-level representations that enable transfer learning (e.g., BERT, RoBERTa, XLNet, AlBERT, and DistilBERT), proposed fusion-based approaches, and compared the developed models with several traditional machine learning, including deep learning, approaches. Using a public dataset, we evaluated the performances of the classifiers on their abilities to classify the non-majority "abuse/misuse" class. Our proposed fusion-based model performs significantly better than the best traditional model (F1-score [95% CI]: 0.67 [0.64-0.69] vs. 0.45 [0.42-0.48]). We illustrate, via experimentation using differing training set sizes, that the transformer-based models are more stable and require less annotated data compared to the other models. The significant improvements achieved by our best-performing classification model over past approaches makes it suitable for automated continuous monitoring of nonmedical PM use from Twitter.

show abstract

“…In our past work [16], we aimed to investigate the opportunity of using social media as a resource for the automatic monitoring of prescription drug abuse by developing an automatic classification system that can classify possible abuse versus no-abuse posts. In some studies [23], 23, deep learning models were developed to detect drug abuse risk behavior using two datasets. The first dataset was manually annotated, and a deep learning model trained on the first dataset was applied to annotate the second dataset automatically.…”

Section: Related Workmentioning

confidence: 99%

Text classification models for the automatic detection of nonmedical prescription medication use from social media

Al-Garadi

Yang

Cai

et al. 2021

BMC Med Inform Decis Mak

View full text Add to dashboard Cite

Background Prescription medication (PM) misuse/abuse has emerged as a national crisis in the United States, and social media has been suggested as a potential resource for performing active monitoring. However, automating a social media-based monitoring system is challenging—requiring advanced natural language processing (NLP) and machine learning methods. In this paper, we describe the development and evaluation of automatic text classification models for detecting self-reports of PM abuse from Twitter. Methods We experimented with state-of-the-art bi-directional transformer-based language models, which utilize tweet-level representations that enable transfer learning (e.g., BERT, RoBERTa, XLNet, AlBERT, and DistilBERT), proposed fusion-based approaches, and compared the developed models with several traditional machine learning, including deep learning, approaches. Using a public dataset, we evaluated the performances of the classifiers on their abilities to classify the non-majority “abuse/misuse” class. Results Our proposed fusion-based model performs significantly better than the best traditional model (F1-score [95% CI]: 0.67 [0.64–0.69] vs. 0.45 [0.42–0.48]). We illustrate, via experimentation using varying training set sizes, that the transformer-based models are more stable and require less annotated data compared to the other models. The significant improvements achieved by our best-performing classification model over past approaches makes it suitable for automated continuous monitoring of nonmedical PM use from Twitter. Conclusions BERT, BERT-like and fusion-based models outperform traditional machine learning and deep learning models, achieving substantial improvements over many years of past research on the topic of prescription medication misuse/abuse classification from social media, which had been shown to be a complex task due to the unique ways in which information about nonmedical use is presented. Several challenges associated with the lack of context and the nature of social media language need to be overcome to further improve BERT and BERT-like models. These experimental driven challenges are represented as potential future research directions.

show abstract

An insight analysis and detection of drug-abuse risk behavior on Twitter with self-taught deep learning

Cited by 29 publications

References 34 publications

A Machine Learning Approach for the Detection and Characterization of Illicit Drug Dealers on Instagram: Model Evaluation Study

A Machine Learning Approach for the Detection and Characterization of Illicit Drug Dealers on Instagram: Model Evaluation Study

Text Classification Models for the Automatic Detection of Nonmedical Prescription Medication Use from Social Media

Text classification models for the automatic detection of nonmedical prescription medication use from social media

Contact Info

Product

Resources

About