Researchers Find Social Media and Dark Web Activity can help Deep Learning Models Generate Substance Abuse Data for Policy Makers

https://arxiv.org/pdf/2103.15764.pdf

A team of international researchers conducted a study on the US opioid epidemic by assessing the data from social media and crypto market activities of opioid users.  They used deep learning models to study substance abuse by collecting data from crypto markets and performing sentiment analysis by correlating this data to social media hashtags. 

According to the Global Drug Survey, the percentage of participants who purchase drugs through cryptomarkets has tripled between 2014 and 2020. 

The purpose of this study is to assess social media data from active opioid users to identify what types of feelings they express and compare it to the drug usage data obtained from crypto markets.

How was the research conducted?

Step 1: Data Collection

Researchers collected over 70,000 opioid-related listings by sending a dedicated crawler to three crypto markets on the dark web- Dream market, Tochka, and WallStreet Market – from March 2018 through January 2019. 

In addition, they collected 290,458 opioid-related posts from six subreddits using custom-built crawlers. 

As subreddits do not allow users to tag their emotions, they crawled Twitter. They collected tweet hashtags that correspond to the seven emotions: Joy, Sadness, Anger, Love, Fear, Thankfulness, and Surprise.

Step 2: Data Analysis

Raw data from crypto markets was processed using previously-trained deep learning Named Entity Recognition (NER) to extract information such as substance names, product weight, and shipment information. This data was used to identify drug entities and categorize them into eight categories by referring to an existing database of drug entities. 

The researchers classified subreddit posts as positive, negative, and neutral. They then implemented the Python library Textblob to generate related sentiments for each subreddit post. 

Deep learning algorithms LTSM, BERT and CNN were used to generate emotion labels for the Tweets and correlate them to the data generated by Textblob. Among the three algorithms, LTSM showed the highest accuracy of 91.2%.

What are some significant results of the research?

Kratom, Heroin, Fentanyl, Morphine, Cocaine, Methadone Suboxone, and Oxycodone are the most widely discussed drugs on the six subreddits that the researchers studied. The researchers were able to find out three emotions most associated with the widely discussed drugs.

Top 3 emotions associated with each drug based on the analysis of data from subreddits and Twitter hashtags by deep learning algorithms

Source: https://arxiv.org/pdf/2103.15764.pdf

The results also show that we can derive slang terms, brand names, novel drugs, mental health symptoms, and medications from the data generated from social media. 

Comparison for drugs Kratom, Heroin, and Oxycodone by seven emotions

Source: https://arxiv.org/pdf/2103.15764.pdf 

Future Application of the Findings

The findings from this research paper can be used to bring together policymakers from the field of AI and healthcare to formulate policies allowing for timely intervention in cases of substance abuse. Such approaches have the potential to prevent overuse-related deaths and combat addiction.

It is also a first step in understanding the relationships between mental health and drug usage. It also helps us design better technology to regulate the sale of drugs through crypto markets.

The challenges ahead

Researchers realize that they need more precise deep learning algorithms to crawl the crypto market for drug sales data. These algorithms also need to be trained better to identify more categories of drugs and classify data more accurately. 

Artificial intelligence is slowly shifting from just building businesses to addressing deep-rooted social challenges such as drug abuse. We must support this shift and continue to develop better versions of AI that can help human societies evolve towards a better future. 

Paper: https://arxiv.org/pdf/2103.15764.pdf

Reference: https://thenextweb.com/neural/2021/03/30/scientists-artificial-ai-link-cryptomarkets-heroin-fentanyl-substance-abusers-reddit-twitter/