Detecting Novel and Emerging Drug Terms Using Natural Language Processing: A Social Media Corpus Study
Bibliographic record
Abstract
BACKGROUND: With the rapid development of new psychoactive substances (NPS) and changes in the use of more traditional drugs, it is increasingly difficult for researchers and public health practitioners to keep up with emerging drugs and drug terms. Substance use surveys and diagnostic tools need to be able to ask about substances using the terms that drug users themselves are likely to be using. Analyses of social media may offer new ways for researchers to uncover and track changes in drug terms in near real time. This study describes the initial results from an innovative collaboration between substance use epidemiologists and linguistic scientists employing techniques from the field of natural language processing to examine drug-related terms in a sample of tweets from the United States. OBJECTIVE: The objective of this study was to assess the feasibility of using distributed word-vector embeddings trained on social media data to uncover previously unknown (to researchers) drug terms. METHODS: In this pilot study, we trained a continuous bag of words (CBOW) model of distributed word-vector embeddings on a Twitter dataset collected during July 2016 (roughly 884.2 million tokens). We queried the trained word embeddings for terms with high cosine similarity (a proxy for semantic relatedness) to well-known slang terms for marijuana to produce a list of candidate terms likely to function as slang terms for this substance. This candidate list was then compared with an expert-generated list of marijuana terms to assess the accuracy and efficacy of using word-vector embeddings to search for novel drug terminology. RESULTS: The method described here produced a list of 200 candidate terms for the target substance (marijuana). Of these 200 candidates, 115 were determined to in fact relate to marijuana (65 terms for the substance itself, 50 terms related to paraphernalia). This included 30 terms which were used to refer to the target substance in the corpus yet did not appear on the expert-generated list and were therefore considered to be successful cases of uncovering novel drug terminology. Several of these novel terms appear to have been introduced as recently as 1 or 2 months before the corpus time slice used to train the word embeddings. CONCLUSIONS: Though the precision of the method described here is low enough as to still necessitate human review of any candidate term lists generated in such a manner, the fact that this process was able to detect 30 novel terms for the target substance based only on one month's worth of Twitter data is highly promising. We see this pilot study as an important proof of concept and a first step toward producing a fully automated drug term discovery system capable of tracking emerging NPS terms in real time.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".