Bidirectional Long Short-Term Memory–Based Detection of Adverse Drug Reaction Posts Using Korean Social Networking Services Data: Deep Learning Approaches
Bibliographic record
Abstract
Background: Social networking services (SNS) closely reflect the lives of individuals in modern society and generate large amounts of data. Previous studies have extracted drug information using relevant SNS data. In particular, it is important to detect adverse drug reactions (ADRs) early using drug surveillance systems. To this end, various deep learning methods have been used to analyze data in multiple languages in addition to English. Objective: A cautionary drug that can cause ADRs in older patients was selected, and Korean SNS data containing this drug information were collected. Based on this information, we aimed to develop a deep learning model that classifies drug ADR posts based on a recurrent neural network. Methods: In previous studies, ketoprofen, which has a high prescription frequency and, thus, was referred to the most in posts secured from SNS data, was selected as the target drug. Blog posts, café posts, and NAVER Q&A posts from 2005 to 2020 were collected from NAVER, a portal site containing drug-related information, and natural language processing techniques were applied to analyze data written in Korean. Posts containing highly relevant drug names and ADR word pairs were filtered through association analysis, and training data were generated through manual labeling tasks. Using the training data, an embedded layer of word2vec was formed, and a Bidirectional Long Short-Term Memory (Bi-LSTM) classification model was generated. Then, we evaluated the area under the curve with other machine learning models. In addition, the entire process was further verified using the nonsteroidal anti-inflammatory drug aceclofenac. Results: Among the nonsteroidal anti-inflammatory drugs, Korean SNS posts containing information on ketoprofen and aceclofenac were secured, and the generic name lexicon, ADR lexicon, and Korean stop word lexicon were generated. In addition, to improve the accuracy of the classification model, an embedding layer was created considering the association between the drug name and the ADR word. In the ADR post classification test, ketoprofen and aceclofenac achieved 85% and 80% accuracy, respectively. Conclusions: Here, we propose a process for developing a model for classifying ADR posts using SNS data. After analyzing drug name-ADR patterns, we filtered high-quality data by extracting posts, including known ADR words based on the analysis. Based on these data, we developed a model that classifies ADR posts. This confirmed that a model that can leverage social data to monitor ADRs automatically is feasible.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
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 source (direct Gemma or distilled Codex), 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".