Extracting Adverse Drug Reactions from Social Media

Authors: Andrew Yates, Nazli Goharian, Ophir Frieder

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We propose three methods for extracting ADRs from forum posts and tweets, and compare our methods with several existing methods. Our methods outperform the existing methods in several scenarios; our filtering method achieves the highest F1 and precision on forum posts, and our CRF method achieves the highest precision on tweets.
Researcher Affiliation Academia Andrew Yates, Nazli Goharian, and Ophir Frieder Information Retrieval Lab Department of Computer Science Georgetown University {andrew, nazli, ophir}@ir.cs.georgetown.edu
Pseudocode No The paper describes the methods (LLDA, Naive Bayes, CRF) in text but does not provide structured pseudocode blocks or algorithm listings.
Open Source Code No The paper states that 'The annotations and a list of the crawled URLs are available on the authors website' and 'Both the tweet ids and the tweet-drug mappings for the tweet-drug mappings for the tweets in our corpus are available on the authors website', which refers to data. It does not provide concrete access to the source code for the methodologies proposed in the paper.
Open Datasets Yes The forum corpus ground truth consists of a random subset of the corpus that was annotated to indicate the ADRs expressed in each post... The annotations and a list of the crawled URLs are available on the authors website.
Dataset Splits Yes Five-fold cross-validation is used with all reported results.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions software tools such as 'Stanford Parser' and 'Twit IE tagger (Bontcheva et al. 2013)', but it does not specify version numbers for these or other software dependencies used in their experimental setup, nor for the programming language or libraries used for their own implementations.
Experiment Setup Yes We use a sliding window size of 5 (n=5) with the Sliding Window method and with multinomial Na ıve Bayes (Multinomial NB); we empirically chose n=5 for both methods. Additive smoothing is used with α = 1.