Annotated Drug Use Tweets

We introduce an English Twitter dataset designed for the detection of online drug use, comprising 112,057 tweets accompanied by metadata. This dataset underwent manual annotation by a team of expert annotators consisting of around 30 members, these annotators, possessing diverse multidisciplinary backgrounds and expertise, committed over six months to meticulously label each tweet. In order to classify a tweet as related to drug use or not, specific criteria must be met, including references to actions like using, smuggling, promoting, encouraging, selling, or buying illicit drugs of any kind. Within the drug use dataset, we found a total of 112,057 tweets posted by 90,621 unique users. Among these, 48,080 tweets (43%) were classified as drug use within (T) label. In contrast, 63,977 tweets (57%) were classified as non drug use within (F) label. To assess the dataset's balance, we applied Shannon’s entropy measure, yielding a result of 0.985, which indicates a well-balanced dataset.