Help
This tool is a collaboration between the Canadian Centre on Substance Use and Addiction (CCSA) and the DaTALab at York University.
The goal of the tool is to identify posts made on X (formerly Twitter) related to emerging substance use trends, risks, and harms. The tool leverages artificial intelligence (AI) to select and display relevant tweets in real time across Canada.
The goal of the tool is to identify posts made on X (formerly Twitter) related to emerging substance use trends, risks, and harms. The tool leverages artificial intelligence (AI) to select and display relevant tweets in real time across Canada.
Searches all tweets captured by the tool for the words or phrases in the search text. The Enter key can be pressed to apply multiple filters at once. Searches can be for substances (e.g., xylazine), issues (e.g., drug poisoning), data sources (e.g., drug checking), or other terms. Results are reflected on the map and in the summarized results panel to the left of the map.
Selects tweets that fall under one of the three category labels (public health, public safety, drug supply) (Labels filter), one of the three account types being monitored (public health units, harm reduction agencies, police detachments) (Accounts filter), tweets that originate in a certain province or territory (Provinces label), and/or tweets that fall within a certain time frame (Time filter). Filters can be cleared through the “Reset filters” button.
Users who are not logged in can access tweets up to 1 month back. Users who are logged in can access tweets back to January 2023.
Users who are not logged in can access tweets up to 1 month back. Users who are logged in can access tweets back to January 2023.
Displays clusters of posts that meet the current filter or search criteria. Users can select regions of interest and zoom in or out. Moving around the map or changing the zoom level will dynamically change the Summarized Results panel to the left of the map.
Displays a summary of posts selected through the search, filter, or map functions. The tweet counts can be clicked on to zoom the map into the corresponding location.
Located in the bottom half of the tool, this provides the detailed results currently visible on the map and summarized results panel. The panel shows the exact tweet text, the account from which the tweet came, the Label assigned by the AI algorithm, and voting buttons to provide feedback on the relevance and accuracy of each tweet (see sections below for more information). Any changes to the filters, map, or search will update this panel.
The tool currently monitors 305 Twitter accounts across all provinces and territories. These primarily include the accounts of local public health units, harm reduction organizations, and police detachments, as well as the accounts of relevant organizations with a pan-Canadian mandate. Accounts to monitor were identified in a systematic search conducted by CCSA's information specialist. Users who are logged into the tool are able to suggest additional accounts to monitor.
The algorithm works by classifying tweets from monitored accounts into one of 14 categories. Each category indicates if the tweet is about some aspect of public health, public safety, or the contents of the drug supply. If the tweet does not fall into one of the predefined categories, it is discarded as irrelevant.
Classification decisions are based on previous training of a "zero-shot classification" algorithm, described in the publication Journal of Medical Internet Research - Automating Detection of Drug-Related Harms on Social Media: Machine Learning Framework (jmir.org), along with a large language model that can assess and refine the decisions of the zero-shot classifier.
The tool does not display the individual categories but rather the higher-level areas to which the tweet is likely referring. This breaks down as follows:
Note: Both proper and stigmatizing terminology were used to improve the accuracy of the findings.
Classification decisions are based on previous training of a "zero-shot classification" algorithm, described in the publication Journal of Medical Internet Research - Automating Detection of Drug-Related Harms on Social Media: Machine Learning Framework (jmir.org), along with a large language model that can assess and refine the decisions of the zero-shot classifier.
The tool does not display the individual categories but rather the higher-level areas to which the tweet is likely referring. This breaks down as follows:
Label displayed on the interface | Label meaning | Categories under this Label |
---|---|---|
Public Health | Posts that point to drug-related health risks, harms, and adverse events, including drug poisoning events | increased drug overdoses or risk or poisoning, opioid emergency or crisis or advisory or overdose signs |
Public Safety | Posts that point to drug manufacture and supply routes through law enforcement responses | drug trafficking or possession, drug-related seizure or investigation |
Drug Supply | Posts that point to the contents of the unregulated drug supply, including adulterants and emergence of NPS | drug supply or mixing, drug supply alert |
Note: Both proper and stigmatizing terminology were used to improve the accuracy of the findings.
We are continuously refining and updating the performance of the AI algorithm. To help with this, users can submit feedback on the accuracy of the algorithm in classifying tweets. If a tweet label is accurate (Public Health, Public Safety, or Drug Supply), please mark it as accurate. If a tweet is not relevant to emerging substance use risks and harms, or if it is relevant but mis-classified, click "Inaccurate" and select the appropriate Label it should have been given. If a tweet is not relevant at all, click "Inaccurate" and then select "Other."
This tool is a prototype and is continuously being updated and improved. New features will be added as they are developed. For feedback or questions please contact socialreporting@ccsa.ca.
Help
|
|
|
Classification counts over timeA plot of the cumulative classifications in the map view over time
|
Canadian Centre on Substance Use and Addiction
75 Albert Street, Suite 500Ottawa, OntarioK1P 5E7 Canada
Toll free: 1-833-235-4048Phone: 613-235-4048
Follow Us | on Twitter | on Instagram | on Facebook | on LinkedIn |
© 2024 Canadian Centre on Substance Use and Addiction