Can you think of someone familiar who has been affected by alcoholism in some way? For many of you probably the answer is yes, since about 6% of US adults ages 18 and older suffers from Alcohol Use Disorder. For many of these affected people, the Alcoholics Anonymous (AA) program has been providing a venue where they can get social support. They can share any problems they experience along the way as well as get inspired from other individuals who have built a successful recovery. Many people continue going to the meetings even though they have been sober for many years.

Even though sharing in the meetings is considered a vital component of recovery, searching for this information can often become overwhelming. AA currently maintains the list of meetings by regions. That is, individual local groups maintain their own websites that contain meetings happening in a particular city or in a region. For instance, if you want to look for meetings in Minneapolis, one has to search on, whereas meetings in Saint Paul are listed on Therefore, searching for nearby meeting information becomes particularly difficult for someone new in recovery who may already be overwhelmed with a lot of recovery related resources and information, or someone traveling or relocating to a new city who is not familiar with the area. Using search engines to find out information doesn’t help much. Try Googling for “AA meetings near me”, and you will see hundreds of different websites and meeting finder apps in the result like the image below. You will keep wondering which link you should click on. Not only that, many of these websites and apps contain static or outdated meeting information, which may often send you to wrong meeting locations. This is very frustrating for many alcoholics who might be on the verge of a relapse and/or in desperate need of attending a meeting.

Search results for "AA meetings near me"
Search results for “AA meetings near me”

With an aim to provide a reliable mechanism for finding AA meetings, we extracted and validated data from diverse local websites to make  the information available in one place in a searchable format. We were interested in providing a countrywide list of meetings that does not rely on local groups continually updating it (which is the current reason why meeting finders are useless and are immediately outdated for many areas).

Different regional websites have different types and structures (maps, calendars, documents, images, etc.). Extracting data from websites of different formats and structures posed technical challenges for us, and we approached these challenges with a combination of automated information retrieval and human computation techniques. The process starts with automatically gathering content of all regional AA websites and extracting meeting day, time, and locations from those pages using an information retrieval algorithm that leverages patterns in the structure of the page. In order to be more certain of the extracted information, we use “crowdsourcing,” which involves making bite-sized tasks for online human workers to validate the output from the automated technique. We have recruited workers from an online crowdsourcing platform: Amazon Mechanical Turk. They complete tasks that involve categorization of AA webpages into meeting-page or non-meeting page and validating whether the automatically extracted time and location of the meetings are accurate or not. We named this process HAIR (human-aided information retrieval).

Steps of HAIR (Human-Aided Information Retrieval)

Combination of automated information retrieval algorithms and crowdsourcing yielded better results than applying automated methods only. Our automated retrieval technique failed to detect some pages containing a small number of meetings (this is common for small cities), whereas human eyes could easily identify and categorize these meeting pages. Additionally, in many cases where the algorithm extracted wrong bits of information about a meeting, crowdworkers could modify them with correct information. This is particularly important because otherwise “partially correct” information about a meeting (e.g., right day and time but wrong address) can be potentially confusing, frustrating, and can potentially send people to a location where there is no meeting, misdirecting people at a critical stage of their recovery.

However, if we want to keep the list updated we have to continue this extraction and human validation periodically to pull the latest meeting information. Automated approaches do not cost much to run but recruiting workers periodically will cost us a lot of money. As an alternative, we plan to leverage the AA members’ willingness to help the newcomers in the program by voluntarily validating meeting information. To know more about how HAIR works, read the full paper here.

Written by

Sabirat Rubya is a PhD candidate at the GroupLens Research of University of Minnesota. Her research focuses on recommending and designing technology for recovery from alcoholism. She completed her bachelors in Computer Science and Engineering from Bangladesh University of Engineering and Technology.

Comments are closed.