David Banks, Text Analysis Interest Group Chair

The Council of Sections recently approved the petition for the Text Analysis Interest Group to become a section. The interest group will officially become a section on January 1, 2023, and Wendy Martinez, a former president of the ASA, will serve as the first chair.

More than 200 ASA members signed the petition and promised to become members of the new section. The following is a list of the sorts of problems that arise in text analytics:

• It is important to study US political blogs over time to identify changing themes and rhetorical styles, understand sentiment, and quantify polarization. Tim Au, now at Google, did significant work on this.
• To the extent Wikipedia is a mirror of human knowledge, it may be possible to find gaps. Also, the network structure and topic modeling give independent pieces of information about Wikipedia—if the clique structure in the network corresponds to distinct topics, it corroborates the underlying organizational coherence. Dave Blei at Columbia has worked on aspects of this problem.
• One can apply latent Dirichlet allocation to more than text. In particular, Qiuyi Wu, a graduate student at the University of Rochester, has applied LDA to music, and it could probably be applied to ecological diversity, visual arts, and other fields.
• Many statisticians are attempting to mine medical records to find patterns of symptoms, comorbidities, or treatments. Text analysis can also be used to discover medical insurance fraud or improper treatment. Bethany Percha at Mount Sinai and Kasper Jensen at the University of Warwick have independently done text analysis of medical records, but there are many others, often working in multidisciplinary teams.
• Text analysis can apply to many data sets in the federal government. The Consumer Protection Agency receives written reports about washing machines that catch fire. The FDA’s Adverse Event Reporting System has records of bad outcomes associated with specific drugs and drug combinations. The US Department of Transportation’s Fatality Analysis Reporting System contains reports on all fatal accidents involving vehicles. There are statisticians at all these agencies who either use or want to use text analytics to find early warnings. This is likely to be one of Wendy Martinez’s pet projects when she takes charge.
• In surveys of all kinds (federal, state, commercial, political), response rates are falling. Transcribed conversational interviews may be more robust and better able to collect nuanced information. Nick Fisher, a private consultant, collected such survey data, and Christine Chai, who is now at Microsoft, analyzed it.

There is no shortage of work in this field, and members of the Text Analysis Interest Group are keen to transition to full section status.