New AI Tool Makes Sense of Public Opinion Data in Minutes, Not Months
A groundbreaking AI tool named DECOTA has been unveiled, promising to revolutionize the analysis of public attitudes through its ability to effortlessly identify recurring themes in open-ended survey responses and policy consultations. This innovation provides a rapid, cost-effective solution to understanding public sentiment.
Developed by a team at the University of Bath, DECOTA—short for Deep Computational Text Analyser—marks the first open-access method capable of analyzing free-text survey responses and consultations on a large scale. As detailed in a research paper published on April 7, 2025, in Psychological Methods, this tool performs analyses up to 380 times faster and at a cost 1,900 times lower than traditional human analysis, with a 92% concurrence rate with results coded by humans.
The AI harnesses fine-tuned large language models to discern key themes and sub-themes from open-ended responses, where individuals express their views in their own words. This qualitative data, despite its insightful nature, has historically been challenging and time-consuming to process, often leading to it being underutilized.
Origins and Applications of DECOTA
DECOTA was created by a multidisciplinary team from the University of Bath, spearheaded by recent PhD graduates Dr. Lois Player and Dr. Ryan Hughes, with support from Professor Lorraine Whitmarsh. While initially developed to understand public opinions on climate policies, its potential applications are vast. Already it has attracted the interest of four UK governmental bodies, academic institutions, and global think tanks.
Dr. Lois Player, an alumna of Bath’s IAAPS Doctoral Training Centre with a PhD in Behavioral Science, notes: “When thousands of respondents contribute to surveys or consultations, manually analyzing all that free-text data becomes impractical. DECOTA enables an efficient summary of prevalent themes across extensive populations—achieving feats otherwise deemed unfeasible.”
Remarkable Speed with Human-Like Precision
Grounded in the well-established technique of thematic analysis—where researchers manually categorize open-text data into themes—DECOTA mirrors this process through a six-step methodology involving two refined large language models and a clustering approach to identify the primary themes and sub-themes underlying the data.
In testing, DECOTA’s performance was measured against human analysts using four example datasets, revealing that DECOTA detected 92% of the sub-themes and 90% of the broader themes identified by humans. Astonishingly, DECOTA generated insights within a mere 10-minute timeframe compared to an average of 63 hours when conducted by human analysts, representing a staggering 380-fold increase in speed.
The implications of these time savings extend to substantial cost reductions. DECOTA processes responses from approximately 1,000 participants for just $0.82, opposed to roughly $1,575 needed to employ a human research assistant at $25 per hour. Furthermore, it surpasses existing computational methods, such as topic modeling, by being 240 times faster and 1,220 times cheaper.
“It’s crucial to emphasize that DECOTA isn’t meant to replace human thematic analysis, but to augment it,” highlights Dr. Player. “We aim to unlock the vast swathes of unanalyzed data, amplifying the breadth of voices engaged in policy and decision-making processes, while freeing valuable researcher time for in-depth, interpretative work.”
Beyond Thematic Analysis
Further extending its capabilities, DECOTA evaluates which demographic groups are more inclined to mention particular themes. For instance, it can determine if a specific issue is more commonly cited by women than men or whether younger demographics are more likely to highlight certain topics. The tool also draws out representative quotes for each sub-theme, facilitating a nuanced interpretation of results.
Built-In Transparency
Dr. Ryan Hughes, whose PhD concentrated on Mechatronics and Data Science, adds: “Beyond just summarizing data, DECOTA offers depth, detailing who said what, and how frequently. It’s designed with transparency in mind. The data processing is open for inspection and customization at each stage, and all code is accessible on the Open Science Framework.”
Professor Lorraine Whitmarsh applauds the tool: “DECOTA represents a significant advancement in analyzing open-ended questionnaire data. By employing machine learning to manage vast text volumes, it saves time and resources for researchers and policymakers eager to understand public attitudes, reinforcing public engagement in policy formulation.”
DECOTA is freely available online, detailed in the paper “The Use of Large Language Models for Qualitative Research: the Deep Computational Text Analyser (DECOTA),” published in Psychological Methods today (DOI: 10.1037/met0000753). The team plans to further refine DECOTA, including developing a user-friendly web application for those without coding expertise.
