Prediction of outcomes in online self-help for alcohol problems with NLP analysis of free text
| Reference number | |
| Coordinator | Region Stockholm - eStöd, Maria Ungdom |
| Funding from Vinnova | SEK 375 270 |
| Project duration | April 2020 - May 2021 |
| Status | Completed |
| Venture | AI - Competence, ability and application |
| Call | Start your AI-journey! For public organizations |
Important results from the project
As per the intended goals and objective, this project developed and evaluated a prediction model to be able to predict outcomes of a popular, anonymous online support service for alcohol problems, with a high degree of accuracy, using natural language processing of free text answers and machine learning.
Expected long term effects
A measurable goal of the project was to achieve 85% accuracy in predictions of outcomes. In combination with demographic variables, the model we developed resulted in 70% accuracy of outcomes in the short term and 86% in the long term. We have secured other external funding to continue developing these methods and results. The project has entailed a substantial skill and knowledge increase (both in specific methodologies as well as applicable laws / regulations) for the coordinating partner and has begun to give the intended systemic effects and results dissemination.
Approach and implementation
After arranging ethical permits and the necessary legal agreements, we used the Gavagai service for natural language processing of already collected free-text answers from program exercises, which resulted in a thematic model at the user level. The variables from this model (30*4) were then exported to predict outcomes using machine learning.