Communications-based early detection of gambling-related problems in online gambling

Most algorithms developed for the early identification of gambling-related problems rely on predictors aggregated out of transactional gambling data. However, as a notable extension, one algorithm uses predictors derived from written correspondence with players and thereby opens up a so far unused resource for the early detection of gambling-related problems. In this article, a sample of 1008 emails from self-excluders and controls to the customer services of an online gambling operator was reanalysed to explore the possibility of using automated text analysis software to extract quantitative markers from written player correspondence. For this purpose a text analysis tool, using psychometrically validated English and German dictionaries, was applied. While the classification results that were based solely on automated text analysis were nearly on a level with those attained by human assessment, the application of an automated prediction model can even add incremental validity to human judgements. A combined model, relying on human rating as well as the scales Anger, Time and Causation, derived from automated text analysis, displayed improved validity and classification rate. Discussed in the light of practical application, the results indicate that automated text analysis can be deployed as an expert system to prioritize cases and to support human judgement.