In this study, the researchers identified linguistic markers that could be used to identify people at risk of problem gambling (PG). They analyzed 19,000 customer service chats sourced from an Italian gambling operator. The researchers found that people at higher risk of PG had longer and more complex conversations. They more frequently discussed financial losses and used heightened emotional and threatening language compared to those with lower risk. A model was developed to predict risk of PG based on such linguistic markers. The model was able to correctly identify the risk level of users 78% of the time.
In the second phase of the study, the model was further refined using a new dataset of 5,900 conversations. A four-level risk classification was applied (no risk, low/moderate risk, high risk, and severe problem gambling). Finally, an interactive system and digital interface was developed to help with early risk detection.
