Image analysis considering textual correlations enables accurate user switching tendency prediction

Predicting likely-to-churn users employing surveys is a challenging task. Individuals with different personalities may make different choices in the same situation, so we introduced social media avatars that reflect the user’s psychological state when analyzing their churn tendency. In this paper, we propose a multimodal framework that jointly learns image and text features to establish correlations among users with low net promoter score (NPS) and those likely to churn. We conducted experiments on actual data, and the results show that our proposed method can identify NPS-degraded users in advance, promoting the commercial development of the operator.