User trust is a fundamental determinant of success for digital platforms such as mobile applications, financial technology services, e-commerce systems, and Software-as-a-Service (SaaS) platforms. When trust deteriorates, users gradually disengage, leading to churn, negative reviews, and revenue loss. Traditional evaluation mechanisms such as customer satisfaction surveys, churn analysis, and Net Promoter Score assessments are reactive in nature and detect problems only after trust has already declined. To address this limitation, this study proposes a Trust Loss Early Warning System for Digital Products that proactively identifies early signals of declining user trust. The proposed system integrates behavioral analytics, sentiment analysis, and machine learning techniques to continuously monitor user interactions and detect patterns that indicate potential trust erosion. Key behavioral indicators such as login frequency, feature usage trends, privacy setting modifications, subscription activities, and session duration are analyzed along with textual feedback obtained from customer support interactions and user reviews. A hybrid predictive architecture combining Temporal Convolutional Networks (TCN), Long Short-Term Memory (LSTM) networks, and Bidirectional Encoder Representations from Transformers (BERT) is used to evaluate temporal behavioral patterns and sentiment polarity. These insights are aggregated to generate a Trust Health Index (THI) that ranges from 0 to 100, representing the overall trust level of a user. Based on this index, users are categorized into risk levels, enabling organizations to implement targeted interventions before churn occurs. The proposed system improves transparency through explainable artificial intelligence techniques and supports proactive customer relationship management strategies. Experimental results indicate that the system can accurately detect early indicators of trust degradation, allowing organizations to take preventive actions and enhance long-term user engagement.
IJDIM (Sat,) studied this question.