Multiplicative Emotion Recognition in Power Marketing Using Optimized Hybrid Intelligence with Causal Inference

Recognition of customers' emotional state is of high importance when applied in contemporary power marketing to personalize services and improve customers' satisfaction. In this work, a new approach to recognize the emotions of customers using Multiplicative Emotion Recognition Framework that combines optimized Naive Bayes and Decision Tree models with causal inference to create an efficient emotion recognizer in power marketing has been proposed. Multiplicative hybridization and causal inference are used in the framework to optimize feature interaction and construct effective causal relationships between customer interaction data and customer emotions. The above-mentioned approach was assessed through its implementation on an altered customer complaint dataset, extracted from an open-access complaints archive. In order to enable further analysis in the context of power marketing, specific transformations relevant to the domain such as complaint-to-domain mapping, substitution of keywords, and feature engineering relevant to utilities were carried out to produce service-related features, including outage time, voltage fluctuation, billing errors, and delays in response times. Finally, the processed data consisted of 12,458 customer interactions categorized into four emotional groups: Satisfaction, Frustration, Anger, and Trust. Obtained experimental results show that the approach surpasses benchmark models. Accuracy rate was calculated as 94.7%, precision reached 93.8%, recall was 92.9%, while F1-score was equal to 93.35%. Proposed model significantly outperforms its components such as Naive Bayes and Decision Tree achieving 81.4% and 87.6% accuracies respectively. Moreover, integration of causal inference helped reduce prediction bias by 28.4% and increased model transparency. Real-time testing has revealed that proposed framework works with the average delay of 47 ms, which allows applying it in online marketing. Overall, obtained results prove applicability of the multiplicative hybrid intelligence model in emotion recognition for power marketing and other similar problems.