A Data-Driven Approach to Predict “Freebie-Seeker” Behaviors in Digital Marketing

With the popularity of digital marketing, the identification and prediction of the Freebie-Seekers has become particularly critical. This study focuses on two scenarios: live broadcasting and code scanning, aiming to explore and accurately identify such behavior. Traditional identification methods require a lot of manual intervention in model design and parameter adjustment, which not only increases the computational cost, but also makes it difficult to guarantee the accuracy of the final model. We adopt a black-box optimization method to search the parameter space of the model, estimate the conditional probability distributions of the hyperparameters by constructing a Gaussian mixture model, and select the hyperparameter combinations with the best performance based on these distributions. Through feature engineering construction and model selection, we evaluate the LightGBM and XGBoost models. The experimental results reveal that after hyperparameter optimization, both models achieve significant performance improvement, with the F1 value of XGBoost reaching 0.8685 in the QR code scanning scenario, while the F1 value of LightGBM in the live streaming scenario is 0.7662. From the business perspective, our model successfully predicts a large number of Freebie-Seekers in the live streaming and live streaming scenarios. From the business point of view, our model successfully predicts a large number of Freebie-Seekers and reduces the potential economic loss by 20.77% and 1.78% in the live streaming and QR code scanning scenarios, respectively. Therefore, this study not only provides an effective prediction method of Freebie-Seekers, but also provides a strong data support for the strategic decision of digital marketing.