Service Recommendation Based on the User Behavior Sequence

With the development of information technology and the growing communication needs of users, telecom operators have launched various products for the individual market. How to recommend the appropriate products to the right people in a timely and accurate manner is the key to increasing revenue and improving user experience. The previous works mainly based on product features to subjectively filter target users or create traditional machine learning models by integrating different features, the former is limited by subjective influence and has low accuracy rate, the latter is incapable of digging deeper into the hidden meaning of feature crossover, so none of these methods can accurately capture the marketing opportunity brought by user behavior changes. In this paper, we obtain user portrait data and user behavior data, and use Transformer model to capture the deeper intent in user behavior sequences in order to provide accurate and timely services to users, and finally demonstrate through experimental results that the model outperforms traditional Random Forest and LightGBM algorithms.