User Profile Construction Based on Multi-Channel Fusion

Aiming at the shortcomings of existing user profile construction algorithms that do not fully utilize contextual structural information and multidimensional feature information., a user profile construction model based on the pre-training model NEZHA, Deep Pyramidal Convolutional Neural Networks (DPCNN), Gated Recurrent Unit (GRU), and attention mechanism is proposed, NCDGA. The method uses multiple channels to process the text, capturing different dimensions of information and constructing features to reduce the prediction bias that occurs when a single model captures a single feature. The experimental results show that the NCDGA model has better accuracy and effectiveness compared with other modeling methods.