Classification of Mobile Data Plan Recommendation Failure Reasons Based on BERT and Text CNN

To provide better services and enhance customer satisfaction, telecom operators often call users to recommend suitable telecom packages. Classifying the reasons for recommendation failures helps analyze user needs, allowing for adjustments in operational strategies and package attributes. BERT generates contextually relevant word embeddings through its deep bidirectional Transformer encoder, capturing rich semantic information, while Text CNN extracts key local features through convolutional operations and max-pooling layers. By using the word embeddings generated by BERT as input to Text CNN, we can combine the strengths of both models, thereby enhancing performance in text classification tasks. To improve long-text classification performance, this paper proposes a fusion model combining BERT and Text CNN. In the experiments, we compared BERT, Text CNN, and the BERT-Text CNN model across different datasets. The results show that, compared to using BERT or Text CNN alone, the BERT-Text CNN model improves classification accuracy across various datasets, validating the effectiveness of this fusion model in enhancing classification performance.