In order to solve the problems of poor data processing effect and large resource consumption in natural language data processing model, a fusion algorithm based on multi-scale feature extraction and attention mechanism is proposed. Through the feature data extraction of different scales and the application of weighting algorithm on the feature map, the attention to some specific scale features is enhanced, and the natural language data processing model is optimized based on the fusion algorithm. The results of simulation experiments show that the feature extraction effect of the fusion algorithm is better, and the ability of computer data processing is significantly improved. Comparing the performance of the optimized natural language processing (NLP) data processing model with CSAMT data processing model, BETG data processing model and NLP data processing model before optimization, it can be seen that the performance of the NLP data processing model optimized by CBAM-MS-CNN is better than other models. The research results show that the fusion algorithm can meet the business requirements of high reliability and intelligent processing in the field of unstructured data management in the electronic transfer process, improve data processing efficiency and data quality, and reduce the workload of manual input data and manual review data.