Abstract:In order to solve the problem of finishing and filing 95598 customer service complaints, including: the problem of archiving caused by negligence in the process of manual archiving, that is, the accuracy of archiving; the cost of manually inaccurate the complaints Time problem, that is, efficiency problem; manual analysis of customer service complaints is insufficient, and it is impossible to accurately and quickly locate the problem of user appeal hotspots, that is, depth problems. This paper gives a solution to the above three problems, using the combination of word2vec and XGBoost to achieve accurate summary of the 95598 customer service complaints. In the process of vectorization of text words, the word2vec method is used to obtain the text word vector of the word; the XGBoost algorithm is used to classify and archive 95598 customer service complaints, and the responsible department, professional classification, appeal event, and error point of the historical complaint work order Four aspects are marked. The classification accuracy of this model is about 83%-91%, which has a good effect. Based on the results of the work order classification, and related complaints board, the data is displayed more intuitively.