Research On Machine Learning Based Intelligent Customer Identification Technology

In recent years,domestic and foreign trade companies have made full use of the Internet to find requisite information of foreign trade and develop corresponding business around the world,leading to the continuous improvement of the scale and level of foreign trade.However,current foreign trade companies mainly rely on manually crawling and identifying information from the Internet.Their high cost,low recognition efficiency,and poor accuracy have severely restricted the rapid development of foreign trade companies and the sustainable development of their scale.Therefore,there is an urgent need to use a new generation of artificial intelligence technology to replace the traditional human-based customer information identification.The thesis focuses on a intelligent identification method of foreign trade customers.On the basis of in-depth investigation of current foreign trade customer identification technology requirements,a new generation of artificial intelligence technology and machine learning methods are introduced to study the accurate and efficient identification of massive potential customer information crawled from the Internet.In order to solve the inaccurate representation problem of text semantic information in the traditional single model-based customer intelligent identification algorithm.Firstly,the thesis innovatively proposes a customer intelligent recognition algorithm based on a multi-text representation model by using weighted improved GloVe(Global Vectors for Word Representation,GloVe)model and BERT(Bidirectional Encoder Representations from Transformers,BERT)model,the text data is vectorized in parallel to enhance the context information.Compared with the traditional models,the improved GloVe model can better capture the grammatical and semantic information of words.The thesis merges the text representation results of two models and classifies them by the use of random forests to obtain customer information identification results.Then,based on the multi-text representation model,the thesis conducted in-depth comparative testing and evaluation of the performance of the proposed customer intelligent recognition algorithm.The results show that the proposed multi-model identification algorithm can effectively identify customers,and the performance evaluation indicators are better than traditional single model method based on GloVe orBERT.Finally,the thesis designs and implements a foreign trade customer data recognition prototype system according to the software engineering concept to show the actual effect of the proposed algorithm.This research can effectively save labor costs with potential applications in foreign trade technology in the context of the current era of artificial intelligence.