The Research On Language Analysis Of Social E-commerce Based On Machine Learning

With the rapid development of online shopping,social e-commerce shopping has become more and more common.Social e-commerce usually uses text descriptions to share and disseminate products.The resulting large number of social e-commerce language text content carries a wealth of information,including the brand,type,name and features of the products,etc.Extracting these valuable information from massive amounts of social e-commerce language text data contains great analytical value.Therefore,this paper studies the language analysis model of social e-commerce.Based on the BERT pre-training language model proposed by previous researchers,this paper proposes a social e-commerce text classification algorithm and a social e-commerce named entity recognition algorithm,and improves it.Then,the paper also proposed a model optimization method combining the two algorithms.The main content and results of this paper are as follows:1.Research the BERT pre-training language model and the neural network model in language text processing,select the appropriate language model and network structure,and optimize and improve it according to the characteristics of the social e-commerce language text data,so that it can be used in the analysis task of social e-commerce text.2.Propose a social e-commerce text classification algorithm based on BERT-Softmax and a social e-commerce named entity recognition algorithm based on BERT-BiLSTM-CRF.In the social e-commerce text classification algorithm,20 category tags are set for the social e-commerce language text,and in the social e-commerce named entity recognition algorithm,three entity tags of "Brand","Product" and"Feature" are set.The purpose of the algorithm in this paper is to effectively identify these categories and entity tags.3.A model optimization method combining two algorithms is proposed.There is a rule for the category tags and entity tags of social e-commerce language text,that is,after one type of tag is determined,the range of another type of tag can also be determined.Using rule-based and training-based optimization methods can reflect this law,thereby further improving the prediction accuracy of the language model.This paper trains a language model suitable for social e-commerce text,analyzes the effect of the language model on classification tasks and named entity recognition tasks through multiple sets of comparative experiments,and at the same time proves the effectiveness of the improvement and optimization methods of this paper.