Sentiment Classification Of Ecommerce Platform User Comments Based On Deeping Learning

With the rapid development of information technology,the deep integration of the Internet and big data has become a significant feature of current technological development.The development of e-commerce platforms has exploded with massive user-generated content.Mining these huge amounts of data and information is essential to grasp consumer behavior and enhance the core competitiveness of enterprises.However,there are still many problems that restrict the mining of potential information.It mainly comes down to three points: one is that the user-generated data is huge,and relying solely on manual analysis is often laborious;the second is that the Chinese language is complex,and the machine's understanding of the language is not satisfactory;the third is that the country started late in the field of text mining and the technical system It is still not perfect,and it needs continuous innovation and development.In order to simplify the text mining workload to a certain extent,improve the accuracy of text mining,and enrich the relevant technical system,this article is driven by large-scale e-commerce platform user review text,combined with BERT(Bidirectional Encoder Representations from Transformers)pre-training model,breakthrough The defects of the static representation of traditional word vectors use Transformer to describe the global semantic information of the text.On this basis,the BERT + Bi-LSTM + Attention classification model is constructed and designed to improve the accuracy of traditional classification tasks.Finally,mine the user's "pain points" and their degree distribution in the user's "bad reviews" data.The main work done in this article is divided into the following aspects:(1)At present,text sentiment classification mostly uses word2 vec for vector space representation.The sparseness of short text features causes the vector space to be sparse,and the word vectors are mapped in the same space,which cannot solve the phenomenon of polysemy.In response to these problems,this paper proposes a model migration method,fine-tuning the BERT pre-training model on the Chinese review short text data set to capture the bidirectional word information in the text,which better solves the polysemy word mapping in Chinese text Problems in the same vector.This paper combines BERT and BiLSTM,and introduces an attention mechanism to extract key text features that affect the classification of emotion polarity,which effectively improves the classification accuracy of the model and obtains better classification results.(2)Propose a mining method for comment text "pain points" based on sentiment dictionary and rules.At present,most of the existing pain point mining methods are based on theoretical knowledge,based on marketing theory,and few computer analysis techniques are used.Based on this,this paper proposes a pain point calculation method based on sentiment dictionary.By extracting feature words and using sentiment dictionary to calculate the sentiment value of the text,the user 's pain points and concerns are obtained,so as to propose consumer perception quality improvement strategies and consumers Desired control strategy.(3)Conduct experimental research and analysis on the above models respectively.The text sentiment classification model uses different models for comparison.The data obtained by the web crawler and the basic data set IMDB are used as two sets of experimental data to classify the text.Using crawleracquired data as the data source,the model is used to mine the user's "pain points",analyze the experimental results,and provide targeted opinions and suggestions.Experiments show that the classification model proposed in this paper can effectively improve the accuracy of the classification task,and the "pain point" mining model is also convenient and valuable.