Emotional Analysis Of E-commerce Comments Based On Deep Learning

Compared with traditional commodity transactions,e-commerce provides online comment function,which has a crucial impact on consumers' choice in the context of fierce competition on e-commerce platforms.It is of great commercial value and significance to research how to accurately tap into consumers' emotions in mass reviews and improve product quality,sales strategy and customer service accordingly.This paper adopts the method of deep learning to study the emotional tendency reflected by e-commerce comment data.The accuracy of emotion analysis was improved by combining LSTM,attentional mechanism,BERT and CNN.The main work of this paper is as follows:(1)in view of the traditional neural network Long Short-Term Memory(LSTM)in the analysis of coarse-grained emotion understanding semantics is not comprehensive,capture the inaccurate problem of emotional information,this paper proposes a fusion Attention mechanism(Attention)combined with two-way LSTM network BiLSTM-Attention network model,using both short-term and long-term Memory neural network for temporal sequence analysis of the structure,fully excavate potential information in this sentence,effectively overcome the problem of Long forgotten.BiLSTM model can better capture the information before and after the sentence by bidirectional analysis of the language text compared with the ordinary LSTM model.The use of attention mechanism can assign weight to different features in the sentence,pay attention to the feature information that tends to the user's emotion in the sentence,and effectively improve the recognition efficiency.The experimental results show that the neural network model of fused attention mechanism and bidirectional LSTM is more effective than other commonly used models in the review data set of a mobile phone of jd.(2)in the fine-grained emotional analysis of e-commerce comments,several common language models,such as Word To Vector(Word2Vec)and Global Vectors for Word Representation(GloVe),can only generate immobilized semantic Word Vectors.Aiming at this problem,this paper puts forward a preliminary training of Bidirectional Encoder Representations from Transformers(BERT)network and convolution the BERT of combiningneural network(CNN)--CNN network model.Compared with Word2 Vec,BERT model changes the context-independent static vector into context-dependent dynamic vector,further increases the generalization ability of the word vector model,fully describes the relationship between character level,word level and sentence level,and better realizes the mapping of words in high dimensional space.In this paper,BERT model is used to obtain the word vectors in sentences,and the convolutional neural network is used for further feature selection and dimensionality reduction,so as to obtain the user's emotional orientation.The experiment shows that in the multi-granularity review data set of jingdong e-commerce,the recognition rate of bert-cnn network model is higher and the effect is improved greatly.However,compared with the traditional neural network model,it takes a lot of time to train the model.