Commodity Evaluation's Sentiment Analysis And Research Based On Deep Learning

With the rapid development of E-commerce,more service scenarios happened both offline and online.E-commerce has changed from simple goods transaction to a much more complex online service transaction,which called “new retail”.For example online service like Didi's car reservation,Cainiao logistics' express reservation and Ctrip's travel logistics reservation are playing a more and more important role in people's daily life.Under this tendency,the more customer comments we get,the more important of extracting valuable information from customer comments become.Those comments are so important in the areas of the iterative optimization of services,purchase decision making and market data mining.From there,the sentiment tendency of these text reviews becomes even more important,because it could reflect the user's feedback of using products and services.Since the previous manual or machine without intelligent can't deal with the huge amount of data,it's time to use artificial intelligence-related technologies to get the final results.This article addresses these issues and coming up the following ideas.(1)First,this paper proposes a construction method of sentimental lexicon based on word embedding for E-commerce evaluation.This construction method could also be used in many other fields of quick construction sentiment lexicon.In order to use it,firstly crawling,collecting,preprocessing data and tagging high-frequency sentiment words from online comments.Then,construct a dictionary by using the full-word embedding's,which established by the word2 vec tool and the combination of both words and reference words.Finally,verity the effectiveness of sentiment lexicon construction method and the constructed E-commerce sentiment lexicon through comparing the above with the sentiment classification effect of the traditional sentiment lexicon.(2)This paper also come up with an sentimental comprehensive word embedding based on sentiment calculation.First,calculate relevant sentiment quantification value for each single sentence review based on E-commerce sentiment lexicon and proposed sentiment calculation method.Then,integrated the quantified value into its own vocabulary based neural network Skip-gram model to form a sentimental comprehensive word embedding with more prominent sentiment features.Finally,design a comparison experiment between experiment sample and normal word embedding to verify the superiority of the sentimental complex word embedding.(3)Third,this paper proposed an emotional sentiment classification model based on sentimental comprehensive word embedding and convolutional neural network models.First,complete the sentiment feature extraction,learning and sentimental tendency classification by using the sentimental comprehensive word embedding as the input of text expression data and the convolutional neural network model of self-constructed deep learning.Then,compare the result with the traditional machine learning method support vector machine design to verify their superior performance.(4)Finally,the Shunt-C&RNN text evaluation sentiment classification model was proposed.In the same way,use the sentimental comprehensive word embedding as the input data structure of the text,independently designed the deep learning convolutional neural network model and the recurrent neural network model.It is to implement a shunting rule(splitter)to implement the judgment input to the data.The operation of the deep learning network utilizes the advantages of the local feature description of the convolutional neural network and the sequential features of the recurrent neural network through the shunt to achieve a more efficient and accurate E-commerce evaluation of text sentiment classification.This paper designed several sets of comparative experiments to verify the excellent overall performance of the Shunt-C&RNN model.