Sentiment Analysis Of Wooden Furniture Reviews Based On Mixed Features

Sentimental factors play an increasingly important role in the process of selecting furniture materials and designing furniture style.In this study,with the help of sentiment analysis method,we extracted and analyzed the sentimental information of wooden furniture reviews of e-commerce website.This paper mainly includes three experiments: the experiment of constructing domain lexicon,the experiment of training sentiment polarity classifier and visual analysis according to different type of reviews,and the experiment of aspect-based sentiment analysis.First of all,a domain lexicon was built based on reviews of wooden furniture.The first thing to do is to select the seed words according to our corpus,and expand the sentiment words by SO-PMI algorithm based on these seed words,so as to construct the domain sentiment lexicon and farther enhance the applicability of our lexicon.The gradient descent formula is used to assign different weights to degree words so as to improve the calculation accuracy of sentimental score.Some reviews were randomly extracted from our corpus and labelled the sentimental polarity of these reviews manually.Finally,lexicon-based method which using traditional sentiment lexicon was used to test the effect of generated domain lexicon.Secondly,the mixed vector model is used to construct the sentiment polarity classifier.Firstly,the sentimental features of the corpus are extracted,and the matrix of sentimental feature is formed according to the semantic rules.On this basis,the word frequency feature matrix generated by TF-IDF model is spliced to the sentiment matrix and a mixed vector model can be obtain,and then the mixed vector model is used to train the classifier.Afterwards,the constructed classifier is used to classify the unlabeled reviews into three categories: positive reviews,neutral reviews and negative reviews.Visual analysis and statistical analysis of word frequency are carried out on the three types of corpus respectively,and then the information in different emotional tendency reviews can be mined.Thirdly,in order to calculate the sentiment score of attribute features in our corpus.This study extracts part of the experimental data for syntactic dependency analysis.The product feature trees were constructed based on the experimental data.Referring to the product feature trees,the extracted <attribute,adjective > binary is manually divided into different primary-features.According to the number of attribute pairs contained in each primary-feature and the average sentiment score of adjective words,the sentiment score of each primary-feature is calculated.Finally,the results of sentiment analysis are visualized.The experimental results show that after integrating the domain lexicon,according to the results of visual analysis of different sentiment polarity corpus,the following suggestions can be put forward: Firstly,the appearance of wooden furniture can meet the aesthetic needs of most consumers;Secondly,consumers' attention to price factors is mainly reflected in cost performance;In addition,manufacturers and distributors should put the quality of products in the first place;Finally,customers have the ability to attract potential new consumers for manufacturers and distributors.By constructing the wooden furniture feature tree and e-commerce website feature tree and calculating the emotional value of each feature,it is concluded that the wooden furniture features that consumers pay attention to from high to low are appearance features,quality features,economy,comfort and safety,while the e-commerce website features that consumers pay attention to from high to low are selectivity,consumer satisfaction,service and shopping experience.The ranking order of emotional scores of each level feature is basically consistent with the results of visual analysis of different emotional polarity.