Research On Sentiment Word Vector Based On Supervised Learning And Its Application

The rapid development of Internet has produced a large number of texts with opinions which are easily accessible.The proper use of these texts will make great benefit to business and society,which makes sentiment analysis technology important.Word embedding,which is commonly used in various deep learning models,has also been used in sentiment analysis widely.Most of the existing word embedding learning algorithms,such as word2vec,train word embedding in an unsupervised method,which makes the word embedding capture rich syntactic and semantic information but disregard the sentiment information.As a result,words with similar context may have similar vector representations but opposite sentiment polarity,such as delicious and tasteless,leading to bad influence on sentiment analysis.Based on existing research,in this paper we propose a method to embed sentiment information into word embedding and learn sentiment word vector.To evaluate the quality of sentiment word vector,we apply it to sentiment analysis.The work of this paper is as follows:1)To overcome the lack of sentiment information in word embedding,we propose a supervised method to generate sentiment word vector.To overcome the problem in related research,we train the neural network with labeled sentences and words based on skip-gram model and convolutional neural networks,2)We use web crawler to crawl Jingdong for product review data.After the webpage text extraction and text preprocessing,we get product review data with sentiment tags.Then we train sentiment word vector on it with the method we propose.3)We apply the sentiment word vector to learning sentiment lexicons.We purpose a method to construct sentiment lexicons based on sentiment word vector and label propagation algorithm.We firstly construct a word graph with sentiment word vector and then use label propagation algorithm to generate sentiment lexicons.4)We apply the sentiment word vector to sentence-level sentiment classification tasks as inputs.Experiments have shown that the sentiment word vector in this paper has a positive effect on various sentiment classification algorithms.