Knowledge Embedding For Sentiment Classification

With the growth of e-commerce and social platforms,a lot of review texts are emerging rapidly.How to automatically analyze these massive review texts is a challenging topic in the natural language processing.Sentiment classification is a powerful tool to analyze these review texts.Sentiment classification has important application value in fields of business decision-making,information retrieval and information security.Sentiment knowledge has a strong emotional color compared with ordinary features.Embedded knowledge can make full use of the emotional factors in the Chinese texts to simplify the construction of classification model at a certain extent.Combining the idea of knowledge embedding,this paper mainly studies the feature representation and classifier construction for Chinese text sentiment classification.The main research work is as follows:1.A feature-weighting method based on knowledge embedding is proposed.In the Chinese text sentiment classification,traditional feature representation methods usually ignore the importance of language knowledge.By constructing a feature embedding model,the contributions of part-of-speech or sentiment words are embedded in the traditional TF-IDF(term frequency-inverse document frequency)weighting,where the best contribution value is obtained by particle swarm optimization algorithm.Finally,support vector machine is used to classify the Chinese texts.In the experiment,the performances of different knowledge,such as part-of-speech,sentiment words and their combination,are compared.The experimental results show that the method based on part-of-speech embedding obtains the best classification performance and improves the accuracy of Chinese text sentiment classification dramatically.2.A Chinese text sentiment classification method is proposed based on part-of-speech embedding and kernel extreme learning machine.According to the disadvantages of traditional classification algorithms for sentiment classification,such as complicated parameter learning and low classification performance,a novel Chinese text sentiment classification approach based on kernel extreme learning machine is proposed.Firstly,feature selection of training data by information gain is used to reduce the input dimension,and then a classifier based on the wavelet kernel extreme learning machine is constructed for the Chinese text sentiment classification.The experimental results show that the model parameters of the proposed method are easier to learn.Moreover,its classification performance is usually superior to support vector machine and naive Bayes.Moreover,the part-of-speech embedding and kernel extreme learning machine are combined to further improve the classification performance.Experimental results show that the addition of part-of-speech embedding further improves the classification performance of the kernel extreme learning machine.