Text Sentiment Analysis Based On Deep Network Word Embeddings

Sentiment analysis is one of the most basic tasks in natural language processing(NLP),which is usually realized by dictionary or machine learning model.But the appearance of new words will lead to the lack of completeness of affective dictionary,and the problem of semantic representation based on machine learning model will appear.With the development of deep learning,compared with traditional emotion analysis model,deep learning model has improved the generalization ability and robustness of emotion analysis model.Aiming at the problems existing in affective analysis,this paper proposes two models to improve the effect of affective classification by integrating affective dictionary and deep learning model.Specific tasks include:(1)Aiming at the problem of single semantics,a text sentiment analysis model(SLP-ELMo)which integrates context language model ELMo and sentiment lexicon was proposed.Firstly,the sentiment lexicon was used by model and the words in the sentence were filtered;Secondly,the character vector of each word was inputted into Elmo for training which generated by the filtered words were inputted into char-CNN model;In addition,attention mechanism was added to the last layer of Elmo vector to train word vector better;Finally,the word vector and Elmo vector were combined in parallel,and inputted to the classifier for text sentiment classification task.The experimental results show that the model can effectively improve the accuracy of text segmentation..(2)Aiming at the input optimization problem of word embedding layer,a text sentiment analysis model based on emotion polarity sorting of sentiment lexicon and fusion of Bi-GRU and CNN neural network was proposed.First,the sentiment lexicon is used to sort the lexical sentiment scores of the data set,and then the data is aligned as the input of the downstream model.The time cost and the over fitting problem of neural network were considered,a model which integrates Bi-GRU model,CNN model and attention mechanism was proposed by the downstream model.Because of the use of Bi-GRU and CNN,the model not only has good effect on representing the local features of sentences,but also has good advantages in representing the features of long sequence sentences.The whole model is optimized at the input This model can not only capture the semantic information features between words in short text data set,but also represent the semantic information features between sentences in long text.The experimental results show that the model improves the classification effect.