Short Texts Sentiment Anaysis Based On Deep Learning

Nowadays,internet,especially mobile internet,is full of short texts with sentiment due to its rapid development.Mining the sentiment information contained in these texts can obtain a lot of business information and social information.This paper mainly to mine the text sentiment information,two methods based on deep learning algorithm are proposed.The first method aims to alleviate the learning bias of a single algorithm through assembly using different types of deep learning algorithms.Boosting and bagging algorithms are used to research the combination of several convolutional neural network with slightly modified.Multiple sampling algorithms are also used to improve the diversity of the basic algorithm.The second method generates better vector representation of text by reducing the noise in the representation.The features of the short text are often sparse when the deep learning algorithm used directly,the vector representation of the texts will have too much noise.Through the multi-task algorithm that training a number of tasks related to sentiment classification simultaneously,more useful information is encoded to the text representation.Multiple test sets are used to evaluate the proposed methods.The experimental results show that these algorithms have strong generalization ability,and the overall effect is basically in line with expectations.