Emotion Analysis Of Fine-grained Based On Multi-features

The neural network model captures the emotional features in the text by constructing a non-linear network structure,which has a good performance in the text emotion classification task.However,this kind of method has many areas worthy of improvement.(1)Each model extracts different features by virtue of its own characteristics.How to design a model to use these characteristics to capture as many emotional features as possible and avoid the waste of text information is worth studying.(2)User comments are often not long speeches,but short and powerful texts.How to improve the accuracy of emotion analysis in a short amount of texts.(3)How to analyze the emotion of different objects in the text instead of the whole sentence,and so on.In view of the above problems,this paper carries out the following research.(1)A multi feature extraction model is proposed to solve coarse-grained emotion analysis.This paper proposes an effective feature pre extraction model(AFPEM),which extracts different emotional features by using the characteristics of various models from the dimensions of fixed word collocation,key words and context information;Because of the particularity of Chinese text,word segmentation is needed in emotion analysis.The emergence of new words such as network terms and place names makes the word embedding model unable to achieve the expected effect.Therefore,this paper proposes a word fusion mechanism,introduces the word vector based on the word vector,and fuses the two through the adaptive model to improve the availability of the word embedding model.The experimental results show that AFPEM model contributes to the improvement of emotion classification accuracy.(2)A two-way attention model is proposed to improve the interaction ability between object and text.For fine-grained emotion analysis,there is a correlation between object vocabulary and short text.This paper proposes a multi object interaction model(MOIM)based on object interaction to calculate the importance of each word in the object vocabulary and each word in the text to both sides,and adjust the weight of its own vocabulary according to the importance,To achieve the interaction between text and object.Preprocessing the object vocabulary to improve the recognition of the object vocabulary is conducive to the subsequent emotion classification.An improved loss function is proposed.Different penalties are given to different error classifications to improve the training speed of the model.This paper sets up a control experiment to verify the feasibility of the above views,puts forward a general model structure to improve the accuracy of fine-grained emotion classification,and applies it to the website of computer basic teaching resource platform of Shanghai colleges and universities.