Aspect-level Text Sentiment Analysis Based On Hybrid Neural Network

With the development of various social media e-commerce platforms on the Internet,more and more users are posting comments on these platforms.According to statistics,the amount of text information in comments is growing at a speed of tens of millions.Analyzing text data on Internet platforms is of great value to both users and platforms,and will also greatly promote the development of aspect-level sentiment analysis tasks.The purpose of aspect-level sentiment analysis is to identify aspect items in the text and judge the sentiment polarity of these aspect items by learning contextual text information.After an in-depth analysis of the existing aspect-level sentiment analysis solutions,this paper finds that the existing sentiment analysis algorithms have the following problems.First of all,most of the existing aspect-level sentiment analysis tasks only describe the sub-task of sentiment classification,the model is relatively simple,and the extraction task of aspect items is rarely described,and the importance of the adjacent context information of the aspect item and the importance of the contextual information of the aspect item are ignored in the aspect item extraction task.Information about part-of-speech features.Secondly,the existing algorithm models are not enough to learn the dependencies between context text and aspect items,ignore the connection between aspect items and context text context,and the model is relatively simple.In view of the above problems,the main research contents of this paper are as follows:(1)A hybrid neural network-based aspect extraction model is proposed to solve the problem of aspect extraction in aspect-level sentiment analysis tasks.The model uses a pre-trained language model to learn the word vector representation of text,and then convolution The network and the bidirectional long-term and short-term memory network are mixed to obtain a hybrid neural network.The network uses CNN to obtain the local features of the vectorized text,and fuses the part-of-speech features and pre-trained word vector features together as the input of Bi＿LSTM for further learning.The text semantic features of a variety of information are combined,and finally the CRF is used as a constraint to extract the final aspect item.Compared with other benchmark models,the accuracy and F1 value of the model on the Co NLL-2003 dataset are improved,which proves the effectiveness of this model.(2)An aspect-level sentiment analysis model based on Roberta and attention mechanism is proposed to obtain the sentiment polarity of different aspects in the text.The model uses the Roberta pre-trained language model to represent the text word vector,and then uses the multi-head attention mechanism.The interdependence of each word in the context text is obtained,and deeper semantic features are learned.At the same time,Bi＿GRU is used to extract the semantic information of aspect items.In order to fully learn the dependencies between context text and aspect items,the interactive attention mechanism is used to enhance the two aspects.influence of the person.Finally,compared with the existing models to prove the superiority of the model proposed in this paper.