Fine-grained Sentiment Analysis For Internet Home Decoration Reviews

With the popularity of the Internet and the booming development of emerging technologies such as big data and artificial intelligence,many industries are using the Internet as a platform for innovation,forming the rapid development model of "Internet plus".Internet home decoration is one industry that has flourished under this model.Its rise has generated a lot of comments about home decoration on the Internet.These reviews contain a lot of important sentiment information,and sentiment analysis of such information not only helps Internet home decoration platforms and companies to understand users’ needs and improve their services or products in a targeted manner,but also provides decision support for users.Fine-grained sentiment analysis is an important task for sentiment analysis.However,most existing datasets for fine-grained sentiment analysis are in English and there is a lack of Chinese datasets,and only a few domain-specific datasets are available.Based on this background,this thesis adopts a deep learning approach to fine-grained sentiment analysis of Internet home decoration reviews.The main research contents are as follows:1.Fine-grained sentiment analysis datasets based on Internet home decoration reviews are constructed.The structure of an Internet home decoration web page is studied,and the corresponding data collection scheme is designed.The collected comment data is stored in a MySQL database after data pre-processing.The target-oriented opinion word extraction dataset and aspect-based sentiment classification dataset are constructed by annotating the data.2.An attention-enhanced object-oriented opinion word extraction model,A-IOG,is proposed.It first uses Inward-LSTM and Outward-LSTM to generate contextual representations and target-related contextual representations,respectively.Then an attention mechanism is introduced into Att-Global-LSTM to make the model pay more attention to words related to the target word and generate enhanced contextual representations.Finally,the three context representations are spliced into Softmax for classification to extract opinion words.Comparative experiments are conducted on multiple datasets to verify the effectiveness of the proposed model.3.A sentiment classification model,SKDGCN,is proposed based on a dual-channel graph convolutional network with sentiment knowledge.It firstly strengthens the syntactic relationship between words through the sentiment knowledge of SenticNet,and builds a Sentiment-enhanced Dependency Graph Convolutional Network(SDGCN).Then,it learns the semantic correlation between words by introducing an attention mechanism,and builds an Attention Graph Convolution Network(AGCN).The two graph convolutional networks are used for interactive learning for sentiment classification.Comparative experiments are conducted on multiple datasets,and the experimental results show that considering sentiment knowledge can effectively improve the classification performance of the model.4.A fine-grained sentiment analysis service based on Internet home decoration reviews is designed and implemented.First,the visualization design of the information overview of Internet home decoration and the analysis of Internet home decoration companies are completed,and then the proposed models A-IOG and SKDGCN are encapsulated and presented to users in the form of interfaces.