Research On The Method Of User Comment Sentiment Analysis Based On Deep Learning

The Internet has gradually become the main medium for people to express their opinions and insights.Various e-commerce and social networking platforms have a huge amount of users,and with them comes a huge amount of text and information,and people’s online activities and real life have become inseparable.There are huge social and commercial values in these user comment information,and sentiment analysis on them to grasp users’ sentiment tendency can help government and other organizations to deal with public opinion,and commercial companies to quickly analyze product quality and make improvements,as well as to assist them in business decision making.Traditional sentiment analysis methods include recurrent neural networks(RNN),which are temporal in nature,cannot be computed in parallel,and cannot directly extract the contextual semantic features of sentences;convolutional neural networks(CNN)can be computed in parallel and have advantages in extracting local features of sentences and computation speed,but they do not perform well in expressing global features of sentences.Therefore,this paper builds a model based on attention mechanism to solve the problem of sentiment analysis of sentence-level and aspect-level text comments,and the main work of this paper is summarized as follows.(1)The methods of text preprocessing are summarized,the mainstream models for generating word vectors are compared,and the deep learning-based sentiment analysis models are analyzed and organized.(2)To address the problem of insufficient feature extraction capability of the traditional model,the Light-Transformer model is proposed,which improves the structure of the Transformer,keeping only the Encoder module,based on word vector representation and positional embedding,and finally the extracted sentence features are input to two fully connected layers for classification.(3)For the problem of fine-grained aspect-level sentiment analysis,the LightTransformer-ALSC model is proposed,where the context vector and the aspect vector are stitched twice and a separate attention network is used to model the aspect words,and finally the fused vectors are used for classification to better evaluate the impact of the aspect words on the overall text sentiment polarity.(4)According to the practical requirements of sentiment analysis,we built a sentiment analysis system based on Django development framework in Python language,designed and implemented the business architecture and database module of the system,applied the proposed algorithm to it,and tested the system.The results show that the accuracy of the model proposed in this paper is improved by 2.48%-9.76% on the NLPCC2014 Task2 dataset and 1.3%-5.5% on the Sem Eval2014Task4 dataset compared to the RNN and CNN-based models;at the same time,the model structure is simpler compared to other models based on attention mechanism with similar accuracy and the model structure is simpler and the number of parameters is smaller than other attention-based models.The sentiment analysis system in this paper has reliable performance and practicality,and can analyze the sentiment tendency of the input text more accurately,which supports the use of downstream tasks such as opinion analysis and data annotation.