Research On Xiaohongshu Text Sentiment Analysis Based On RoBERTa Model

Xiaohongshu is currently one of the most popular social e-commerce platforms in China,where users can post their shopping tips and product usage experiences.As the number of users continues to grow,the text data on the platform is also growing rapidly.Therefore,for the notes text of Xiaohongshu,this thesis uses natural language processing techniques to dig deeper into the positive and negative sentiment expressions in it,which can not only enable merchants to understand consumers’ attitudes towards products and services,but also provide reference for consumers,with certain theoretical and practical research significance.The traditional sentiment analysis method performs poorly in extracting semantic features in the face of the high degree of colloquialism,grammatical irregularities and endless internet phrases expressed in the Xiaohongshu text.This thesis improves on the traditional sentiment analysis method and presents a sentiment analysis model for the Xiaohongshu text,and the main work is as follows:(1)The Xiaohongshu text dataset is constructed.After acquiring the text of notes posted by users through crawler technology,and then performing text cleaning and manual tagging of sentiment,a dataset containing 10747 texts is finally constructed.(2)For the problem that the text expression of Xiaohongshu is not standardized and the semantic features are difficult to extract,a sentiment analysis model based on RoBERTa-BiLSTM-Attention is proposed.The model introduces RoBERTa based on BERT improvement to obtain vectorized representation of text,solves the problem that traditional Word2 vec cannot represent multiple meanings of a word,uses the improved network BiLSTM to make up for the shortage of LSTM that cannot use the following information,and at the same time integrates with the Attention mechanism to highlight the important sentiment information in the text,and the experimental results show that the model has better classification performance compared with other benchmark models.(3)For the problem that the commonly used pre-trained models with fine-tuned sentiment analysis methods have inconsistent upstream and downstream training objectives,resulting in not giving full play to the capabilities of the pre-trained models themselves,a sentiment analysis model based on RoBERTa fusion of keyword extraction and prompt learning is proposed,which incorporates the extracted keywords into the input text to increase the contextual information of the text,so that the model can better understand the sentiment of the text,and uses prompt learning to improve the utilization of knowledge in the pre-trained language model by designing a prompt template for the downstream task,and the experimental results show that the model has better sentiment analysis results.(4)Based on the proposed model,this thesis designs and implements a prototype system for sentiment analysis of Xiaohongshu text,which mainly includes user management module,data processing module and sentiment analysis and result display module.