| Sentiment analysis is an important task of natural language processing,which refers to the process of analyzing,processing,and summarizing text with emotional colors.With the rapid development of the Internet,a large amount of commodity-related information has appeared on the Internet,and they all affect the future trend of commodity prices to a certain extent,and commodities are at the bottom of the national commodity trading chain.Therefore,study this Some of the information has important implications for studying commodity prices.However,at present,sentiment analysis mainly focuses on commodity evaluation,Weibo,Twitter text and other fields,and there are few researches on sentiment analysis of commodity information.Therefore,this paper studies the texts about commodities,and establishes a sentiment classification model for commodity information based on its corpus characteristics,which can analyze the sentiment tendency of commodity information.Finally,this paper studies the correlation between the sentiment tendency of commodity information and its market price trend.The main research contributions of this paper are as follows:(1)Establish a commodity information sentiment classification model based on the custom network layer ARCNN.For the complex relationship between Chinese words and words,a custom network layer ARCNN is designed and added to the model for training and verification.The model includes an input layer,an ARCNN network layer,a fully connected layer,and a classification layer.Experiments show that the accuracy rate reaches 92.4%,and it has a good classification effect.After comparing with the traditional CNN model,it was found that the classification accuracy of the ARCNN model was 5.3%higher than that of the traditional CNN model on the same samples and models,and the accuracy of the ARCNN model on the general corpus reached 83%and 98%,respectively.The effectiveness of ARCNN is demonstrated.(2)Establish a pre-training-based sentiment classification model for commodity information.In view of the effectiveness of the pre-trained model,the pre-trained model is used as a pattern extractor,followed by a sentiment classification layer,and then the model is fine-tuned using the commodity information corpus with sentiment classification as the training task,so that the model parameters can extract the characteristics of the commodity corpus.Features and adaptation to downstream tasks to improve subsequent training efficiency and improve classification accuracy.The experimental results show that the model parameters of the pre-trained model tend to be stable after training,and its accuracy rate reaches 97.6%,which proves that the pre-trained model is usable.(3)Design emotional technical indicators and price technical indicators,and study the relationship between emotional tendencies of commodity information on the Internet and commodity prices.Using the experimental data results,the correlation between the sentiment index and the price index is calculated to be 0.916,which proves that the sentiment classification result of commodity information has a correlation with the price trend of commodities. |
