Research On Key Technologies Of Text Sentiment Analysis For Financial Field

In recent years,Internet platforms have accumulated a large amount of text data.Analysis and mining on it can effectively help people understand the semantic emotional information contained in the data.Especially for the financial field,industry practitioners and investors usually analyze and study corporate financial reports,news announcements and other information to study and judge the company's development status and industry trends,thereby assisting business decision-making and helping relevant departments in corporate supervision.Therefore,according to the characteristics of text data in the financial field,this paper focuses on document-level and entity-level sentiment analysis tasks,and proposes two different model methods to solve the problem that the existing pre-training language model(BERT)cannot handle long text encoding.At the same time,the attention mechanism and standardization method are used to improve the accuracy of the model in entity-level sentiment analysis tasks.In coarse-grained document-level sentiment analysis,although the existing BERTbased sentiment analysis model can effectively use the prior knowledge embedded in the pre-training process,the BERT model cannot provide effective position coding for long texts,so this paper proposes a truncated extraction method.According to the emotional distribution characteristics contained in the data,combined with the extractive summary model,the text input length can be greatly reduced without losing the main emotional information.In addition,consider using different upper-layer networks to realize the fusion of full-text semantic information through the sliding window splicing method.The abovementioned method can effectively avoid the limitation of the existing BERT model on long text encoding,thereby effectively improving the accuracy of sentiment analysis tasks.In fine-grained entity-level sentiment analysis,this paper uses the fine-tuning method of reading comprehension technology to extend the existing BERT model to calculate the emotion of a specific entity.In order to make up for the model's insufficient attention to entity information,the model considers adding entity feature information so that the model can more easily capture entity-level emotional information.In addition,this paper also proposes an attention calculation model based on entity perception.In the model,text features are weighted and summed to make the model pay more attention to key entity information in sentences.At the same time,it considers the use of conditional normalization methods to integrate entity feature information into standardization,which can improve the model's ability to perceive and judge the entity's emotional information,thereby further improving the accuracy of text sentiment analysis in the financial field.Experiments on real financial news data sets show that the related model proposed in this paper has better performance than the original BERT model in terms of accuracy and F1 indicators.