| Financial news has a huge impact on the stock market and is closely watched by various of participants in stocks market.Computer networks and portable devices provide convenient channels to release and disseminate information but lead investors fall into "Information Trek." Thus,to extract the part related to a certain stock from the massive financial news and classify the information levels is an interesting question.The research can make information orderly and mine the value of news deeply.And it can also be a base to expand various application-oriented knowledge service projects.Therefore,this article intends to catch and classify financial news from a vast ocean of information based on theoretical foundation and Machine Learning automatically.To measure financial news’ value related to stocks,this article constructs a quantitative indicator system based on the Theory of News Value.Then,we extract features of news and stock by Natural Language Processing,Named Entity Recognition,Topic Word Extraction and statistics methods.Relying on the above work,a research context-oriented Stock Persona and News Profile are constructed.The “classification”and “grading” of financial news correspond to the "recall" and "refinement" process of Recommendation Systems.Eventually,our framework can select and grade the related news for each stock automatically.In theory,this paper conducts a value quantitation system related to stocks as a new way to classify news.The system is based on the most fundamental scientific theories in fields of Informatics,Journalism,Finance,etc.Further,the introduction of Recommendation System Framework is another innovation for solving the problem of Text Classification.Firstly,it can dig out the complex connections between stock and news.Secondly,it improves interpretability,scalability and portability by integrating various external auxiliary information into the framework.In practice,it serves to achieve significant application services as a ground,such as financial news recommendation,stock market forecast,etc.The limitations of this research are mainly reflected in the process of discovering and extracting the connections between each stock and news.In the process,we utilize content features including entities,topics and so on.However,they can also generate connections through key people,investment and financing behaviors.Follow-up researches may consider to construct Knowledge Map to organize information of the Stock Persona and News Profile.Confidently there will be an improvement in extracting more implicit and measuring complex connections between a piece of news financial new and a stock. |
