Improving News Traders using CRF: Using Conditional Random Fields to reduce Feature Space

摘要: Since financial markets react to news very quickly it is necessary even quicker in order make money. Normal text data has high dimensionality and reducing the number of features needed classify a document reduces time do so. This thesis looks at way reduce feature space by use Conditional Random Fields. To this, new set made using mandatory stock messages released Oslo Stock Exchange. The are combined with on all trades completed three-year period. A Field trained textual used extract important features. then train Support Vector Machine classifier Forest classifier. Both evaluated against randomly selected find that results 4 percentage point reduction accuracy 81,25 run time. We conclude possible without significant accuracy. also this method not good enough for making profit market. consistent earlier work reduction.