Change search
ReferencesLink to record
Permanent link

Direct link
Improving News Traders using CRF: Using Conditional Random Fields to reduce Feature Space
Norwegian University of Science and Technology, Faculty of Information Technology, Mathematics and Electrical Engineering, Department of Computer and Information Science.
2013 (English)MasteroppgaveStudent thesis
Abstract [en]

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

Place, publisher, year, edition, pages
Institutt for datateknikk og informasjonsvitenskap , 2013. , 60 p.
URN: urn:nbn:no:ntnu:diva-23248Local ID: ntnudaim:8794OAI: diva2:659653
Available from: 2013-10-26 Created: 2013-10-26 Last updated: 2013-10-26Bibliographically approved

Open Access in DiVA

fulltext(774 kB)723 downloads
File information
File name FULLTEXT01.pdfFile size 774 kBChecksum SHA-512
Type fulltextMimetype application/pdf
cover(184 kB)10 downloads
File information
File name COVER01.pdfFile size 184 kBChecksum SHA-512
Type coverMimetype application/pdf
attachment(363 kB)13 downloads
File information
File name ATTACHMENT01.zipFile size 363 kBChecksum SHA-512
Type attachmentMimetype application/zip

By organisation
Department of Computer and Information Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 723 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 71 hits
ReferencesLink to record
Permanent link

Direct link