Where academic tradition
meets the exciting future

You are here: TUCS > PUBLICATIONS > Publication Search > Mining Textual Contents of Qua...

Mining Textual Contents of Quarterly Reports

Antonina Kloptchenko, Camilla Magnusson, Barbro Back, Ari Visa, Hannu Vanharanta, Mining Textual Contents of Quarterly Reports. TUCS Technical Reports 515, Turku Centre for Computer Science, 2003.

Abstract:

A huge amount of electronic information concerning companies’ financial performance is available in organizational databases and on the Internet today. Numeric financial information is important for many stakeholders and has been extensively analyzed for many decades with advanced computational methods. Textual financial reports and news contain not only the factual information about events, but also explain why they have happened. Exploiting finance and business related textual information in addition to numeric financial information could potentially increase the quality of decision-making. Researchers are searching for effective and computationally fairly simple tools that would be able to handle the sophisticated text-related tasks without thorough linguistic preprogramming.
The message, stylistic focus, language and readability of financial reports are good indications of the perspectives and developments of any company. These indications can guide companies’ decision makers to more efficient actions in the dynamic business environment. Although, financial experts and experienced readers can detect those indications and make more precise financial decisions, the manual analysis of textual reports require a lot of time, and time is a costly asset in a financial community. Text Mining methods aim to offer the automatic ways for analyzing and discovering previously unknown patterns in text.
In this paper, we have studied the language and contents of quarterly reports using linguistic and text mining methods. We have compared the results obtained from linguistic analysis of quarterly reports by means of collocational networks and the results obtained from automatic text mining analysis of quarterly report by means of prototype matching clustering. Our objective was to look at how well the computer-aided text-mining tool can perceive the content of quarterly reports in comparison with linguistically motivated collocational networks that outline the most frequent and significant words in the texts. The purpose was to see how meaningful the prototype matching clustering from a perspective of collocational networks linguistic analysis. We performed the study on the quarterly reports from three leading companies in the telecommunications sector, Motorola, Ericsson and Nokia, for years 2000-2001. Our results are somewhat controversial: some of the reports from the companies have as their closest matches the reports with similar collocational networks and some do not have.

Files:

Full publication in PDF-format

BibTeX entry:

@TECHREPORT{tKlMaBaViVa03a,
  title = {Mining Textual Contents of Quarterly Reports},
  author = {Kloptchenko, Antonina and Magnusson, Camilla and Back, Barbro and Visa, Ari and Vanharanta, Hannu},
  number = {515},
  series = {TUCS Technical Reports},
  publisher = {Turku Centre for Computer Science},
  year = {2003},
  keywords = {text mining, annual reports, prototype-matching clustering, collocational networks, collocations },
  ISBN = {952-12-1138-5},
}

Belongs to TUCS Research Unit(s): Data Mining and Knowledge Management Laboratory

Edit publication