Designing Information Systems Through Text Mining: A Case Study of Fire Service Documentation

The Challenge

After every intervention by the State Fire Service (PSP), the Incident Commander (KDR) prepares a paper report documenting the operation. This report, structured as a form called “Information from the Incident,” includes a section titled “Descriptive Data for Incident Information,” where the KDR describes various aspects of the rescue operation using natural language. While the paper form divides this section into six subcategories (e.g., rescue actions, equipment used, weather conditions), the electronic version in the EWID system merges everything into a single unstructured text block.

This unstructured format poses significant challenges. For example, searching for “hydrants on Mickiewicz Street” might return irrelevant results, such as all rescue operations on that street, rather than specific hydrant information.

The Solution: Text Mining

To address these issues, I proposed a text mining (TM) approach—a specialized branch of knowledge discovery in databases (KDD)—focused on analyzing textual data. While existing TM tools often process documents as a whole or rely on natural language processing (NLP) for tasks like lemmatization, they overlook the importance of individual segments (e.g., sentences) as standalone units of information.

My research introduced a novel method called text-driven software design, which emphasizes:

  1. Segmenting text into meaningful units (e.g., sentences).

  2. Classifying segments into semantic categories (e.g., “operations,” “equipment,” “weather”).

  3. Extracting structured information to build a functional information system.

Key Achievements

  • Developed a dedicated rule-based segmenter (SR) that outperformed existing tools (95.5% accuracy vs. 87–88.7%).

  • Designed a semantic classifier (SKSS) achieving 90% accuracy in categorizing segments.

  • Created a system for topic-specific information extraction (SEIt), enabling precise retrieval of data (e.g., hydrant locations or equipment status).

  • Demonstrated the method’s versatility through case studies, including fire incidents and traffic accidents.

Why It Matters

This approach transforms unstructured fire service reports into actionable insights, improving data retrieval and decision-making. It also lays the groundwork for similar applications in other domains where textual data is underutilized.

For further details, explore the full autoreferat.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.