Knowledge Base Discovery Tool*

AAAI 1999

TL;DR

This paper introduces the Knowledge Base Discovery Tool (KBDT), a system designed to move beyond keyword-based document indexing and searching towards a meaning-based approach. By leveraging an ontology to represent concepts and their relationships, KBDT aims to enable more intelligent and context-aware information retrieval.

What Problem Does This Paper Address?

Traditional search engines and document indexing methods often rely on matching keywords, which can lead to irrelevant results due to polysemy (words with multiple meanings) and synonymy (different words with the same meaning). This paper addresses the challenge of improving information retrieval by understanding the meaning of words and their context, rather than just their literal presence. The authors highlight the limitations of keyword matching, especially when dealing with the nuances of human language, and propose a system that builds a knowledge base to capture these semantic relationships.

Methods

The core of KBDT is its knowledge base, which is built upon an ontology. This ontology is initialized using the WordNet database and is structured around "concepts." A concept in KBDT represents a collection of synonyms along with a description of its usage, effectively disambiguating words with multiple meanings.

The paper describes two key prototypes:

  • Knowledge Base Editor: This tool allows humans to browse and interact with the ontology. It facilitates finding concepts, selecting concepts with multiple senses, and navigating between related concepts via links.
  • Intelligent Information Retrieval Engine (IIR Engine): This engine goes beyond simple keyword extraction. When a user enters a query, the IIR engine retrieves relevant concepts from the knowledge base. If a word has multiple meanings (concepts), the user is prompted to clarify which meaning they intend. This semantic context is then used to conduct a more focused search, aiming to retrieve documents that specifically match the intended meaning.

Datasets

Not discussed in the paper. The paper mentions that the initial ontology was derived from the WordNet database.

Key Insights and Ideas

  • Meaning over Keywords: The fundamental idea is that understanding the meaning and context of words is crucial for effective information retrieval, moving beyond simple keyword matching.
  • Ontology as a Semantic Backbone: Using an ontology to represent concepts, synonyms, and their relationships provides a structured way to capture and leverage semantic knowledge.
  • Disambiguation through User Interaction: The system actively involves the user in disambiguating word meanings when multiple concepts are identified, leading to more precise search results.
  • Concept-based Indexing: Storing indexes based on concepts (senses) rather than just individual words significantly enhances the representative power of the knowledge base.

Key Results

The paper focuses on demonstrating the functionality of the prototypes rather than presenting quantitative results. The key qualitative findings are:

  • The Knowledge Base Editor successfully allows users to browse and navigate the ontology, identifying concepts and their relationships.
  • The IIR Engine demonstrates the ability to retrieve concepts related to query terms and to use user-provided semantic contexts to refine searches.
  • The authors suggest that the quality of semantic-context information retrieval depends on the richness of the knowledge base.

Limitations and Future Work

Limitations:

  • The paper acknowledges that the quality of results is dependent on the amount and kind of information in the knowledge base.
  • The current prototypes are early demonstrations, and extensive quantitative evaluation is not presented.

Future Work:

  • Development of a parsing engine for extracting concepts directly from documents.
  • Creation of a concept builder to assist users in expanding the knowledge base by defining new terms found in documents.
  • Prototypes of these future tools are expected by mid-year 1999.

Summary updated

BibTeX
@inproceedings{1999knowledge,
  title = {{Knowledge Base Discovery Tool*}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1999},
}
Figure