Abstract In knowledge management systems, ontologies play an important role as a backbone for providing and accessing knowledge sources. They are largely used in the next generation of the Semantic Web that focuses on supporting a better cooperation between humans and machines. Since manual ontology construction is costly, time-consuming, error-prone, and inflexible to change, it is hoped that an automated ontology learning process will result in more effective and more efficient ontology construction and also be able to create ontologies that better match a specific application. Ontology learning has recently become a major focus for research whose goal is to facilitate the construction of ontologies by decreasing the amount of effort required to produce an ontology for a new domain. However, most current approaches deal with narrowly-defined specific tasks or a single part of the ontology learning process rather than providing complete support to users. There are few studies that attempt to automate the entire ontology learning process from the collection of domain-specific literature and filtering out documents irrelevant to the domain, to text mining to build new ontologies or enrich existing ones.
The Knowledge Acquisition Bottleneck
In the context of the Semantic Web, an ontology is a formal, machine-readable map of concepts and their relationships (e.g., "a frog is a type of amphibian"). These maps are the backbone of intelligent systems, but they have a major flaw: they are incredibly tedious to build. Manual construction is slow, expensive, and often outdated by the time it's finished.
Automated "ontology learning" aims to solve this, but most existing tools are either too narrow or struggle with polysemy (words having multiple meanings). For example, if an automated system sees the word "cell," should it add it to a biology ontology or a telecommunications one? This paper tackles the challenge of end-to-end ontology enrichment—finding new relevant words and accurately placing them in the right "neighborhood" of the knowledge tree.
Methodology: A Dual-Track Framework
The authors propose two distinct approaches within a single framework to automate the growth of ontologies.
1. Lexical Expansion (WordNet-based)
The system takes "seed" words from an existing ontology and looks them up in WordNet. To solve the ambiguity problem, the researchers developed a Reference Hypernym Tree. By comparing the "family tree" (hypernyms) of a WordNet sense to the top-level structure of the existing ontology, the system can mathematically determine which sense is the correct one. Once the correct sense is identified, its synonyms and parent terms (hypernyms) are added to the ontology.
2. Text Mining (Web-based)
- Focused Crawling: Instead of a blind search, the system uses the ontology to generate specific queries for search engines and digital libraries.
- SVM Filtering: A Support Vector Machine (SVM) classifier acts as a "noise filter," discarding documents that contain the keywords but aren't actually about the target domain (e.g., filtering out "cell phone" articles when looking for biological "cells").
- Information Extraction: The system uses a Vector Space Model (TF-IDF) to rank and extract the most important nouns from the relevant documents.
- Mapping via kNN: To figure out where a new word belongs, the system compares the "context" (the words surrounding it in the text) to the context of existing concepts using the k-Nearest Neighbors algorithm.
Experimental Setup
The researchers used a specialized domain to test their framework: Amphibian Morphology.
- Seed Ontology: A manually created ontology from the AmphibAnat project containing 1,986 concepts and 570 properties.
- WordNet 3.0: Used as the primary lexical resource.
- Web Corpus: 10,000 documents from the Open Directory Project (ODP) were used to establish a baseline for word frequencies (IDF), while a focused crawl provided the domain-specific text.
Key Results and Insights
The study revealed a fundamental trade-off between discovery and placement accuracy:
- Lexical Expansion: The "Depth + Common Elements" metric achieved a 74% precision in identifying the correct word senses from WordNet. It was highly accurate at placement, with a 71.9% precision for adding synonyms to the correct concepts.
- Text Mining: The system was highly effective at finding relevant words, achieving 88% precision for word extraction. However, it struggled to place them in the tree, with only 19.3% precision for correct concept assignment.
The most significant insight is that the structure of the ontology itself is the solution to ambiguity. By looking at the "ancestors" of a word in WordNet and comparing them to the "ancestors" in the ontology, the system can navigate ambiguity without human intervention. Additionally, restricting extraction to nouns (the "L1N" method) significantly boosted precision in technical domains.
Conclusion
The primary limitation is the "narrowness" of structured dictionaries. Many highly specific biological terms simply don't exist in WordNet, rendering the Lexical Expansion approach useless for the most technical parts of the ontology. Conversely, the text mining approach's struggle with "placement" shows that context-based similarity (kNN) isn't yet a perfect substitute for structured hierarchy.
The authors suggest that the "Holy Grail" will be a hybrid system: using text mining to discover new terms from the web, and then using a combination of WordNet and automated reasoning to place those terms into the ontology.
Summary updated