1. Learning Taxonomic Relations from Heterogeneous Evidence Philipp Cimiano Aleksander Pivk Lars Schmidt-Thieme Steffen Staab (ECAI 2004)

2. Purpose  To examine the possibility of learning taxonomic relations by considering various sources of evidence  Main aim: 1. To gain insight into the behavior of different approaches to learn taxonomic relations 2. To provide a first step towards combining these different approaches 3. To establish a baseline model for further research

3. Introduction  Taxonomies or conceptual hierarchies are useful in many NLP applications.  However, the development of suitable ontologies is time- consuming.  Automatically acquiring ontological knowledge is required.  The approach proposed in this paper learns taxonomic relations (is-a relation) by considering four different evidences: 1. Hearst-patterns matched in a large corpus 2. Hearst-patterns matched in WWW 3. WordNet 4. The ‘ vertical relations ’ -heuristic

4. Introduction  Goal: Learning is-a relations in tourism domain  Training Corpus: Domain-specific:  http://www.lonelyplanet.com http://www.lonelyplanet.com  http://www.all-inall.de http://www.all-inall.de General:  British National Corpus  The ontology for evaluation: A tourism reference ontology modeled by ontology engineer. A few abstract concepts are removed. 272 concepts, 225 direct is-a relations, and 636 non-direct is-a relations

5. Hearst Patterns  Lexico-syntactic patterns proposed by Hearst (1992). N such as N 1, N 2, … such N as N 1, N 2, … N 1, N 2, … and other N N, (especially | including) N 1, N 2, …  From these patterns, we could derive is-a(N i, N).  Numbers of Hearst-patterns between different terms are recorded and normalized to 0~1.  Different thresholds are set and experimented.

6. Hearst Patterns

7. WordNet  WordNet is not “ unstructured ” source of evidence.  However, it is general and domain-independent.  One term may have several senses and there may be more than one hypernym relation between two terms.  Two different strategies are used: 1. Normalizing all hypernym paths between two terms: 2. Considering only the most frequent sense of t 1

8. WordNet

10. ‘ Vertical Relations ’ -Heuristic  Given t 1 and t 2, if t 2 matches t 1 and t 1 is additionally modified by certain terms or adjectives, the relation is-a(t 1, t 2 ) is derived.  Ex. is-a HEURISTIC (international conference, conference)

11. World Wide Web  Google API (http://www.google.com/apis/) is used to count the matches of certain Hearst- patterns in the Web.http://www.google.com/apis/  The sum of the number of Google hits over all patterns for a certain pair (t 1, t 2 ) is normalized by dividing through the number of hits returned for t 1.

12. World Wide Web

13. Combining Evidences

14. Conclusion and Further Work  A simple combination strategy improves the results.  It remains further work to find out if other sources of evidence could be integrated into this approach.  It could turn out to be useful to only consider domain-specific text collections instead of a general corpus such as the BNC and to consider only pages in the World Wide Web related to the domain.  It remains as a challenge to determine the optimal strategy to combine the different approaches.  In order to apply machine learning techniques for this purpose, it is necessary to cope with the high number of negative examples.