Kongressbeitrag zum International Congress of Psychology
Kaser, Armin, Kolar, Gerald & Sachse, Pierre (2008). Google’s Sense of Semantics.
International Journal of Psychology, Abstracts of the XXIX International Congress of Psychology, Berlin (Germany), Volume 43, Issue 3/4, p. 145.
Transkript:
Introduction
Modern search engines display parallels to the functioning of the human brain (cf. Griffiths, Steyvers & Firl, 2007).
Semantic distance may be inferred from the amount of hits which two terms obtain in a query. Thereby, semantic networks can be automatically constructed.
The present study aims to show:
- which search engine may be the best for the construction of full semantic networks;
- if web catalogues are better suited than search engines for the purpose of constructing semantic networks.
Whereas in search engines a robot autonomously includes new websites into the query, a web catalogue derives from the efforts of human beings compiling the information.
Methods
Starting point of the study was a natural (i.e. “human”) data set amassed over several years. About 10,000 times two terms at a time have been selected from this data set in order to correlate their seman-tic distance with that of the search engines.
Phase 1: Semantic distance in the data set
Two terms being directly connected in the data set are semantically nearer than two terms being only connected via other terms in be-tween.
Phase 2: Semantic distance in the search engines
it two terms are entered into a search engine, they together obtain a certain amount of hits. Should those two terms be semantically near, they obtain more hits than terms that are more distant.
Phase 3: Correlation “semantic distance in data set — semantic distance in search engines”
The correlation gives information an how well search engines may be suitable for automatically constructing semantic networks.
Results
All results are significant.
Discussion
– Results by categories
The examined search engines differ in their suitability of constructing
semantic networks.
The best results are:
– Search engines vs. web catalogues
Web catalogues conceived by humans clearly obtain better results than search engines which have their indices automatically compiled by bots. However, search engines may have an advantage due to their substan-tially (arger indices if the required terms are quite rare and therefore not included in web catalogues.