Using Clusters of Concepts to Extract Semantic Relations from Standalone Documents
The extraction of semantic relations from texts is a hot topic. However, a large number of current methods are language and domain dependent, and the statistical and language-independent methods tend to work only with large amounts of text. This leaves out the extraction of semantic relations from standalone documents, such as single documents of unique subjects, reports from very specific domains, or small books.
We propose a statistical method to extract semantic relations using clusters of concepts. Clusters are areas in the documents where concepts occur more frequently. When clusters of different concepts occur in the same areas, they may represent highly related concepts.
Our method is language independent and we show comparative results for three different European languages.