Topic Detection -- Kea

 

gate-5.2.1-build3581-ALL\plugins\Keyphrase_Extraction_Algorithm\src\gate\creole\kea\Kea.java

 

 

Why top detection?

 

Generally, one post focus on one topic which may mainly from the first poster. We should reflect the relationship between posts which under the same thread. But now we just save every post into DB, but have no relationship between these posts.

 

So, the default level for detecting topics is Thread rather than one single post. As we can see, most posts just have several lines of context or less and that's not enough for detecting topics from. My idea is, for detecting topics, first, we should combine posts which under the same thread .

 

 

Kea is based on machine learning algorithms, so it need to be trained before you use it. For training Kea, you need to provide some sample files (.txt and .key pairs).