Thursday, July 9, 2009

Information retrieval: conceptual problem

From my experience with grep for example, I come to a conclusion that the main problem of current search engine
interfaces (not grep ones as grep serves other aims) is that they lack one major feature. It is the feature of visualizing the search space or from other perspective -- a feature of search hints.

Suppose I want to search *something* about HTTP protocol (some very specific detail) and I know very little about HTTP (an artificial example). So for example, I even don't know which category does that particular detail belongs too.
Needless to say, that I don't know the term which I'm looking for. Now I'm stuck. I should have some starting hints, which would lead me to first query terms apart from HTTP. Because HTTP is very wide term in a tree of terms (like the tree root), I will spend hours and hours reading through millions of returned hits.

Instead, I would use a concept graph, visualized, which directly corresponds to the search index of a searching engine.
Now I can easily jump in a concise way over those categories getting deeper and deeper into HTTP topic and discovering
some previously unknown terms, which I can further search for. Thus narrowing down the search space with the parallel learning I would come to a result faster.

No comments: