Language modeling and Vector space are so similar, pretty much the only difference is term weighting, frequency vs probability. So which one is better? Is there any real data showing it, like it’s 5% better, or something like this, not just theoretical comparison? Because I don’t see how adding hundreds of irrelevant terms (even though with very small weights) into your query LM (on smoothing stage) will help you to achieve better result.
No comments:
Post a Comment