Thursday, March 22

NLP search and resume-ing my file transfer

Powerset, Google, are you listening?

Today I ran across a frustrating example of polysemy - resume. I was searching for some help getting my linux server to resume file downloads so I searched for

vsftpd file resume

I got back Seth Vidal's Resume. Hmm, not exactly what I was looking for, but perhaps I'll give him a call and ask for his help. Maybe it's a sign from Google. Ahh well. A few more iterations and I found what I was looking for, but even in several of my refinements it seemed impossible to entirely escape Seth.

I thought this was a good example of where I wanted to disambiguate my query -- I did not want someones resume, I wanted to resume my file transfer. A good example of polysemy and where a better understanding of my query would have been nice.

Please, give me a way to disambiguate or do a better job of reading my mind.

1 comment:

  1. Do you think this is purely a problem of google not being able to differentiate between the possible senses of 'resume' could it be more that there aren't a lot of good results for the topic you were searching for?

    I notice that in google's index there isn't anywhere (apart from here) that the phrase "vsftpd file resume" appears. Google's algorithm usually works for these types of examples because it seems to rank query terms clustered together in the document more highly then the same words scattered through out a large document. When the query terms don't occur close together in any of the documents then google is forced to just give you what ever it can find.

    The findory version of your blog post is now the topic result for that query. My feeling is that this is just a case where either you used a poor query or the information you want isn't readily available. It might be out there but no-one has really bothered to link to it?

    It may be that a smart linguistic algorithm link powerset could disambiguate the query and provide better results, but I'm yet to be convinced...