Comments on: Database query and ranked results

By: John Woods

John Woods — Fri, 30 Jan 2009 19:27:07 +0000

I’ve been thinking about similar things. How would Google Biology–or, rather, Bioogle, or Biologoogle, or something clever–look?

I’m not sure I have any answers yet. It’s a tough problem. But that’s why I’m in systems biology. =)

Greetings, by the way. I just found you through Google Reader.

John

By: Mr. Gunn

Mr. Gunn — Wed, 28 Jan 2009 01:36:42 +0000

You and I are looking for similar things, Pawel.

I think people are just starting to realize that, in this new paradigm of Big Data, it’s more likely that you’ll get too many results than not enough. Being able to comprehend the results through sorting, filtering, and visualizations that allow you to make sense of everything and how it’s related is becoming increasingly important.

By: Pawel Szczesny

Pawel Szczesny — Thu, 22 Jan 2009 22:06:04 +0000

Kay, thanks. I agree – my examples are about sorting, not doing fuzzy search. But I’ve made them such way, because I was afraid running into a need of explanation what I mean by “abstract _more related_ to disease X”, which I’m not sure I could do. I just couldn’t come up with a fuzzy search example that wouldn’t in my mind raise a question “why one would want to do such thing?”

On the other hand, both approaches seem to me closely connected. One thing is a filter (and sorting options). SRS does it quite well, and probably that’s about it among biological databases. The other thing is fuzziness (and weighting of filters). I believe both should be implemented and available as user switchable options – not only to improve search, but also (as you’ve pointed out) to avoid situations that the server does more than one has asked it to do.

By: Kay at Suicyte

Kay at Suicyte — Thu, 22 Jan 2009 21:12:09 +0000

I don’t quite see the connection between your query examples and your description of an ideal search engine. I am not mistaken, all of your examples could be dealt with by SQL-access to a relational version of Pubmed. The business with fuzzy searches, term weights and sliders would be useful for solving different questions, such as finding papers that are more related to disease X than to disease Y.

I am generally not a friend of fuzzy searches and weighting, but rather prefer pure boolean searches that retrieve only those entries that fully match my search criteria. I must admit however that there are situations where a certain fuzziness comes in handy.

What I really hate is PubMed’s habit of searching for other (similar looking or similar sounding) term than those I specified – with the pathetic excuse that there are more hits. I guess that most people consider this a useful feature, but I don’t. If there are no valid matches to my query, I can stand the answer ‘no hits’. There is no need to search for the next best thing just to be able to report something.