Patrick Herron

Mon, 23 Oct 2006 01:27:36 -0800

the fashion in my graduate program was to enumerate the myriad ways in which google stinks. some of those were decent but ultimately insufficient reasons. there were more reasons for why it was good, certainly. after a couple of years down the road building text mining systems, and after a long talk with a senior engineer at google, i can see they are heading the game theory-predictable way of the market leader: choosing product diversification over innovation. the windows search beta is more innovative than the google search. the market leader's best strategy is to say the winning course. ultimately google is very good for a user who has wide-open lossely defined needs with little background knowledge or a user with background knowledge of a subject who has a ton of time to waste. you can see google's glaring shortcomings when you work for a company in a highly specialized field that uses a google box to provide their search. it's all in how google judges quality (via PageRank). morville is right. you need to add faceted classification menus for searches just to get started. clusty already does this and microsoft is going to bring theirs forward. meanwhile google is gobbling up competitors rather than innovating. i'm not sure the answer to your problem is with disambiguation. if you add contextual information you might, even in your case, send 'seattlest' farther down the relevance list and push 'seattle st' higher. adding context is crucial, but managing possibilities is always better than eliminating them. after all, IR systems and their descendants can do nothing more than make suggestions to users. in this respect google has made the right choice by showing the possibilities in different groupings. maybe you could argue that google is doing the faceted search *better* than vivisimo/clusty or microsoft....

Seattlest: Findability, or Why Google Still (Kinda) Sucks

Patrick Herron