Slashdot Mirror
The Un-Google - The Search Competition
WinEveryGame writes "The Economist is running an article on the state of the competition for Internet Search. While Google clearly dominates, and continues to have positive momentum, its leadership is still vulnerable. The search-engine battle is not over yet." From the article: "In terms of momentum — mass times velocity — Google's lead indeed looks daunting. It has by far the most mass, with an American market share of 43% as of April, which reaches 50% counting AOL, an internet property that uses Google's search technology. This compares with 28% for Yahoo!; 13% for MSN, which belongs to Microsoft; and 6% for Ask, which is owned by IAC/Interactive Corp, a conglomerate of about 60 online media brands. Google also has velocity: its market share grew by 17% in the four quarters to this spring, whereas Yahoo! and MSN both lost share. Only Ask has more velocity — its share grew by 35% — but then again it has little mass."
There are some customers (government/military included) that are aware of the two concepts of precision and recall. Before you groan and skip this post because you recall those words from all classifying algorithms, you should take note that there are two stages we have yet to meet in this respect.
One is simply improving precision without sacrificing recall. When I search for 'horn' in Google, how many of those searches are relevant? I was thinking about a French horn (instrument) and the first link brings me to a society about them. The next three links, however, do not. You might say, "Well, gee, you should have put 'French' in your search" but is this really necessary? So there is some money to be made in "learning" search engines that tailor themselves to the user or perhaps the results could be displayed intuitively in domains of knowledge (a la Clusty). So that I can select a node that applies to the correct searching term and see all results returned below that. Have you ever wished to view your search results in a format other than a linear display of ranked results? The documents are related in more than one dimension, you know. As computing power increases, I suspect there will be room to display them in two dimensions (heat/area mapping, nodes & vertices on a plane) and three dimensions (spatial 3D engines with nodes & vertices in space).
The second stage is giving the user the power to adjust precision versus recall. Even a graphical interface that shows the F-measure relationship between precision and recall would be helpful to consider in the search engine wars. Say you give the user some control through a slider AJAX interface of a threshold ß. But the threshold isn't simply the "Google score cut off" or even a term frequency cutoff. Instead, it's applied to be a "relevance" threshold. You would score relevance by fingerprinting frequency, specificity, clustering and other useful tools by using a domain ontology or taxonomy.
Another big thing that is missing is identifying what kind of data you are searching. Social data? Scientific data? Historical data? etc. Perhaps I'm only interested in who's who to Stephen Hawking. I'd search for him and flip through nodes of separation from him to other people.
The current search sites also only tend to favor key-word regular expressions. What about searching with raw text or entire paragraphs? If you want to see an interesting demo of this, visit Collexis' Demo Site which alludes to a whole new kind of searching.
The key to entering the market as a competitor with Google is to pick up Google's slack and to try to pose yourself as a complimentary service to Google. Google is terrible at closed domain searches but amazingly efficient at open domain searches. You don't want to compete with them so fill a different part of the market. Google benefits from simple design, so go to an advanced flashy complex design. Most people aren't looking for that but the people that are have nowhere to go.
The Economist is alluding to potential leadership problems inside Google. Who cares? That's not going to be Google's downfall. Google's downfall will be an new intuitive way to search and the only thing that will prevent their downfall is if they buyout the company or bone up on the technology.
The search-engine battle hasn't even hit its stride.
My work here is dung.
It seems more and more when I try to find something on google all I get are a bunch of link farms. This morning I was trying to find a bike jersey for a friend of mine and on the first page of results, it took getting to the second page to find any actual results. I did much better using Yahoo and found what I was looking for on the first page of results.
This is just one example, but it happens constantly...
Well, actually, almost all current search engines suck. There is waaaaaay too much noise in the results they return. Let's say I'm doing a search for "product X", search for it in Google and what do you get? Several links to ebay (which may or may not be current), tons of links to various "rate it" sites such as epinions/nextag/msn/etc, and maybe a few smatterings of other sites mixed in. Typically the manufacturers own site won't even appear in the first couple of dozen results!
So basically, I agree with the general position of the article, that there is still a TON (actually several tons) of work to be done and room for someone else to move in with a truely superiour solution. While it's great that Google is tinkering with lots of other technologies, I wish they'd actually make some real advances in their core business (and actually, I'm slowly starting to come to grips with the fact that their core competency may not be searching, but really it's in creating low latency widely distributed computing infrastructures). For all the years and the massive sums of money, my search experience is not significantly better than it was 5 years ago.