It seems that all big search engines are now focusing on personalized search. One may be lead to believe that there isn’t much evolution left on general indexed search, but the recent boom of alternative search engines disproves it. Google has been researching on this subject for a long time now and their results are being slowly integrated with the main engine. Now all users with a Google account are already receiving customized results. More on Google efforts on this Read/WriteWeb article. But is personalized search real good for the users?

Alternative Search Engines

Recently there’s been a impressive growth on the number of alternative search engines. They’re alternative because, instead of trying to compete with the popular please-everyone search portals, they focus on specific features, niches, sources, aiming for The Long Tail of web search. Some depend on personalization, normally in an explicit way, others don’t. Lijit for example, works only on your and your network’s content, over personalizing the results. On the other side, Rollyo enables you to specify the sources and topics you want to filter our searches. In Spock you can search people and on Feedster you can search RSS feeds’ channels. There’s even a blog dedicated to this kind of services: Alt Search Engines. So I predict there’s more into search niches than we’re still used to think.

There’s no bullet-proof search

On a ideal search, results would come exactly by their relevance. However, while this seems simple from a machine point of view, humans are far more complicated. And the human language, as its creators, has a great degree of complexity and ambiguity. Software has to literally guess what are the user thoughts. One simple example is the query word python. For a normal person, that word would mean nothing different that a specie of snake. But from an computer literate point of view, it would probably the known programming language. There’s hundreds of situations like these. The only way to present the right results is to know a little about who queried. Is he a biologist? Did it searched for other snake species? Or is he into interpreted programming languages and searched for PHP and ruby before?

We already found shortcuts to let the software know what do we want to get. We usually help a bit by querying for ‘python programming’, to ensure our results. But there are certainly searches for very underground links, that could only be found inserting that specific keyword we just didn’t though about.

Is there hope?

Even when all the information about the user searches is gathered, mining all that information and integrating it into the engine ranking system isn’t easy. And humans are unpredictable too, and computer programmer might also want to know about snakes and other animals. Can there ever be a piece of software that does a job that well, or will search personalization only lead us to get exact results sometimes and completely confuse us when we change a bit our behavior?

One of the approaches I found most promising is the categorization of results. Analyzing specific tags, the search engine can group a type of results in ‘Biology’ and other in ‘Informatics’ (using the python example). This way, the user could simply ignore one group and look through the other for what he’s looking for. However, machines have an hard time trying to figure out what a website is all about, while humans can do it much faster. Folksonomies are the best solution for these case.

Can you trust the machine?

The last issue on personalizing search results concerns the user privacy. In order to present the best results the machine must have a large number of information about the user. This is not a problem when we know it’s only for our benefit, but we never do know right? A recent study on CNET compares five main search engines on their privacy policies. Ask.com comes out first by only saving data for some hours and then deleting it, although this prevents all kind of implicit personalization of their results. And maybe that’s the right path. Only the future can tell…

If you enjoyed this post, consider subscribing to my RSS Feed