Who Wants To Beat Google?

If you try your hand at machine learning, you learn very quickly that 95% of your amazing bright ideas don't work. The thing you learn by reading about popular algorithms isn't how ingenious they are - their popularity tells you the key fact that they often actually work.

Google is an AI company, they just don't advertise the fact. It's quite possible that someone in Google already tried an approach like this, and found that it didn't really work. A lot of people had similar ideas about the Netflix Prize and it didn't work for them.

I don't mean to sound too discouraging here. It's just that brilliant new notions for data you can integrate into your AI algorithm (and search is an AI algorithm) more often than not turn out to simply... not... work. So you don't get enthusiastic about an idea like that because it sounds good - you really do have to try it. If it turns out to work great in practice, like PageRank, then you might have a Google killer on your hands. It's just that the practical performance is hard to guess by eyeballing what a good idea it sounds like.

I disagree with a lot of this article - firstly, I wouldn't call Microsoft extremely diversified. Sure - they have a lot of products - but the cornerstones are Windows and Office.

Microsoft is fixated on their platform - Windows, Office. Almost every other play Microsoft has is really a funnel for revenue into these (and their closely associates). Microsoft's play is all about "seamless, easy integration". That is, you get SQL Server, it runs on Windows, it integrates nicely into VS.NET - which is integrated into your IE browser.

To innovate, Microsoft needs to move beyond this. They may have the best engineers in the world, but it's hard to break away if the mentality around these products persist.... Everything needs to work with Outlook, or Live Messenger, or search, or whatever. Microsoft is all about pre-canned and all the advantages and disadvantages that brings.

On the other hand, these are hugely profitable products for Microsoft. It's hard (and perhaps reckless) to let go of this as a strategy.

In my view, Microsoft needs to break itself up a little in order to create niches and to innovate. This means - don't integrate everything - open things up - attack markets that aren't already Microsoft mainstays.

The microformats and social network integration as an idea is fine - very semantic web - but Microsoft simply couldn't execute it at the moment. It would get tangled up in this machinery.

I'm really interested in how the personalization stuff and link graph are merged at Google. Part of what we're doing at Directed Edge is merging graph-based (PageRank and HITS-like algorithms) and matrix-based (collaborative filtering) into a single model that will allow us to do information filtering for recommendations. Once you've got both of them loaded up onto the same model you can do some interesting things clustering wise. One of the biggest problem with typical matrix clustering, i.e. what Google News uses, is that you're limited in trying to move out to neighboring clusters. If you combine both of them into a single graph model you've got more predictive power in terms of what you expect for a user to know about and what you expect to be new information for them.

> If my mom does a search for something, or if I do a search for something, Google treats us exactly the same. My mom loves crafts, and I love software development. We should get completely different results.

This is actually not true. Different users do get different results, not only based on whether they are searching using their local google site or e.g. google.com, but also based on their search history.

For example, when I searched for GAC (I'm signed in with my google account), the first result I got was:

http://en.wikipedia.org/wiki/Global_Assembly_Cache

Imagine the frustration if you are a software developer and are actually searching for Geological Association of Canada and Google/MS only returns for Global_Assembly_Cache because of your 'profile'.

Fail.

The author might want to take a look at some of the existing clustering search engines (e.g. Clusty, http://clusty.com/). If that could somehow be bound to a user's social network, then it might, maybe, possibly could work.

What was his actual idea?

Integrate more personal information into search?

That is already being done to some degree, not being done to a bigger I would think has to with complexity of algorithms and lack of reliable/valuable data?

Doesn't Google learn peoples interests with the personal search?

The approach explained in the article is similar to the one used by Popego, the "interests graph".

If you use Google Bookmarks it supposedly does take such things into account when searching.

PageRank = Wisdom of Crowds

AHHH GAWD i had to stop reading that to maintain my self respect. This guy does not know the meaning of "logic" and unbiased trials for comparison.

In a nutshell: He actually entered 'dance club' in google and was ANNOYED by the lack of RELEVANCE AND DETAIL. No no, rather than take .45 seconds to actually THINK about refining his search query to "popular dance clubs in miami" he proposes that THE SEARCH ENGINE should sift though all of his billions of "social utilities" to figure out he was planning a trip to miami and so OBVIOUSLY he was looking for "dance clubs in miami. But wait, there's more. His Google-killing idea further specifies that the results should list clubs that 'his friends' have recommended and like. So forget about putting "popular" in the search term, and forget that any NORMAL PERSON would just use yelp, no no no no

the answer is to blame Google for its blatant disregard for peoples search stupidity and FIX EVERYTHING about your query, because you know...its brain is bigger than yours. duhhhhhhhhhhhhhhhhhhh

Good god that was bad. Just for kicks, uh yeah, its called the semantic web, people are working on it, and uh no, you did not invent it.

WOW