MSN Search has the most powerful and promising searching technology. It is based on Neural Networks and NOT on an Algorithm (like Yahoo and Google).
Difference between Algorithm, AI Algorithm, and Artificial Neural Networks:
1) Algorithm is a flow controlled logic, that works "Perfectly" IF implemented "Perfectly". It cannot adapt, and rests on the human brain to develop logic.
2) AI Algorithm is the one, that can perform intelligent actions, based on situation and conditions. The overall flow controlled logic is always the same. It uses task completion algorithms like the "Hill Climbing" etc to accomplish a task.
3) It is based on the Architecture of human brains, implementing Neurons. Neural Networks do not use algorithms, but generate results on the basis of inputs fed to the network. The network is interconnected neurons with weighted links.
A Artificial neural network is a series of computers which are supposed to learn based on input provided.
Think about that for a second – a learning computer. One that just doesn’t follow rules assigned to it (which is what the more traditional algorithmic search engines like Google and Yahoo! do) but actually can learn from its results.
Essentially MSN search learns from input given to it. For example, if the search engine is told that Ebay is considered an authoritative site on online auctions, then when a person performs such a search they should see Ebay.com at the top of the search results.
Upon analyzing Ebay.com the search engine can then learn why it is considered an authority and apply that learning to other sites to see if they are also authoritative.
The biggest advantage of such a platform is the engineers at MSN can “train” the system to understand what is considered relevant and important and what isn’t. As time goes on we would expect to see MSN search become one of the most relevant of all the search engines simply because the system is designed to improve itself over time.
Of course like any search engine, MSN could be tricked. If we knew what those factors were, we could create a page which could be considered highly relevant, based on the MSN search criteria but would in fact be a garbage page. However because of its ability to learn, the system could quickly adapt to such spam content and readjust rankings “on the fly” to filter out these bogus results.
Another advantage to MSN is that the system should be infinitely scalable. Which means as the use of the search grows, it should only be a matter of introducing new hardware, or requirements into the system, having it adapt to the additions and begin using them as if they’ve existed all along.
Therefore, as new spam techniques are developed, its simply a matter of training the system to watch out for the new technique, flag it as potential spam and even potentially react to it by filtering all sites using the new technique.
By now you are probably saying “holy cow that type of technology must use a ton of resources” and you’d be correct.
The amount of computational power required by such a system would be immense. Just the storage capacity needed to store what the system has “learned” would have to continue to grow. In addition, the system also has a great crawler out indexing more and more content all the time.
It’s not your typical algorithmic based engine. With most algorithmic systems, the ranking algorithms are finite in size. With this system, one would expect the Neural Net to continue to grow as new pathways are created.
Consider this structure as similar to a human brain – as we develop new thoughts and ideas, new synaptic pathways are developed linking areas of the brain to other areas where links previously didn’t exist. Essentially this is what a Neural Network does. While its pathways may not be physical, it does nonetheless develop relationships between previously unrelated sections.
Therefore, the engineers at MSN have developed ways to “shortcut” the requirements for ranking. Essentially they have said “sure there are over 500 factors determining the page quality, but in this category only 150 are used, therefore you can use the same 150 associated with this category.”
Overall, as long as Microsoft can continue to support such a system, I would think that it could win out in the “search engine wars.” The system appears (at least on paper) to be superior to algorithmic based systems, and appears to be able to adapt more quickly to changes on the web because it doesn’t have to wait for an algorithm change to adapt, it only has to learn of the change and apply itself.
Edited by CaptainRon, 04 December 2006 - 10:32 AM.