Forum Moderators: mack
I am very impressed with the improvements in MSN search quality in the past 2 months.
OK msn are working on it and we can only hope they can get it right in the end but msn is not relevent and you can do almost any search request on its engine and you wont get relevent serps, they are miles off!
I just wont use the msn search now its so bad, i only look in on it to see if its been fixed yet!
I use Msn an Google side by side everyday, for stuff ranging from
very techie programming stuff to legal matters, cos i wear two hats
Techie an auditor/accts
and I must tell you that its rare for either search engine to let me down, in both cases, relevance is remarkabley high, basically i'm tell you i do real work using both, not just doing widget searches
Google & Msn both have superb algorithims
Google has a natural advantage in depth of files, plus I suspect a lotta folk naturally submit to google an simply ignore other search engines
That's actually the bottom line. Google doesn't have a whole lot of spam at the top of competitive SERPs. It's just directories rank better because of IBLs. When you think about it, it's pretty amazing to pull a relevant result from billions of pages in a matter of seconds.
It's not just recent stuff either. Just one example, I had a problem with gentoo portage a few days a go. Stuck the error message into google...zero results. Tried msn and around a dozen good results (a few from about 2 years ago). If I hadn't tried msn and stopped at google I would have still have a slightly borked server but thanks to msn it's all fixed.
Google's failure to list (and removal of) good pages from its index is making it continually worse, whreas msn's quick spiderng and fast indexing is taking it leaps ahead.
thats why i prefer google because i know i can find it there faster.
I think he means that we give too strong a preference to geolocal information. We've also heard a lot of dissatisfaction from people here on Webmaster World about this, so we realize it doesn't always work perfectly. Like anything else, if we take it out, other things get worse, so it stays in, but we are working on improving the precision of it.
Time will tell, but relevancy is one small part of search, albeit an important one that google overlooks at times. MSn needs to get better at scoring pages.
The reason I use MSN is I *know* there are so many sites missing/bannned/etc from Google that I often go to MSN now just to see a wider more complete selection of what is available.
In fact the results often seem very similar.
I still don't like the local targetting but at least that does seem to have improved over the summer.
Keep at it MSN! I installed the MSN search bar last week and discovered this MSN live thing. I don't like that much though. It is slow and I see inferior results there.
Give me MSN.com any day.
So just who will be fed these Live results and why?
Try "adventure travel" again. Not even close in terms of importance and relevance. I have no idea what MSN is doing or looking for, and I programmed with many Microsoft products for 20 years and continue to... Many of their programming tools are great, but the search engine algorithm of late just seems off.
And I really like to see competition.
"I'm looking at them right now, and I'm not seeing any differences"
I am in Europe (Madrid) and the results are definately different.
For a travel search I monitor about half the first page are different sites to MSN.com's first page and those that are the same are in a different order.
Does not seem to be geo-targetting either.
I have a very successful website, with thousands of visitors per day.
Most of the visitors are Americans and most of the content is written by American professionals. The website has many top positions for important keywords on other search engines (not doing too well on msn lately, except during the recent update which has been rolled back).
I live in Belgium, but this website is hosted in the States.
If it would have been hosted in Belgium and with geotargeting, American surfers would not find the way to this information which however is very useful for them.
My opinion is that geotargeting is wrong because of the globality of the Internet. What if an American website owner purchases hosting in for instance Holland because he can get a good deal there?
I think targeting based on languages is ok, based on location of the server is wrong.
As Microsoft has often stated, we using "Machine Learning" to generate our algorithm. This means that we don't actually write the algorithm directly; we write a program that "learns" from examples, and then that program writes the algorithm (which we call a "net"), trying to get the best result it can for the training examples. We test the new net on a separate set of examples (so it can't just memorize the answers), and if we like the performance, we ship it.
When we write the training program, we give it as much information as we can think to. For example, it knows how many links are on a page. It knows how many words are on a page. It knows what language we think the page is written in. etc. Each of these tidbits of information is called a "feature." (In the sense that a face has features, not the way a new product has features.) Our system considers many hundreds of features.
One of those hundreds of features is the country we think the site is hosted in.
Based on the training data, the program is free to decide how important a given feature is. As you surmise, it thinks language is really, really important. As you also know, it thinks location is pretty important too. As we have improved the system in other ways, it has come to rely less on this, but at present, it still gives location a lot of weight. (Note, though, that it does not "ban" a site; all it does is give a bonus or penalty.)
We do not try to adjust these weights manually. Among other things, since a net actually uses combinations of features, it's not humanly possible to do that, but even if we could, any naive change would make the measured result worse. That is, a change that fixed the results for one site would likely break them for dozens (or hundreds) of others.
As a result, when someone reports a problem (such as you have), we cannot simply go fix it. Instead, we have to try to figure out why the net isn't "seeing" the right answer. We can add new features (or improve old ones), and we can add training examples that show right and wrong results, but we can't just change a few lines of code and fix a specific result.
When it works, this technique (machine learning) has tremendous power. The computer literally learns from its own mistakes, and it learns rules that no human being would ever have found. When it doesn't work, though, it can be very frustrating -- for all parties involved.
That said, we actually do have some ideas for improving this particular problem (geolocal preference). I realize it's hard to be patient and trust that we're working on it, but that's what we have to ask you to do.
Sorry again if you thought I was making fun of you.