Understanding Intentions and Microsoft Search Personalization
A woman says to a man, “It’s cold in here.” The man puts his arms around her and holds her. He could have turned up a thermostat, or brought her a sweater, or asked her if she wanted to go somewhere warmer. But imagine that the man and woman are in a relationship and have had that conversation a number of times in the past, and the intention behind the words was easily understood by both.
We can sometimes understand the intent behind certain words even though the words don’t actually match up well with the intentions of the people who voice them, though the intentions behind words can be difficult things to understand. Sometimes nonverbal communication that accompanies words can be helpful in interpreting them, though humans aren’t necessarily that good at reading nonverbal communication either.
Sometimes past experience can be informative in understanding what certain words might mean, like the man and woman in my example above.
If human beings can grow easily confused about the intentions behind words, how well can a computer understand the intent behind a handful or less words in a query at a search engine?
That’s a challenge that search engines face as they try to strive to serve one set of search results to one visitor, and a different set of results to another based upon things like location or preferred choice of language or past searching and browsing history.
Erik Selberg provides his thoughts on an upcoming Microsoft search upgrade, in his post “first impressions”. The codename for the upgrade appears to be “Kumo,” which may also end up being the new name of Microsoft’s search. The name “Kumo” can mean “spider” or “cloud” in Japanese.
Dr. Selberg is one of the founding members of Microsoft Search Labs, and worked on helping to create and define the algorithms behind the relevance of MSN Search and Live Search. His thoughts about a possible screen shot of upgraded search results from Microsoft are interesting, especially this one:
Then there are related searches, and then a very interesting section: “your history.” So, looks like they’re finally going to start to show you your previous searches and possibly let you do something with them! It’s about time some of that work started to see the light of day!
Interestingly, a patent application filed this week from Microsoft lists Erik Selberg as the inventor, and gives us a glimpse at one kind of personalization from Microsoft that might someday be seen the light of day. Whether or not it’s the personalization that we see with Kumo is unknown. We might also see something like Microsoft’s U Rank, which shares many similarities with Google’s Search Wiki.
Best Answers For Different Searchers
The patent filing is worth looking at more deeply, regardless of whether it’s part of what Microsoft rolls out when they release their search upgrade.
The basic premise is that when two different people are searching for the same query term, chances are that the answers that they are trying to find or the sites that they might want to see are different, and that a search engine might be able to help each of those searchers find what they are looking for based upon past experience, and past searches and search result selections.
If someone has performed the same, or a substantially similar search a few times in the past, and has a tendency to choose a particular result in response to those searches, then that site might be seen as a desired result for that searcher in the future. A similar search might be, for instance, one in which 8 of the top 10 search results are the same for each query term.
What if, in future searches for that searcher, the site that they tend to pick is displayed in a different manner, such as bolded, or highlighted, or shown with a border, or in a different font than the other search results? What if it is placed at the top of the results instead of where it might be normally? The idea behind the different presentation being to make it easier for the searcher to pick out that site as one that might be preferred.
The patent application is:
Presenting Result Items Based Upon User Behavior
Invented by Erik Warren Selberg
Assigned to Microsoft
US Patent Application 20090063460
Published March 5, 2009
Filed August 31, 2007
The concept of “Personal Definitives” is at the heart of the patent filing.
A personal definitive is the “best” answer to a given search query as it relates to a specific searcher. Information collected to identify a searcher’s personal definitives include such things as:
- Cookie or login information,
- Terms or phrases used to search upon in the past,
- Search results or a representation of those results in response to queries, and;
- Selection counts for previously presented search results.
If you tend to search using the same or a substantially similar query term or phrase, and tend to select the same page or pages in response to that search, don’t be surprised if at some point it might be highlighted or bolded or placed at the top of the search results in the future.
While this is something we might see at Microsoft, it wouldn’t be surprising to see something similar spring up at Google or Yahoo in the future as well.
The question though is, whether past searches are a good indication of intent for searches in the future?
Sometimes the statement “It’s cold in here,” isn’t an invitation to a hug, but rather a request to turn up the thermostat.