Google and Guha: The Semantic Steamroller
April 17, 2009
I hear quite a lot about semantic search. I try to provide some color on selected players. By now, you know that I recycle in this Web log, and this article is no exception. The difference is that few people pay much attention to patent documents. In general, these are less popular than a printed dead tree daily paper, but in my opinion quite a bit more exciting. But that’s what makes me an addled goose, and you a reader of free Web log posts.
You will want to snag a copy of US20090100036 from our ever efficient USPTO. Please, read the instructions for running a query on the USPTO system. I don’t provide for free support to public facing, easy to use, elegant interfaces such as that available from the Federal government.
The “eyes” of Googzilla. From US20090100036, Figure 21, Cyrus, in case you want to see what your employer is doing these days.
The title of the document is “Methods and Systems for Classifying Search Results to Determine Page Elements” by a gaggle of Googlers, one of whom is Ramanathan Guha. If you read my Google Version 2.0 or the semantic white paper I wrote for Bear Stearns when it was respected and in business, you know that Dr. Guha is a bit of a superstar in my corner of the world. The founder of Epinions.com and a blue chip wizard with credentials (Semantic Web RDF, Babelfish, Open Directory, etc.) that will take away the puffery of newly minted search consultants, Dr. Guha invented, wrote up, and filed five major inventions. These five set forth the Programmable Search Engine. You will have to chase down one of my for fee writings to get more detail about how the PSE meshes with Google’s data management inventions. If you are IBM or Microsoft, you will remind me that patents are products and that Google is not doing anything particularly new. I love those old eight track tapes, don’t you.
The new invention is the work of Tania Bedrax-Weiss, Patrick Riley, Corin Anderson, and Ramanathan Guha. His name is spelled “Ramanthan” in the patent snippet I have. Fish & Richardson, Google’s go-to search patent attorney may have submitted it correctly in October 2007 but it emerged from the USPTO on April 16, 2009, with the spelling error.
The application is a 33 page long document, which is beefy by Google’s standard. Google dearly loves brevity so the invention is pushing into Gone with the Wind length for the GOOG. The Fish & Richardson synopsis said:
This invention relates to determining page elements to display in response to a search. A method embodiment of this invention determines a page element based on a search result. The method includes: (1) determining a set of result classifications based on the search result, wherein each result classification includes a result category and a result score; and (2) determining the page element based on the set of result classifications. In this way, a classification is determined based on a search result and page elements are generated based on the classification. By using the search result, as opposed to just the query, page elements are generated that corresponds to a predominant interpretation of the user’s query within the search results. As result, the page elements may, in most cases, accurately reflect the user’s intent.
Got that? If you did not, you are not alone. The invention makes sense in the context of a number of other Google technical initiatives ranging from the non hierarchical clustering methods to the data management innovations you can spot if you poke around Google Base. I noted classification refinement, snippets, and “signal” weighting. If you are in the health biz, you might want to check out the labels in the figures in the patent application. If you were at my lecture for Houston Wellness, I described some of Google’s health related activities.
On the surface, you may think, “Page parsing. No big deal.” You are not exactly right. Page parsing at Google scale, the method, and the scores complement Google’s “dossier” function about which Sue Feldman and I wrote in our September 2008 IDC client only report. This is IDC paper 213562.
What does a medical information publisher need with those human editors anyway?
Stephen Arnold, April 17, 2009
Comments
One Response to “Google and Guha: The Semantic Steamroller”
-
- Subscribe to Beyond Search
Feature archive
News archive
-
Vertical Searching does have some serious potential. Getting exactly what you want out of searching is the point. I found a website called Cazoodle.com that utilizes vertical search in one of its products that gets listings for apartments. The results were very accurate, relevant and abundant. One search I tried for a 1 bedroom apartment got me 1009 apartment listings for the Chambana area in IL.