Edison, Ask.com's New Algorithm Options

[rustybrick]

I got word that Ask.com is migrating Teoma and Direct Hit to create a new search algorithm named Edison.

More details at:
http://www.seroundtable.com/archives/013086.html
http://searchengineland.com/070412-154815.php

[bragadocchio]

I've been sitting on a couple of patent applications that recently came out from Ask that sound like they fit the description that you've been given, Barry.

I guess I should pay a little more attention to those.

Great picture with you and AG!

I have a brief overview on one of their recent patent applications here, and it seems like a very good match for what you've been told, such as the tagging of pages based upon queries and clicks:

Ask.com Patent Application Discusses Responding to User Queries
http://www.seobythesea.com/?p=561

I'm in the midst of writing about some interesting Google stuff that's come out, but I'll see if I can share more about the other patent applications from Ask after that.


Edison is an interesting choice of code names. Interesting that it isn't Princeton instead.

Maybe I'll skip the Google stuff momentarily and look at the other indexing related patent application first.

[bragadocchio]

This one sounds like it matches what you were describing, Barry. Of the three recent patent applications I've seen from Ask, one came out of Princeton, one was predominately worked upon by folks in Italy (and focused upon image search and clustering), and one has two authors from Edison, NJ (amonst others).

Here are some details for anyone else who wants to work through this, and break it down into plainer English:

System and method for responding to a user reference query
Inventors: Scot Gregory Zola (Edison, NJ),Eric J. Glover (North Brunswick, NJ),Rohit Mittal (Somerset, NJ), Apostolos Gerasoulis (Edison, NJ), Stephen Lee Orr (La Jolla, CA),Gary Chevsky (Alamo, CA)
United States Patent Application 20070078842
Published April 5, 2007
Filed: December 21, 2005

Abstract

The invention provides a system and method for providing an output in response to a user reference query. A user reference query is received and an answer to the user reference query is determined. A unique entity identifier (EID) for the answer is determined and used to query a reference system to determine one or more reference elements, e.g. a reference summary, a picture, or a video, for example. The one or more reference elements and a text of the answer are provided to an output provider. The output provider combines the one or more reference elements and the answer text into a single output. In one exemplary configuration, the single output is a graphical user interface transmitted as an extensible markup language (XML) file, which integrates the answer text and the one or more reference elements.

[bragadocchio]

Roughly, here is the process that this patent application describes:

1. A query is received
2. A unique identifier is assigned to each concept in the answer
3. Reference elements related to those concepts are collected
4. The answer and one or more of the references are returned in response to the query

What is a User Reference Query?

Rather than matching keywords, this approach means providing a factual answer to a "user reference query," and then supplying reference documents that go with the answer. The answering technology is described further down in the patent application

User queries may be divided into categories. Some examples:

1. Transactional, or queries involving web mediated tasks. These can be things like buying an airline ticket or other e-commerce activities.

2. Navigational - involving locating a site on the network, such as the homepage of the 2006 Olympics.

3. Informational - These can be in the form of a question. The example they provide is "Who won the Triple Crown in 1934?" though they state that it doesn't have to be in the form of a question. This is a "user reference query" in that searchers are looking for answers, and for references to those answers.

They point out that a user reference query often doesn't contain keywords that are included in the text of an answer to the query. The winner of the Triple Crown in 1934 (a baseball term meaning that a batter has the highest batting average, the most home runs, and has driven in the most runs) was Lou Gehrig. The answer wasn't in the query, but it can be used to find references to show to searchers.

The answer to a user reference query may also change over time. The answer to "Who is the President of the United States" may change every four years, for example.

The answer to our triple crown question was: "Lou Gehrig won the triple crown in baseball in 1934". Answering technology would provide that answer.

[eKstreme]

System and method for responding to a user reference query

Thanks for that Bill. Here is my interpretation.

The patent describes a two step process. To illustrate it, let's suppose you want the biography of the CEO of Apple, but don't know who he/she is. The usual method of getting what you want is to figure out who is the CEO of Apple, and then search for biographies. The Ask patent describes a shortcut in these situations.

The Edison system, as I understand the patent, receives a query, say "Apple's CEO". The first step is to actually answer the (implied) question and tell you the name you want is Steve Jobs, an perhaps give you a bit of a biography. This is the answer phase. The answer contains one or more concepts that are extracted. In our example, concepts can be Steve himself, the year he joined, or the story of him joining, etc. All these concepts are linked to the answer, and therefore to your initial query.

So the second step is bring in more data about the concepts. The data can be encyclopedia entry summary, journal paper summaries, videos, images, etc, and are all displayed with the answer.

Quite neatly, the patent talks about ambiguity. If the query can have different answers, then the system is supposed to ask you what you specifically you meant.

This is it in a nutshell. To see what the final thing would look like, skip everything and look at figure 5D. As patents go, it's fairly well written, but still full of legalese. Enjoy

Pierre

EDIT Bill and I were posting at the same time

This post has been edited by eKstreme: Apr 12 2007, 05:43 PM

[bragadocchio]

Looking at the pictures is always a good first step in trying to understand what some of these patent applications are about.

You've described it in a nutshell, Pierre.

Is this what AP meant by the "Edison Algorithm?" I'm guessing that it is.

I thought that it was important to mention the different types of queries that they identified - transactional, navigational, and informational to point out that it appears to be only the informational type queries that look like they are asking a question that go through this process.

A search for "2006 Olympics" won't go through the process of finding an answer, looking at the associated concepts attached to that answer, and finding references that match the concepts. A search for "Who won the triple crown in 1934" will.

[rustybrick]

Awesome stuff Bill. Thanks for the info.

It was awesome meeting AG. The picture is a keeper.