Direct Answers – Taken from Authority Websites

Sharing is caring!

Part One of this series – Direct Answers – Natural Language Search Results For Intent Queries

One of the most important parts of the “Natural Language Search Results for Intent Queries” patent application is how it describes both queries and search results as being in “natural language.”

By natural language, it means ordinary conversational words and sentences, rather than some formalized and highly structured request for results like you might see in a programming language.

As the patent tells us:

A natural language query is a query using terms a person would use to ask a question, such as “how do I make hummus?”

You may remember the search engine, Ask Jeeves, which tried to answer natural language queries. I remember it well.

Back in my early days of working for an agency, they insisted that we send ranking reports to clients each month, showing how well the pages of their sites were doing in a number of search engines, including Google, AOL, Hotbot, Lycos, Excite, and others for total of eleven in all. Ask Jeeves was one of those as well.

So why is Google trying to recapture the style and format of a former rival, like Ask Jeeves? Maybe it wasn’t thinking of Ask Jeeves.

Google may not have had Ask Jeeves in mind when deciding to show answers to natural language questions.

Google may have had a site like MIT’s START Natural Language Question Answering System in mind instead. Try a few queries – but watch out – it’s contagious.

At one point, the professors involved with START and with the artificial Intelligence program at MIT sued Ask Jeeves over their use of the technology involved. They held two patents that aimed at answering natural language questions. We will see a similar process in Google’s patent in tomorrow’s post, but for now, here’s what the news article summarizes those patents as saying:

The two patents which are held by Katz and Winston include the use of natural language searching to retrieve text and database material. A patent filed in 1994 describes “converting natural language queries to structured form, matching the structured form query against stored annotations and retrieving database subdivisions connected to matched annotations.”

Here’s Start answering the question, “what are the symptoms for Mono?” like the question asked in the images for the Natural Language Search Results query.

Start - Answering a question about Symptoms for Mono

Here is the Start question-answering search engine:

Let’s look at the patents from Professors Boris Katz and Patrick H Winston, and the patent that Ask Jeeves didn’t have at the time of this law suit, but ended up getting. One of the interesting things that we haven’t talked about yet from the Google patent are intent templates, and these patents involve natural language queries and use a structured format like an intent template. We’ll delve deeper into the Google patent’s use of such a template tomorrow, but for now, notice how these patents all use a process like that.

The two MIT patents are:

Method and apparatus for utilizing annotations to facilitate computer retrieval of database material
Invented by Boris Katz and Patrick H Winston
US Patent 5,404,295
Granted April 4, 1995
Filed: January 4, 1994

Abstract

A method and apparatus for computer retrieval of database material which may be text, computer programs, graphics, audio, object classes, action specifications or other material which may be machine stored. Annotations are provided for at least selected database subdivisions, preferably with natural language questions, assertions or noun phrases or some combination/collection thereof.

However, the annotations may also initially be generated in a structured form. Annotations are, if required, converted to a structured form and are stored in that form along with connections to corresponding subdivisions. Searching for relevant subdivisions involves entering a query in natural language or structured form, converting natural language queries to structured form, matching the structured form query against stored annotations and retrieving database subdivisions connected to matched annotations. The annotation process may be aided by utilizing various techniques for automatically or semiautomatically generating the annotations.

katz-patent-1

Method and apparatus for generating and utlizing annotations to facilitate computer text retrieval
Invented by Boris Katz and Patrick H Winston
US Patent 5,309,359
Granted May 3, 1994
Filed: August 16, 1990

Abstract

A method and apparatus for computer text retrieval which involves annotating at least selected text subdivisions, preferably with natural language questions, assertions or noun phrases. However, the annotations may also initially be generated in a structured form. Annotations are, if required, converted to a structured form and are stored in that form.

Searching for relevant text subdivisions involves entering a query in natural language or structured form, converting natural language queries to structured form, matching the structured form query against stored annotations and retrieving text subdivisions connected to matched annotations.

The annotation process may be aided by utilizing various techniques for automatically or semiautomatically generating the annotations.

katz-patent-2

Here is a patent from Ask Jeeves that looks like it attempts to answer Natural Language Queries. Note that it also includes the use of Query Templates:

Natural language query processing
Inventors: Tom Harrison, Michael E. Barrett, Swarup Reddi, John Lowe,and Gary Chevsky
Assigned to IAC Search & Media, Inc.
US Patent 7,403,938
Granted July 22, 2008
Filed September 20, 2002

Abstract

An enhanced natural language information retrieval technique tokenizes an incoming query, comparing the tokenized representation against a collection of query templates. Query templates include a first portion having one or more query patterns representative of a query and in a form suitable for matching the tokenized representation of an incoming query.

Query templates also include one or more information retrieval commands that are designed to return information relevant to those query patterns in its first portion.

The enhanced natural language information retrieval technique selects those query templates that are determined to be most relevant to the incoming query (via its tokenized representation) and initiates one or more information retrieval commands associated with the selected query templates.

jeeves-query-template-patent

Authority Websites

In the world of SEO, when people talk about authority web sites, they often are referring to pages to get links from. Discussions in SEO forums often asked about whether or not links from pages with .edu domains or .gov domains where from authority pages because it was usually difficult to get links from such sources.

Though someone who’s been in the SEO world around the same time that Brin and Page brought PageRank into the picture might remember Jon Kleinberg’s Hubs and Authorities. His more technical look at these topics is in the paper Authoritative Sources in a Hyperlinked Environment (pdf).

From the first Kleinberg paper:

Some pages, the most prominent sources of primary content, are the authorities on the topic; other pages, equally intrinsic to the structure, assemble high-quality guides and resource lists that act as focused hubs, directing users to recommended authorities.

The nature of the linkage in this framework is highly asymmetric. Hubs link heavily to authorities, but hubs may themselves have very few incoming links, and authorities may well not link to other authorities.

It’s not the same kind of authority page that the Natural Language Search Results for Intent Queries patent application is referring to when it talks about extracting natural language answers from pages on the Web. Kleinberg’s pages get their “authority status from being pages linked to as examples or illustrations of a topic.

The “intent queries” patent filing instead refers to pages identified by the search engine based upon how people treat those pages.

The Question and Answer, or Q&A engine behind these natural language search results populate a Q&A Data Store by extracting “heading-text pairs found in documents from the authoritative sources.”

I’m going to expand on that part of the process, with some illustrations, over the next couple of days, but I wanted to focus upon how a search engine might determine what an authoritative source might be.

The search engine may process authoritative sources offline, to find answers to “common clear-intent non-factual questions.”

Since these are “natural language questions,” they are just like questions that one person might ask another person, instead of the keyword approach that you often see at search engines. The keyword approach that search engines, with the possible exception of ask.com seemed to prefer isn’t preferred in this patent.

Instead, it wants to “identify clear-intent queries and match the queries to the stored answers”. It also wants to show enhanced search results that have complete answers from one or more authoritative sources.

So, the question remains:

Just What is an Authoritative Source according to Google’s patent?

  • A source that is popular and trusted, as determined by frequent selection of the source in search results, or
  • A source that consistently ranks high in search results for queries dealing in the subject matter of the Q&A data store.

We are told in the patent filing that the authoritative sources may be manually identified or may be automatically selected. They may also be general sources and focused sources.

If you look at direct answers such as information about symptoms of illnesses, or recipes for food, you will often see a link to the source of the information. We are told about these authoritative sources:

The the domains webmd.com, mayoclinic.com, and medicinenet.com may be general authoritative sources for medical subject matter and the domains cancer.org and heart.org may be focused authoritative sources for medical subject matter.

Similarly, allrecipes.com and foodnetwork.com may be general authoritative sources for cooking subject matter and vegetariantimes.com may be a focused authoritative source for cooking subject matter.

So, the pages that direct answers come from are considered to be authoritative based upon the quality and focus of their content, and that is supposedly one of the values in Google providing such answers.

They aren’t answers focused upon specific keywords, but rather based upon query templates and questions that seem to be related to them. That Google provides answers to natural language queries and also provides search results for queries means that as a searcher, you get the benefit of both.

We’ll discuss intent templates, and Q&A Data Stores and mixed in search results in tomorrow’s post.

Sharing is caring!

9 thoughts on “Direct Answers – Taken from Authority Websites”

  1. Hi Bill,

    Thanks a lot for sharing awesome in-depth data and insights with us. That’s all I can say. You are doing a great job.

    Best Regards
    Miraj Gazi

  2. the most interesting part of this for me related to authority sites was : “authoritative sources may be manually identified or may be automatically selected”.

    Google always makes the claim that results are all algorithmically generated. this shows that there is some manual selection/override that goes on. I think a lot of folks (including myself) have always assumed this, but I don’t remember ever seeing this acknowledged at least as it relates to standard organic results.

  3. Hi Tom

    This kind of “authority” sites are pages that might be chosen as the source for direct answers. That is different than the many other “authority” sites that many people hypothesize or guess at. 🙂

  4. Yeah, understood. But they have always said that selection of sites for placement in the answer box was purely algorithmic. I’ve asked this question directly and been told as much. In this patent they are saying that they (someone at Google) can manually select or override what is considered to be an authority site in a topic (like epicurious for recipes). This is the first I’ve seen direct from Google that they will have some manual mechanism to influence what is being displayed in organic search (direct answers in this case).

    In the patent they say that they will “automatically select” (algorithm) or “manually identify” what are authority sites to appear as direct answers by topic. That’s news right? Google saying that there is manual curation to some extent within organic results (outside of like a manual penalty/removal or something like that).

  5. Hi Tom

    This likely is the first time that a Google patent has said that they may decide that certain sites are “authority” sites for direct answer results, and the first one I can recall ever seeing from them. I don’t think they’ve ever said such a thing before anywhere. It is news, but it’s not manual selection for organic rankings of query results for SERPs, and it’s possible that a number of sites might be chosen as authority sites that answers might be selected from (rather than just one.)

  6. Many thanks for sharing some really cool information. I tried fiddling with Start, though it is still very basic, does not cover many areas, like I quizzed it for the recipe of some of my favorite dishes, it drew a blank, but still very nice tool.I remember using ask jeeves, it used to be a great tool.
    But still a great effort and it would be interesting to see how Google evolves this interesting concept further.

  7. Yeah – agree Bill. Thanks. Not saying that they’re picking what appears. Probably wasn’t clear. What appears in the box is determined algorithmically. But the algorithm is/may be working against a manually selected set of “authority” sites by topic or category.

    So it might be that if people at Google don’t select you as an “authority” for the recipe category, then no amount of algorithmic favor can get you in the box. You can still rank #1 outside the box.

Comments are closed.