Click a Panda: High Quality Search Results based on Repeat Clicks and Visit Duration

Posted @ Jun 28 2017 by

Click a Panda

A Panda Patent Returns with New Claims

Three years ago, I wrote about a patent from Google’s Navneet Panda which focused on how pages are ranked in search results. My post was really a summary. and was called, Google’s Panda Granted a Patent on Ranking Search Results. Last week, a continuation version of the patent I wrote about in that post was awarded to Google. I like continuation patents since the claims section of such patents change as the process described in them has changed in some way. Those changes are often an interesting reflection of Google making changes to the process described in the original patent. This one points to repeat clicks as being important.

We have seen changes at Google in recent years involving things such as Rank Brain and Machine Learning, but the updates in this patent appear to be related to user interaction with websites. As I mentioned, one of the inventors involved in this patent is Navneet Panda, who is best known for being involved in the Panda update at Google. It doesn’t come as a surprise that the patent tells us that it involves demoting search results for pages that are low quality.

I looked at the claims from this new version of the patent and was suprised by the changes in how it worked.

The patent focuses upon search results returned in response to a query. It mentions a group-based modification factor that might influence which search results appear for a query:

In general, one innovative aspect of the subject matter described in this specification can be embodied in methods that include the actions of receiving a search query from a client device; receiving data identifying a plurality of search result resources and respective initial scores for each of the search result resources; identifying a respective group of resources to which each of the search result resources belongs; determining a respective group-based modification factor for each group of resources; and adjusting the initial score for each of the search result resources based at least in part on the group-specific modification factor for the group of resources to which the search result resource belongs to generate a respective second score for each of the search result resources.

The Process behind the Patent

The process described in the patent starts off with determining if the result selected in response to a query is navigational in nature or not. If it is, then the process described in the patent doesn’t seem to add much information about the site selected in search results.

The patent tells us about unique clicks and deliberate visits (from a user who has not before clicked upon search results from any resource in a particular group) and how it might look at a count of repeat clicks on the particular group of resources selected to generate a repeat click fraction for resources chosen.

If someone, over the course of more than one search session, performs a search and selects the same resource (possibly using different words), that would be counted as a repeat click, and the more repeat clicks, the higher a signal of quality for that resource – if someone is returning to a page purposefully (a deliberate visit), that is a positive sign, and combined with other signals, such as how long the duration of their stay there might be, before they return to search.

The Importance of Duration of a Visit

While the duration of a visit to a search result is something that I’ve seen hinted at as something that a search engine might use as a positive signal, this is the first time I’ve seen it spelled out clearly in a patent, which tells us:

The average duration metric for the particular group of resources can be a statistical measure computed from a data set of measurements of a length of time that elapses between a time that a given user clicks on a search result included in a search results web page that identifies a resource in the particular group of resources and a time that the given user navigates back to the search results web page.

As I wrote about in Unraveling Panda Patterns at Moz, indications of high quality in search results are a positive signal, and indications of low-quality resources can result in those resources being demoted in search results. The patent tells us the benefit of this is that:

Thus, the user experience can be improved because search results higher in the presentation order will better match the user’s informational needs.

The updated continuation patent is:

Ranking search results
Inventors: Navneet Panda, Vladimir Ofitserov, and Kaihua Zhu
Assignee: Google
US Patent: 9,684,697
Granted: June 20, 2017
Filed: December 31, 2012

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for receiving a search query from a client device; receiving search result data identifying a first plurality of search result resources and respective initial scores for each of the first plurality of search result resources; identifying a respective group of resources to which each of the search result resources belongs; determining a respective group-based modification factor for each group of resources; and adjusting the initial score for each of the search result resources based at least in part on the group-specific modification factor for the group of resources to which the search result resource belongs to generate a respective second score for each of the search result resources.

The earlier version of the patent Ranking Search Results granted in 2014 differs from this newer version, especially in the claims section, which discusses clicks in much more detail.

Take Aways

We’ve been told by many Google representatives that click selection data tends to be noisy and isn’t used in ranking search results, but instead may be used to test algorithms used by Google. Is that true with this patent as well? It appears to be an important part of how the process described in the patent functions, and worth spending some time thinking about. The patent points to clicks from logged-in searchers and is accompanied by cookies or Device IDs. How reliable is it?

But more importantly: (1) Give people a reason to come to your website for the things you care to rank for, and
(2) Give people a reason to spend time on your page, once they come visit you. 🙂

7 Comments

  1. Michael Martinez

    June 28th, 2017 at 2:43 pm

    This appears to describe an algorithm for scoring the quality of test search results (not used for real-time ranking). The type of math involved requires a fixed number of data items, which you would have in a learning set. The group modification score might be the Panda algorithm score that we have speculated Google is using. The original Panda algorithm’s results were compared to the user-generated demotions that Google collected from people downvoting items in live search results.

    Reply

    • Bill Slawski

      June 28th, 2017 at 2:51 pm

      Thank you for sharing your thoughts, Michael. We only see parts of the whole when we are given glimpses like this, so a little context is helpful.

  2. Mark Traphagen

    June 29th, 2017 at 7:05 am

    Interesting, but keep in mind this would be for personalized results only, not a general ranking factor, and other Google search engineers have given very good reasons why they do NOT use CTR and length of visit as ranking factors, but they might be OK to use to give better personalized results as proposed in this patent.

    Reply

    • Bill Slawski

      June 29th, 2017 at 9:43 am

      Hi Mark,

      It is good to see you here. I have to ask, because you raised this point. What is it in the patent that indicates to you that it is only for personalized results? I don’t have records of everything everyone at Google has said about ranking, but I don’t remember hearing someone from Google stating that they don’t use the length of a visit as something that might be used to rank. Whom do you recall stating that and where? Thanks.

  3. Pingback: A Case Study in Google's Use of Visit & Click Data - Rand's Blog

  4. Pingback: Personalizing Search Results at Google - SEO by the Sea ⚓

  5. Pingback: Google Ranking Factors & Quality Rating Criteria in 2017 - Hobo

Leave a Comment