How Google may Create a Reputation Score for People Based Upon User interactions

by Posted @ Jul 24 2015

Twitter

One of the things I was hoping to see with Google’s Social Network, Google+, was Google using reputation scores to rank content from users of the social network system. In the past, I wrote about Google’s Agent Rank, which was originally introduced in a patent application from Google back in 2007. People have been discussing Agent Rank under names such as Author Rank, in many places since, including at Google+. Sadly, the authorship markup that was behind Agent Rank was discontinued.

I was surprised by a recently granted Google patent that tells us that it aims to help people find more trustworthiness from other members of social networks with reputation scores. It describes a reputation score for use in a social network. It is not the Agent Rank I wrote about years ago, and has a number of different elements, including details of how machine learning might be used to calculate reputation scores, which would primarily be based upon information collected about their online interactions with others. It also refers to a social bonus that might be added to such a reputation score.

The closest I’d seen Google come in the past to describing in detail such reputation scores for online users was in a patent I covered in the post, How Google Might Rank User Generated Web Content in Google + and Other Social Networks. It described how content that people might contribute to an online network, and their responses to content submitted by others might be used to better understand topics they are interested in.

In that patent, we were treated to a description of how Google might attempt to understand what things a person might be interested in and have an expertise within topic-wise, in their interactions on a social network, such as the things they posted and shared, and their comments in response to other people.

This new patent tells us that Google may look at user activities from certain data sources, classifying those into categories, generating a base score for each category of user interactions to determine a reputation score for the user. This categories approach is good – it can help identify things it feels these social network users might be experts in.

This scoring process may also consider other user activities including data it refers to as training data that identifies user activities associated with a trustworthy user and user activities associated with an untrustworthy user, to generate a machine learning result based at least in part on the user activities to determining a second reputation score for the user based at least in part on the learning result.

These reputation scores may also be influenced by a social bonus score that would be based at least in part on social affinity data between the first user and other users in the social network. This social bonus score could look at
activity data involving other users’ reactions to user interactions created by the first user.

The patent is:

Generating a reputation score based on user interactions
Inventors:Brandon Bilinski, Alexander Collins
Assigned to: Google
US Patent 8,793,255
Granted July 29, 2014
Filed: October 15, 2012

Abstract

A system and method for generating a reputation score is disclosed. A processing unit processes user activity data from data sources to identify user interactions associated with a user. A categorizing engine categorizes the user interactions into categories. A social bonus engine determines a social bonus score based on social affinity data. A scoring engine computes a first reputation score for the user by combining scores for the categorized user interactions with a social bonus score. A learning engine receives a second set of user interactions and training data and generates a learning result that is used to update the first reputation score.

User Activity Data

The patent tells us that it would track user activity data about users who opt-in to having data collected across a range of data sources.

  • Search (This sounds like what Google tracks of search history data, to personalize search results for searchers based upon things they are interested in.
  • Entertainment (Again, information that Google might track regarding what people search for and possibly browse, and maybe even subscribe to on the web.
  • Social activity (How do people interact with others, and on what kinds of topics, with microblogs, forums, social networks
  • Activity on third-party sites (This could include shopping sites and review sites, places where ratings could be left, and more

Take-Aways

The patent does provide a considerable amount of details about how it might calculate a reputation score for individuals and some on how a machine learning process might fit into that. I don’t know if this is Google’s second bite at a reputation score like Agent Rank, but it could be.

It does look like this reputation score could influence how content from people you are connected to may rank in logged-in searchers, much like I’ve written about recently in Is An Improved Version Of Agent Rank Returning To Google?

Google knowing something about the people who create content, and ranking that content based upon reputation scores for those people sounds like it could be an improvement upon a link analysis based ranking approach such as PageRank, because it would possibly be less likely to be spammed.

Is this return of Author Rank? It might be. Hopefully it will stick around longer.

subscribe to our newsletter

4 Comments

  1. Dallan

    July 25th, 2015 at 12:09 am

    With the success Google has had with G+ in the area of interest pages and GMB accounts Google could leverage this to do a author rank or agent rank.

    Besides G+, would Google match up users with expert authors using their Chrome browser to do this? Can they do this with other browsers?

    Reply

    • Bill Slawski

      July 25th, 2015 at 2:49 pm

      Hi Dallan,

      This patent discusses reputation scores with the context of a social network, such as Google+, so the matching of of expert users seems to be something that fits withing social networks. I’m not quite sure I understand what the difference you are asking about might be.

  2. Soumya Roy

    July 26th, 2015 at 4:08 am

    Another Insightful post Bill Sir, thanks for the share. Now this can be rally interesting to see how it is different from the old Author Rank and finally how Google re-implements this on ranking.
    Will Google only depend on their own G+ system or they will also accumulate the reputation data/ signals from FB, Twitter and Linkedin?
    Will be very much interested to read more on this particular topic.

    Reply

    • Bill Slawski

      July 26th, 2015 at 12:23 pm

      Hi Soumya,

      Thank you. My guess is that Google will rely upon Google+ primarily to interaction data, since they have the best access to that information, and to information that might be behind the scenes, such as where something might have been posted from, when it was posted exactly, what the user agent was that posted it. Google doesn’t necessarily have access to data like that for FB, and LinkedIn, and we don’t know how much data they have access to pursuant to their deal with Twitter.

Leave a Comment