SEO Blog - Internet marketing news and views  

Social networks are Open for profiling

Written by David Harry   
Monday, 06 October 2008 02:00

Can Google solve the social monetization puzzle?

Google tracks social network usersBusiness Week had an article recently about; Making Social Networks Profitable - which talks about Google’s (patent pending?) social user ranking system to identify influencers for more targeted Ad Delivery. It’s altogether unsure exactly which patent they are talking about, but some recent patent publications may shed some light on the matter.

One such approach was from Google; a patent dubbed by some as a ‘FriendRank’ which was released back in July of this year;

Network Node Ad Targeting - filed Dec 2006 / published July 3rd 2008 - Note; (FriendRank is apparently a patent pending  method from SocialMedia)

Another related approach you can look at is Microsoft’s; Targeting social media influencers. (filed Sept.2006/ published March 2008) - reviewed here on the trail.

These essentially target top influencers by connections and related topics or groups for the purpose of placing advertising in said profiles. While these are good ideas for targeted Ad serving, (or content serving in general) they don’t provide for a deeper reach across multiple social platforms, or tight topical relevance.

That's where a new trio of patents from Google may further mine social data for better Ad serving, content delivery, association suggestions and the like (mostly for Ad serving if U ask me ;0).

Open Profile Content Identification – Filed Mar 2007 : Published Oct 02 2008

Related Entity Content Identification - Filed Mar 2007 : Published Oct 02 2008

Custodian based content identification - Filed Mar 2007 : Published Oct 02 2008

(Note: Tomasz J. Tunguz-Zawislak (whom worked on all 4 including the network node patent ) was also named on; Profile advertisements – filed June 2006 : Published Jan 2008 and covered here by Bill Slawski)

Anyway, let's look deeper into the three from last week.

Enter Open Social

Monetizing social networks has been notoriously difficult and these patents from Google, for better Ad targeting based on user profile data, could be a few of the pieces to the puzzle. In the world of present day targeting a variety of signals can be used to better target users with advertising such as;

  1. On page topical (semantic) analysis
  2. Past user interactions (with search results, pages or Ads)
  3. Query revisions (for search results)
  4. Ad performance metrics
  5. Personalization data

While these would give some insight into user intent, social networks present new signals from which further refinements in targeting can be obtained. Because social network users represent a wide variety of user types and interests, ad delivery has been a tricky business to say the least. Social Networks not only offer up topical categorizations, but user information and relations that may also speak to a viewer’s interests.

Profile AnalysisThe network node model, (‘friend rank’) only identifies user types, categorization and associations; it does not give deeper data as to the types of content that would best suit a given viewer to the page. Allthough the network node scoring could eaily be incorporated into elements of these methods.    

What is interesting is that the patents seem to be a perfect match to Google’s OpenSocial offering from last year as the system can be used over multiple social platforms. Getting people on board with this program would certainly ensure a greater reach for ad targeting systems such as the ones we’re looking at today.

Think more about the potential implications for OpenSocial my friends as we'll revisit this trail soon ;0) < moving along... >

Profile Analysis

One part of the process involves identifying user tags to store in the system through open profiles. Each user profile has ‘open data’ which is publicly available and can be analyzed to define interests and preferences of a user as well as demographic information. The process of analyzing the profile can consist of;

  • A natural language processor that extracts one or more phrases from free-form text data and assign phrase weights to the extracted phrases.
  • A sentiment detection processor is that identifies user interests and non-interests from the text data in the user profile.
  • The category processor to associate labels with a user profile based on the extracted one or more phrases, assigned phrase weights, and the identified user interests and non-interests.

Now, depending on what types of data collection you are after, profiles and groups of profiles can be given various demographic and interest related (topical) categorizations.

If we also consider the network node (‘FriendRank’) concepts, then a metric could also be implemented for that as another ‘weight’ in the process. The main goal is to develop signals and categorizations for social network users.


Entity Relationships and Topics

Entity relationships are essentially the users one is ‘friends’ with or part of groups you may be involved in on a given social network. By looking at the open profile data, entity relationships and groups; topical information can be established. That is the approach behind the second part of the series.

“The entity relationship, can, for example, be based on similar interests defined by the user accounts and/or similar interests defined by the user accounts of acquaintances of a particular user 112, and/or memberships of groups, or other identifiable signals.”

Proile dataAnd that’s not all. Other signals could include more passive or implicit relationships.

Such data types can include;

  1. Common behaviour
  2. Similar membership in groups
  3. Similar profile data
  4. Other similarities

In addition to that, data can be taken from content submitted by the user as well as content from friends and groups they are involved in. Taken in part or in the aggregate, the relationships and topical data provide insight into potential advertising targeting.

There are also provisions for providing less weight or discounting altogether profile data, content submissions that may not meet thresholds of the content type. Once again, phrase relationships can be used in the analysis.

Actually, phrase analysis was also mentioned as a tool for finding entity topics. By identifying the most frequently occurring keywords/phrases or even grammatical elements (nouns, verbs) more topical data can be obtained for tagging the user profile.

And how about some data not related to the network? We have that too;

”For example, entity relationships and entities can be identified by processing web logs, e.g., blogs, AND processing web-based communities, e.g., homeowners associations, fan sites, etc., by processing company intranets, and by processing other data sources.”

In short, there are no limits to the types of data that can be mined (not scraping are they?) to find user relationships and associated profile and content signals. By then monitoring the performance of the delivered content (ads or other content) the system could then tweak what it displays as it learns from its hits and misses.

Interestingly, this data can be used not only for targeting advertising, but ultimately for suggesting content, groups and even other users you may be interested in.


Custodian Profiles

The system when targeting a given content type could look at the user profile associated with the content in questions eg. When serving an ad on a given page within the network to another user. These are called, at least in this filing, custodian profiles.

OpenSocialAccount information that can be used include;

  1. user profile data
  2. user acquaintance data
  3. user group data
  4. user media data
  5. user options data
  6. and other user data

The user profile can, for example, also include general demographic data, such as age, sex, location, interests, etc. Or it could include professional information, e.g., occupation, educational background, etc., and other data, such as contact information. Depending on the social network, the types of data that can be used varies.

The elements can then be considered as a ‘custodian profile’. As an example the ‘user media data’ or content submitted to a social network for example, can be assigned to the user custodian account. The system can use information within custodian profiles to targeting signals as such;

“… the custodian profile data may include professional information such as "Fishing Guide," geographic information, such as "Key West, Fla.," and a list of interests related to fishing and boating.

The keywords can be provided to the content serving system, which can, for example, serve advertisements relating to Key West fishing guides.”

The custodian of the content’s profile can be used as can that of the viewer (if known). This can lead to a hybrid delivery based on both the viewer profile and the custodian profile;

“….. the viewer profile data may include hobby information such as "deep sea fishing," geographic information, such as "Seattle, Wash.," and a list of interests related to deep sea fishing.

The keywords can be provided to the content serving system, which can, for example, serve advertisements relating to Key West deep sea fishing guides and travel options between Seattle and Key West. “

Knowing more about the profile of a user means the system can not only target content and ads based on related profile data, but assign performance metrics as well. For example the viewing of a given web page in the network could constitute and interest in that topic. Furthermore content/ad performance can be tracked over time intervals (such as weekends, time of day etc..). Data collected from the profile, associations and behavioural aspects can better target future content delivery.

Taking into account the data from the person related with the content (ie; news submission, discussion etc..) potential relevance can be gained in serving advertisements and so on. Essentially, if you are viewing a page there is every potential that you may also be interested in topics related to the user that created the content. This is especially handy for ad serving to a visitor that may not be part of the social network or not logged in.


You’re not paranoid, you really are being watched

Ultimately each of the methods adds a layer to the user targeting system. If a given web page on a social network has incomplete or a lack of content, on page semantic targeting won’t really produce content suggestions/advertisements that are relevant to the user. By looking at open profile data, associations and custodian signals, elements beyond mere semantic matching, greater targeting can be achieved.

Given all of this here’s a potential framework of elements for targeting social network users;

Open Profile (user identification) - Uses identification and scoring/categorization analysis of a user profiles

  1. demographic data
  2. interest/topic categorization
  3. media data
  4. group associations
  5. influencer score

Custodian profile (content relational) - Looks at inferences between the viewer of a web page and the person that created, or manages it.

  1. user profile data
  2. user acquaintance data
  3. user group data
  4. user media data
  5. user options data
  6. and other user data

Relationships and Topics - For looking at common relationships among users and related topical categorizations (and performance metrics).

  1. Common groups
  2. Common behaviour
  3. Similar membership in groups
  4. Similar profile data
  5. Common acquaintances
  6. Other similarities

As you can see there are a wide variety of factors that can be used in social networks to try and deliver more relevant content (including advertising). While the whole ‘friend rank’ approach can find the influencers, the systems mentioned in these patents seem far more flexible in identifying broader social targeting. Sure, there are those that might have concerns relating to privacy, but that’s why this system is (for the most part) built around publicly available ‘OpenSocial’ information

Combined these tools (along with the network node targeting) certainly have the potential to start monetizing social networks better and deliver tighter, more targeted traffic to advertisers. It can also facilitate targeting of other content, suggest groups and users to members and more. Now let's wait for the other shoe to drop... the prvacy peeps ;0)

... leave you with a fun diagram from one of the associated patents (in case you want to build yer own);

targeting social network users


... and if you made it all the way through, congratulations... barely made it myself (helluva a lot of readin').


Search the Site

SEO Training

Tools of the Trade


On Twitter

Follow me on Twitter

Site Designed by Verve Developments.