Google takes personalization and behavioural to a new level
When you read a lot of patents and papers there is a lot of new layers on the search onion. But every now and then... you come across something that makes you really stop and think. Today is one of those days. We never know when they'll happen, but they inevitably will. Pathetic as it is, I sort of pine for 'em.
We all know that mobile search continues to grow, including geo-localization. Most of us also have seen the rise of personalization over the years. The relationship of temporal elements, if it is adapting older content or preying upon the QDF (query deserves freshness) is also fairly common knowledge. That's what makes today's journey fun, it covers all three!
First off, the patent;
(This application is a continuation application of, and claims priority to, pending U.S. patent application Ser. No. 12/277,432, filed on Nov. 25, 2008. )
“In a computing system, information regarding a plurality of events that use a computing device is obtained, and a time-dependant increase in activity for each of at least some of the events is identified. An observed interest by a user in an event is correlated with an identified increase in activity for the event. Information about the activity at a time related to the event is provided for review by the user.”
How Google might look for predictive behaviour
The core concept of the patent looks at “providing digital content based on predictive determinations that are made in response to observations of user behavior”. Digital content can be many things, including search results, but the obvious connection for Google is of course; ad serving. Some of the examples they gave include;
- promotion information,
- event information,
- reviews, directions, and the like.
What they're looking at here are various behavioural signals that might show a time-dependant increase in activity for certain events. When there is an increase in activity for an event they could analyze the search requests related to the event. They can also look at a location as well as a transaction (potentially made via smart phone).
From there they might look at the increase of activity as an explicit action combined with observed interest surrounding activities related to the event, the obvious one being search queries etc.
“The method may further include identifying a lack of correlation between the observed interest of the user and second non-user-specific events, and identifying the information based on a combination of the correlation and the lack of correlation.
- identifying a time-dependant increase in activity for each of at least some of the events.
- Identifying the time-dependant increase in activity for each of the events may include analyzing search requests related to the event. Identifying the one or more time-dependant increases in activity for each of the events may include identifying a location associated with each of the events.
- the information may include promotional information for the non-user-specific activity, or promotional information for an entity in a geographic vicinity of the non-user-specific activity. “
Human beings are habitual. We live fairly structured lives. We often perform recurring or semi-recurring actions. The temporal elements of these traits that Google talks about looking at include;
- within a minute,
- within an hour,
- over several hours,
- during the course of a day,
- over a few days,
- a week,
- a month,
- a year, or over multiple years, etc.
If we consider this as well as the proliferation of mobile devices, you can see how this might come in handy for a search engine. So let's have a look how...
How geographic and behavioural data could influence search results
This is where it starts to get pretty interesting... here's a few snippets from the offering that best explain the thinking... (bold is my emphasis)
“For example, a user may generally drive from home to work each weekday and may return home from work in the afternoon. Though occasionally the user may depart from this routine, such as by carpooling, taking a vacation or sick day, riding the bus, etc., in general, the user may typically follow the daily routine of driving to and from work.
In other examples, a user may tend to golf, cycle, or go boating each weekend, attend a club meeting the first Tuesday of each month, or wash the car almost every Saturday morning.
Some activities may recur at particular intervals, such as each weekend or on one or more particular days of the week or month, but may additionally be correlated to another time interval, such as a season. “
Are you starting to see where this is headed? Ok, let's make sure....
“For example, a golfer may tend to play golf at a local golf course (or one of several local courses, e.g.) each weekend (or many or most weekends) during the spring, summer, and fall seasons, but may tend to practice instead at an indoor driving range on weekends during the winter.
In another example, a user may engage in activities that recur on an annual basis, bi-annual basis, or some other long-term recurring schedule. For example, the user may travel to visit family during Thanksgiving each year. In another example, a group of six couples may enjoy a tradition of getting together once every six months for dinner, drinks, and fellowship, and may alternate hosting the event such that each couple hosts an event once every three years. “
Catching on? This approach would use the user tendencies to anticipate future actions and provide content tailored to suit over the mobile device. It could be search results, ads or both. This could be provided at the time of an anticipated event, or even a relative period prior to it. Let's imagine that it predicatively knew you'd be in Chicago this weekend. Not only might it add content related to the city, but even tailor it to a degree for a given region of the city.
Further examples of events could include;
- the dates and times that the television program "Law & Order" is broadcast,
- a basketball team's published schedule,
- the schedule of nights that a local dance bar will feature line dancing
Are you curious about that first one? Let's say you use you Android to control your TV or tend to search topics related to that show around the time of broadcast each week. Again, even being at home can have a behavioural geo-location predictive model.
Adapting to the patterns
One thing I found interesting are the elements that look for increased or decreased interest in a given activity, (or inquiries or communications regarding the activities or events). In a few instances they classified these as patterns of “ regular, semi-regular, or occasional user behaviour”. By looking at the day-to-day interactions with your mobile phone, or computing device. Presumably a tablet.
Part of the reasoning for the intuitive nature of the predictive nature of it is that it avoids the more explicit indicators such as listing preferences such as, “ activities, events, causes, associations, characteristics, habits, tendencies, settings and the like”. They also talk of using those types of explicit actions that could be offered in refinement interface elements.
What it means to me
And yes, for those that like to just scan and then roll on to the end of these posts, there is a lesson to be learned. The world keeps evolving, marketing must too. HA! Take that. No, seriously, read the entire post and then try to envision such concepts against how you're doing things now.
For me, I can see an interesting approach as far as combining not only traditional behavioural activities, but adding another layer via geographic tracking via mobile computing devices. As a marketer, imaging a combination of Google Local, street view, indoor maps with ad serving is an interesting twist and could offer far more tailored campaigns. As an SEO, I start to think about how well I not only target a client geographically, but also understanding the demographic better.
As always, a patent is just a patent. One of many. It's more about a cumulative instinct toward the mindset of a search engineer. Google ones in this instance. Future proof SEO is about seeing into the future. It helps one better understand the search world around them. For me, this was a damned interesting ride.
Until next time, play safe.
The Coffee Drinker
“Using observed or determined patterns, the system can determine appropriate information for presentation to the user at a relevant time. For example, the system may use the observed information that the user enjoys a morning coffee before work, and may present a coupon from Coffee Shop 1, or from another coffee shop (e.g., "Coffee Shop 2") located along the route 103a usually travelled by the user to work on a weekday morning. In an example, the system may present the information to the user shortly before the user typically departs from home for work, for example. The user may appreciate this information, because it may be tailored to a preference of the user (e.g., the user's enjoyment of coffee on weekday mornings), and because it may be delivered or presented in a time-opportunistic fashion (i.e., near the time when the user often purchases coffee). As such, the user may save money or time with little or no additional effort expended in obtaining the promotion information.
In some implementations, the system can use information indicative of a first user pattern and information from an external signal to present the user with information at a time outside of a time associated with the pattern. For example, suppose the user in the example above is driving about town on a Saturday morning at 9:00 A.M. The system may use the observed information pertaining to the user's weekday coffee purchases (that is, associated with a pattern of activity occurring on weekday mornings), and may use the external signal that the user is presently driving in the morning (albeit on a Saturday) to present content associated with a coffee shop, such as a coupon for a free bagel with the purchase of a medium or large mocha. One or more additional external signals, such as the user's present location, calendar information, mobile device communications or requests, etc., may further be used to tailor content for presentation to the user, including for example presenting a coupon for a nearby coffee shop. “
The Sports Fan
“In certain implementations, the information-providing system can also present time-related user information based on a combination of global information and an observed pattern. As an illustrative example, the system can observe that the user traveled to the stadium the evenings of Monday, Tuesday, and Wednesday based on events 108a-c. In one example, the system can determine that baseball games are held in the stadium at a time near the occurrence of the events 108a-c by accessing global information, such as the baseball team's schedule, which may be stored in a server and available on a web page, for example. For example, the system may determine that the team plays home games at the stadium and may access the team's schedule from a team or league web site. In various implementations, the system can request that the user upload a game schedule for the baseball team, or send a query to a search engine to search for results relating to the team and its schedule, or access the baseball team's website to obtain the schedule. The system can use the user-related information (traveling to the stadium on Monday, Tuesday and Wednesday), combined with the global information or external signal (that a baseball game was played at the stadium each of those nights), to determine that the user likely attended baseball games at the stadium each of Monday, Tuesday, and Wednesday nights.
Such a determination might indicate any of several meanings. For example, it may indicate that the user is a full or partial season ticket holder for the baseball team, and that the user may be likely to attend many future games at the stadium. Such an indication may be bolstered if the user continues to attend baseball games at the stadium on a regular or semi-regular basis, or if external signals are observed that indicate an interest in the local baseball team (e.g., using one's mobile device to check league standings or read articles on the team). Alternatively, if each of the games that the user attended were against a common opponent, it may instead indicate that the user is a fan of the opponent, rather than the local team, and may portend that the user is likely to attend future games at the stadium when the opponent visits, but may be less indicative that the user will attend games when other opponents come to town. Such an indication may be bolstered if the user never or rarely attends baseball games involving other opponents, attends future games involving the same opponent, or exhibits observable behavior indicating an interest in the opponent baseball team. “
The Movie Watcher
“For example, the system may observe that the event 110, visiting a movie theater on Friday evening, occurred even though a baseball game was concurrently being played in the stadium. This may indicate, for example, that the user prefers seeing a movie on Friday nights to attending a baseball game, even if the user may have previously purchased tickets for the baseball game (e.g., if the user is a season ticket holder). In certain implementations, the system can determine that the user prefers to go to the movie theater rather than watch baseball on Friday nights. This may distinguish Friday evenings, for example, because the user selected baseball over movies during the other days (e.g., Mon., Tues., Wed., as described above) where both choices were available. Based on the user preference, the system may provide movie-related information instead of baseball-related information on Friday evenings when both activities are scheduled, even though another user-observed pattern (associated with events 108) may have indicated that the user attends many or most baseball games played at the stadium.
For example, the system can present movie schedule information, movie trailers, advance-ticket-purchase information, 2-for-1 buttered popcorn promotions, and the like, at 6:00 P.M. on Friday evening, in anticipation that the user will again choose to attend a movie on Friday evening over a baseball game. Similarly, the system may present information on events similar to movies (e.g., a musical play) that may be occurring in a nearby venue (e.g., in a student-run theatre near the movie theater) for user review on Friday evening. In some cases, the system may present information related to the baseball game and information related to movies on Friday evening, given the user tendencies described above.”
Day to Day Life
“the system can observe that a significant portion of users tend to shop for groceries and refill their vehicles with gas during weekends. In some implementations, the system may present the user with information related to grocery stores or gas stations (e.g., advertisements, coupons, promotions, locations or directions) near the user's home, as shown in the map 102d, during weekend periods. Such global information can be combined with observed user tendencies in various implementations, or may be used independently to predict user behavior and provide appropriate content accordingly. “
There's a ton of other examples in this patent, if you read only one patent this week (lol), make it this one. Even my post here doesn't fully do it justice.
- One of the authors of this patent, Sumit Agarwal, (former head of mobile) joined the US Defense Department (for outreach and social media) back in 2010. You can see his bio here., (PDF)
- Another named author, Dipchand “Deep” Nishar, who managed Google’s mobile initiatives worldwide” from 2005 to 2007, was deposed in a 2011 Google case with Oracle, and is now with LinkedIn.
- The third author, Andy Rubin, who served as head of Android since the platform was acquired in 2005, has some 17 patents and continues to work there, although not with Android anymore. No one knows. Google Glass maybe?