Andreas Weigend
Stanford University
Data Mining and Electronic Business
Stat 252 and MS&E 238
Spring 2008

=Class 7: Geolocation data ----

List of Initial Contributors

Karen Ryberg
Arun Saha
Yi Chai
Bin Shen
Bill Whiteley
Sean Sit
Adam Ting
Bo Cowgill

Audio Link

In mp3 format:

Reading for Next Class

We will be discussing the following paper (fresh from the press):

Krause, E. Horvitz, A. Kansal, F. Zhao. Toward Community Sensing,

Final Homework

To be complete in groups of 2, 3, or 4, d
o one of the following:

1. Social networks: Analyze our crowdvine
2. Facebook all: Understanding your social graph
  • FB: How many of your friend put in gender they are looking for?
  • Who are the singles of your target gender who are looking for?
3. Geodata: Visualize location and movement


What do you do with your mobile phone?
What do you want to do with your mobile phone?
What can you do with your mobile phone?

Here's a fact: only 4% of US commerce is online transactions. However much more data that is inference. 80% of people wanting to buy cars do research online. 90+% of people have mobile phones with them. So use auxiliary devices where the results or data are pushed onto mobile phones which people carry with them.
Mobile matters!
Can we attract more internet commerce using this popular device?

Take a look at another example, user generated content, did you know that, in every minute, 8 hours of YouTube videos and 6500 Flickr pictures get uploaded?
How can mobile phones enhance our experience with these activities?

Why Geolocation?

Location Based Advertising
  • tomtom.JPGRestaurants - Advertise to people in the vicinity of a particular eating establishent, such as the specials. Consider the target audience. People out in public are far more likely to eat out than people currently at home, but much of restaurant advertising are television commercials. Also, it would likely be a good idea if people could subscribe to a particular type of advertising rather than be bombarded with any type of ad anytime, which would cause the consumer to tune out the advertisments. Instead they should be able to use their mobile device to query ads by changing their status to hungary, for example.
  • Stores - This is nearly identical to the restaurant example. Local stores could send out advertisements like special limited time sales or coupons to people in the vicinity to attract customers. Once again, this would likely work best if it were an opt in architecture.
  • Apartment Hunting - This is also similar to the previous two examples, but applies to a more narrow niche of apartment hunting.

Combining Location and Context

  • Time Management - Combine your schedule and appointments with your location and potentially other information, such as current or predicted traffic patterns, and receive alerts of notices if you are on schedule or running late.
  • plazes.JPGHelper Applications - Combine your current location with other contextual information to suggest useful actions for the user to take. For example, when your car is low on fuel your location could be used to suggest a local gas station.
  • Friends Localization - Where are your friends located at? If you know where your friends are and maybe what their current status is (similar to an instant messenger status) you could meet up with them or get an idea of what they are doing. Are they at the same baseball game you're at? Users could also receive alerts when friends get within a certain distance.
  • joey.JPGTransportation - Using your current location to help someone find a parking spot, find a bus stop or catch the subway. Consider how much time is spent looking or waiting for these things. If transportation information is combined with geolocation data the amount of wasted time could be greatly reduced.
  • Sticky notes - Leave messages for someone at a particular location. metosphere.JPGMaybe it is a reminder to do something or it could be a flirtacious note for someone. This way the message left for someone will be taken in the context of the location in which they received it, which is contrary to simple text messages now which contain no location context.

Location Data

external image AndrewTurner_100.jpg
Andrew Turner - A neogeographer involved in helping build the geospatial web.
Quick Vocabulary Lesson:
  • Exonym- a name for a place that is not used within that place by the local inhabitants
  • Endonym - A name used by a group or category of people to refer to themselves or their language, as opposed to a name given to them by other groups

Location Revealing Architectures
  • external image ConstellationGPS.gifIP Address - Determine geolocation of Internet users based on their IP address. Currently able to localize people with a reasonably high probability down to a given city. Used for geo-targeted advertising on websites.
  • Global Positioning System - A system of 24 orbiting satellites that provide geospatial referencing using time of flight calculations from multiple signals. This system requires multiple satellites to be visible and often has difficult in urban settings due to the obstructions caused by tall buildings.
  • location_matters.JPGCellular Tower - The approximate location of a mobile phone user can be calculated by using the signal strength measurement to multiple cellular towers.
  • WiFi Base Stations - WiFi base stations can be used to determine your location if they have been tagged with location information. Loki uses this method to localize people.

Hardware devices

  • Ublip- Out-of-the-box geolocation tracking products with web based interfaces.
  • Garmin- One of many manufacturers of handheld GPS tracking devices.
  • Nokia N810- Very small Internet tablet
  • Dash- Internet connected automotive GPS device that utilizes a GPRS connection to enable up to the moment optimized routing based on real-time traffic loads.
  • SPOT- Satellite messenger that reports a person's GPS location periodically. It is also able to call for help if the user gets into trouble.

external image nokia-n810-press-top.jpgexternal image SPOT.jpgexternal image de_ondash_3route_270px.jpg

Location Data Representation and Formats
  • KML - Stands for Keyhole Markup Language and is an xml based language used for geographic annotation and visualization. It is used by Google Earth, which is one of the most popular mapping software packages currently used.
  • GeoRss - Specification that provides a way to encode geolocation data into RSS feeds. It has a simple version and a more complex version based on Geography Markup Language (GML)

GeoLocation Applications and Services

A number of applications provide Geolocation data and services. Data can be broadly classified as implicit data and explicit data both of which can be mined.


Geocoding is gathering information about a place from public photos of people at that place. Since individuals apply tags other than geotags to a given photograph, this metadata can be mined to figure out attributes of a place or a cluster of places with similar attributes.


Flickr has a feature that enables users to geotag a photograph. Geotagging is the process of adding geographical identification metadata to photograph. To date users have geotagged 63 Million photos. One can make good visual representations using this geotagged data to leave a trail of where pictures were taken as in the bike route below
external image map
Flickr Geotagging

However geotags do not have a well defined meaning. There can be multiple definitions of a particular geotag like Boston which can stand for a photograph taken in the city of Boston or a sketch of the city. Here's an example of Auburn also tagged as Awwwburn. Some interesting applications can be to find places satisfying a certain theme e.g geotagging can be used to figure out “Sadness in New York” which would be the places in New York where people are sad.

Intel-Berkeley Lab:

They have an interesting app called Ergo that places little sensors that measure air quality around the area. A user can move around in an area and get a time history of their personal air quality. These are being installed on street sweepers that can collect this data from various points that again provides air quality information in the neighborhood.


Navizon is a collaborative database that maps out the Wifi and Cellular landscape in a given area and acts as a virtual GPS by pinpointing the users’ location using celltower triangulation.


Brightkite is a location-based social network.that allows people to take their online profiles onto their mobiles, view where their friends are and what they’re upto.


Loopt is a similar service to Brightkite that lets you see your friends online and view their recommendationson places of interests around you.


Mologogo is another service that allows you to view your friends online with geolocation.


A very nifty application that lets one view geolocation along with the twitter post.

Twitter Vision


Socialight is again a mobile social networking application. The site allows you to leave virtual notes behind at different points mentioning attributes of the place(like a Post-it) for future visitors. The messages could be anything like "Tuna sandwich is great at this restaurant" OR "great sunset view". This creates interesting social applications like tours for Seinfield fans to view places mentioned on Seinfield, the filming locations of Lost. An example application here on New York history for passer-bys where virtual post-its point out various places of interests.

Project Runway NYC


Some of these are closed social networks while others are open social networks that leverage network connections from other social graphing applications like facebook. Can websites based on open social network can be monetized. Mologogo makes money on some hardware they send out, Brightkite is trying to get into some niche areas but the revenue model is not crystal clear yet. Much like instant messengers, closed social networks will eventually be opened up by third party applications that allow users to connect across social networks.

Modelling Context/Situation on Mobile Devices

The mobile phone is a device that stays very close to the user and constantly follows the user hence there is a great amount of implicit as well as explicit data that can be obtained from the user. A recommender system can then be developed based on the user’s context and his geolocation data. Some points to make are

  • Distance between users is relative dependent on mode of transport.
  • Trajectory of the user is important when measuring distance since it separates the "past" from the "future" and recommendations should be targetted towards the user's future
  • Context is around the user and the behavior of the device should be context sensitive.

Context-awareness in mobile devices partly derives from research in Xerox PARC on ubiquitous computing nicely summarized in this wiki pageand this blog . Some of the parameters of pervasive computing and economics of communication are described below and come from a great discussion in class.

  • Economics of communication: Communication is of no cost to the sender and is partly the reason for the existence of so much spam. A cost imposed on the sender of the message can impose an additional thought process on the part of the sender that would make him reevaluate sending the message e.g. if it is guaranteed that the receiver will not pick up the phone then calling is not only a waste of time, it is also a waste of money. . Sending huge numbers of e-mail messages costs spammers very little, since the costs of e-mail messages are spread out over the internet service providers that distribute them (and the recipients who must spend attention dealing with them). Thus sending out as much spam as possible is a rational strategy: even if only 0.001% of recipients (1 in 100,000) is converted into a sale, a spam campaign can be profitable. This is due to a the architecture of current systems where the spammer gets the attention of the receiver without paying a price for it
  • Attention Markets Attention economics is an approach to management of information that treats information as a scarce commodity. The concept was first articulated by Herbert Simon who found that designers represented problems as "information scarcity" when they really needed to address the problem as "attention scarcity". An economics based approach is to apply a price to "attention rights". It would be expensive to buy "attention rights" to a CEO but cheaper for a student. Also "attention rights" are not the same for all users e.g a professor's spouse has attention rights for free but the students pay a little more and advertisers pay even more.
  • Recipient Interrupt mangement: With regard to mobile communication a recommendation system is required to indicate the urgency of the call. Urgency of a call does not necessarily satisfy the reflexivity property i.e. if the sender thinks the call is important the receiver does not necessarily think the same way.. The sender might have it as a 10 on his/her priority list but it is not so on the receiver’s list. The first-order approach would be to apply a feedback mechanism to refine the priority viewed by the receiver based on the priority applied by the sender on previous calls and how they were viewed by the receiver. This ties in closely with the concept of attention markets.
  • Sensors:
    • The alarm function in the mobile phone could be used to determine whether the user is asleep or awake. Typically the user sets the alarm before going to bed. Some other parameters that can be used are motion sensing and movement sensing.Calls from the sender can be tagged with some metadata that provides additional information to the receiver.
    • Implicit and Explicit sensing of context. The user could explicitly set their status to a certain state or it could be obtained implicitly based on reception quality.
    • : An interesting application of crowdsourcing is to use surrounding mobile phones as a sensory mechanism. If the surrounding phones are in silent mode then there’s a good chance that the phone is in an environment where disturbance is not welcome. There are some concerns about privacy on this if other cellphones can poll for data on your cellphone.
  • Data Sources:
    • A user's addressbook.
    • The call log and importantly the time of each call, length of each call.
    • The set of other mobile phones that surround a given phone provides context on the situation. Are you with your mom or at a party ?
  • Context-Aware Recommendation systems: Traditional recommendation systems have relied on a user's preference history to make recommendations. However, the mobile phone introduces a new dimension, that of immediate context. The same user can use separate recommendations for a restaurant when he is setting up for a party or when he is chilling outside in the rain.
  • Risks: There are concerns about privacy especially when a person's mobile phone can be polled remotely by other users. Not everybody is confortable with this concept


Student Ideas for Mobile Devices


What can we measure? Mobile phone as a sensor (what do we know?)
1) Today, we are able to know when somebody last touches their phone, and what buttons they're pressing/what actions they're doing. We also know checking time vs. sending SMS.
2) We can model someone's walking speed, the level of sound in the room and outside. Perhaps, create an auto-volume keeping in mind both inside and outside volumes.
3) Combined with the above, we have GPS to find out where and when people are calling whom.
4) Using the address book, we can integrate into social networks so that no need to get friend request confirmation if already talk to them a lot.
5) Can infer driving to turn on/off bluetooth headset (among other things).
6) Some ideas for new sensors: Would like to detect the moisture on the hand of the user (are they sweating), blood pressure, body temperature
7) Cell Phones to Model Pollution, Cell Phones Detect Nuclear Threats

Action based on situation

1) Alarm Setting
Use noise levels, light levels, and movement to infer sleeping
2) Elevators
Why don't the signals change to indicate going up or down (laptops too, why not indicate whether plugged in or unplugged)?
3) Interruptions
Should the phone ring or not? What sort of considerations are there when deciding not to interrupt somebody?
Whose demands are more important, those of the caller, or those of the receiver?
Polling other phones in the vicinity for their volume levels to determine your own?
Perhaps, similar system of auto-reply by sending text message, explicit setting of your status
Birth of a reputation system to control voluntary (self-declared) levels of urgency. Spam can be controlled because networks will quickly learn about where spam originates.
OVERALL: By sending meta-data with your call, phones will create a response function for each user who calls you, and the community will do the same for every user.

Examples of Using Location


Shows Wikipedia sites close to your current location






Find restaurants and bars, etc.
They can now auto get location Lightpole

Outside in

Local news and conversations
A gateway to discovering your community
Outside In


"A location-based mashup designed to find out where your friends are." Integrates the already effective uses of Twitter ("for
messaging and your social/attention network"), Upcoming("for event and place names"), and Fire Eagle ("for location queries and
updates"). No need to re-add friends, they're imported from the previously mentioned apps. Fireball


Fire Eagle app that uses geo-location data to provide local weather forecasts and nearby Flickr photos Firewidgets


"Shoppers using will be able to simply click a button to search local store inventories for the products they like and reserve the items for pickup at over 30,000 participating store locations of over 200 leading national retailers."
Also, new iPhone App for navigating shopping centers Nearby Now


Makes mobile ad impressions, using location data to better target consumers eZee



Like Street View, can actually move through cities
Unlike Street View, you can go INSIDE (i.e. check out a hotel lobby, a restaurant, etc.)
Put advertisements at certain points EveryScape


Easy to access government information that can be relevant and interesting to know but is often difficult to find. EveryBlock

In the Works

Friend locater. Also, how close is your boss? Thesis advisor? Ex girlfriend?
Fire Eagle's Missing Apps

Personalization: is your mobile smart enough for you?

How can we make our devices "smart"?

When one hears the word personalization, many people can describe the omnipotent artificial intelligence of their dreams. But let's go back to the basics for now. Do you ever question the fundamentals as a customer:
  • What do you want personalization to do?
  • What is it good for? efficiency? in what sense? time and speed?
  • Where does if fail by necessity?
  • How is it similar and different to human nature?

Now, there is also the product manufacturers' interests:

  • Why do we want to offer the customer a personalized experience?
  • How is a personalized experience different from a customized experience?

Let's venture into depth one by one.

-0- Definitions: Personalization versus Customization

Together, personalization and customization make users' life easier and make users feel welcomed. They also provide users with relevant content based on their needs and interests. Both qualities foster customer loyalty, which vastly increases the chance of repeated site traffic. Repeated site traffic, in turn, has enormous potential for increasing advertising and sales revenue. [reference]

Thus these two properties are vital to online commerce. But what are they? Customization can be driven by personalization, but they are very different.


System makes suggestions
  • Personalization is the process of providing relevant content based on individual user preferences. If you want to sell books online, it might be helpful to know the types of books a customer likes, dislikes, needs, or wants. Personalized sites can obtain that information implicitly, by tracking customer purchases or usage habits, or explicitly, by gathering information through customer forms or questionnaires.


User defines explicitly
  • Customization takes place when users are able to modify a Web site's look and feel. Many sites provide both customization and personalization features. For example, after registering with the Excite and Yahoo! sites, users can create their own customized start pages by choosing their preferred layout, content, and color scheme. Customer information obtained through the registration process, such as customer names, is also used to create personalized greetings within the customized start pages. Thus, these sites combine customization and personalization features to provide users with the information they need, quickly and easily.

-1- Desiderata: What are desired properties of a personalization system?

Value propositions: What’s in for the user?

What’s in for the website?
  • Lifecycle
    • Example: Show different or more complicated features for experienced user vs. first-time user
  • Up-sell
    • Up-selling is a sales technique whereby a salesman attempts to have the customer purchase more expensive items, upgrades, or other add-ons in an attempt to make a more profitable sale. [reference]
    • More features, understand how far user willing to go (e.g., good salesperson!)
    • What cues beyond clicks will be possible?
    • Interaction / Discourse / Mode switching: ask key questions for decision explicitly (rather than let user build a mental model of the system!)

  • Cross-sell
    • Cross-selling is defined by the Oxford English Dictionary as "the action or practice of selling among or between established clients, markets, traders, etc." or "that of selling an additional product or service to an existing customer". [reference]
    • Market basket analysis
      Association rules of stuff bought (or in shopping basket)
    • Info through navigation

  • Increase relevance
  • Increase simplicity: remove irrelevant stuff
    • Automatically things creeping in
  • Not static, but adaptive (long, slow time scale)
  • Detect current situation (fast time scale)
    • Often not called personalization but sessionization / occasionalization [paper]
    • Analyze users’ moods and how they are using the Web at particular moments.
    • History vs session/situation
    • ASK if it makes sense (why guess when the other one wants to tell you anyways)
  • Understandability / Interpretability
    • Does ROOT need to offer tools to create meaning?
    • Others should do that – let others use “our” data

Make things easier
  • M-Commerce
    • Personalization on mobile has advantages that on a limited screen space, because it knows more the suitation you are in, and it can actually show you information more targeted than a laptop.

Save time, increase productivity
  • Especially on repeated tasks
  • “Customization” vs personalization

Help user discover things
  • Alerts
    • Geo-location alert: we can assume the mobile is usually with the user.
    • Cross modal
    • STW (Surf to work) ratio above 1
    • Worth thinking about

-2- Metrics: How do we measure whether the system does a good / bad job?

1:55:30 on lecture video

Define targets and objectives

  • Profit
  • Number of orders
  • Loyalty
    • Re-purchase within 6 months
    • Attention recorder turned on / off
    • Write line into log
  • Churn
    • Can it be reduced?
  • More fine-grained
    • Number of visits?
    • Measures of engagement
      • Inviting friends
      • Using features to annotate
      • Putting itme on wish list
      • Saving to shopping cart
    • Prob of clicking on personalization feature (clicked once, clicked twice)
    • Prob of buying a recommended item
    • Length of session / total number of clicks
    • How easiy is it to navigate?
    • Does the user understand the recommendation (condoms)?
    • Number of alerts sent out
    • Accuracy of recommendations - determine how to define this
      • Is recommendation understood
      • Is it purchased
      • Is it added to wish list
      • Do items sales go up
    • Feedback to easily correct recommendations
      • for example "Not interested" to recommendations
      • also gives users the basis for the recommendations (a previously purchased or recommended product) and provide users a "Fix this link" to refine their previous ratings
    • Robustness
    • Relevance – who determines relevance?
      • Customer, such as in the condoms example above
      • Need to understand demographics and purchasing history and how they relate to relevance for the customer.
      • Ask user explicitly? YES!!!
      • Does user click at it (short-term)?
      • Does the user return?


Can you do a good job estimating the probabability of purchase?
Examine customer satisfaction
  • Should we do a survey?
    The American Customer Satisfaction Index
  • How do we measure customer satisfaction?
    • Look at time scales: short-term vs long-term
    • Will satisfaction differ dramatically according to personality?

Track more things

  • Especially behavior over time.
  • Find proxies for: satsifaction or dissatisfaction.

-3- Inputs: What data are used to build the model (variables)?

B2B examples

Many business-to-business (B2B) marketers now realize that building a customer-centric company is vital to corporate health and profitability, but even the most superb customer-oriented companies may still be far removed from anything resembling one-to-one customer relationships. A possible reason is that it often means changing attitudes, then business practices, and sometimes even the corporate culture.

Examples of Internet Personaliziation:

  1. Dell computer customizes Web pages to meet the specific needs of its major customers. By customizing these "Premier Pages" for each customer, Dell provides a very personalized experience. More important, this customization makes it easy for customers to always know the discounts and other terms and conditions of their relationship with Dell, thus making it easy to do business with the company.
  2. Another example of highly personalized customer service is the way advises its customers on purchasing. Amazon makes "instant recommendations," suggests items that the customer might be interested in based on previous purchases, and offers ideas for complementary items (e.g., select a computer printer, and Amazon will recommend cartridges and printer cables to go along with it). Amazon also provides "1-Click ordering," which customizes the ordering process so returning customers don't have to re-enter basic data already on file.

Some basic variables for personalization

  • Time of day
  • Day of week
  • Recent purchases
  • Uses of these products when available
  • Implicit data such as user’s intention
  • Search terms
  • Exit a page during online shopping, e.g., Amazon; exit after the task is completed or exit in the middle of the session, presumably finishing shopping somewhere else, because the search doesn’t return relevant/interesting results
  • Source: Http-referrer (how do you start the session, which websites refer a user to your page? Looking at Access log or use Google analytics (homework 2) would be useful for such analysis
  • Languages of the site, can be customized
  • Mobile devices ( “…mobile phone has become the closes thing to us, second to our under wears…” quoted from Andreas)
  • Geolocation: Google has launched location-based personalized search results. For those users who have provided a default location in Google Maps, Google will personalize results based on that location. For example, if a user has entered a default location into Google Maps and types in “library”, the results will bring up the user’s local library.
  • IP Address: IBM customers can be recognized based on their company's IP address, even without registering. IBM can identify the company and personalize tasks and marketing messages based on the company itself, or its industry or other firmographic factors.
  • Shopping history, both browsing and the order history
  • Past price sensitivity
  • Explicit ratings (like/dislike), such information could be used as basis for recommendation system; key distinction: Passive observation vs. active collection

Personalization versus Customization Models

  • Implicit: customer has no choice

  • Explicit set of choices, but people are not doing it unless they see what’s in for them
  • Disambiguates the otherwise unary bought vs don’t know

Amount of inference

  • Users are willing to spend time fixing things if they see, in a relatively short time scale, that value comes in for them; to make it explicit for them or to make complicated models for them
  • From just storing past searches to having complicated models
  • Amazon used to have a “show me” button why you are recommending this product to me? Which question is more interesting to user: what gets recommended to me or why Amazon is recommending this item to me?
  • Assumptions about the user behaviors? Some are correct and some are not true.
  • Amazon allowed users to have different passwords for the same account: one password for school use and the other for daily life; by doing so, it assumes two different entities

Multi-modal personalization
  • Can view as schizophrenic, multiple personalities Either user explicitly logs in with different identities, or sorts actions to belong to identities, or


  • Dating site: Room-mate, visit, hook-up, LTR
  • Availability of historical data

Personalization by definition based on persistent data

  • Past history may or may not be good indicator of the future decisions
  • Perform poorly for "spur of the moments" or "occasional sense of adventure"

Range of data collection
  • Privacy
  • See Lecture 8 on June 2, 2008

  • Only current session is available
  • Entire history is available
  • How much is sufficient yet necessary?

Other questions we need to address
  • Do we want to enable the user to edit their history?
  • What people edit out is informative Or at least turn it on or off?

-4- Outputs: What does the model produce (predictions)?

Mode: Push vs. pull for delivery of the result

Example 1:

Google thinks of personalization in 3 parts:
1. Search your own stuff (like Google Desktop Search, Web History)
2. Traditional Search (Pull)
3. Push Search (like recommendations, iGoogle/ personalized homepage
For more information, visit ****

Example 2: The Push e-mail system implemented by many Blackberry phones. Push e-mail is used to describe e-mail systems that provide an "always-on" capability, in which new e-mail is instantly and actively transferred (pushed) as it arrives by the mail delivery agent (MDA) (commonly called mail server) to the mail user agent (MUA), also called the e-mail client. Traditional e-mail access over network connections was and still is Pull based: at login and later in intervals, the Mail User Agent (e-mail reader) polls the Mail Delivery Agent (server) to see if there is new mail, and if so downloads it to a mailbox in the user's home directory.

Exploration versus exploitation

Abstract from paper “Capturing User Interests by Both Exploitation and Exploration”:
“Personalization is one of the important research issues in the areas of information retrieval and Web search. Providing personalized services that are tailored toward the specific preferences and interests of a given user can enhance her experience and satisfaction. However, to effectively capture user interests is a challenging research problem. Some challenges include how to quickly capture user interests in an unobtrusive way, how to provide diversified recommendations, and how to track the drifts of user interests in a timely fashion. In this paper, we propose a model for learning user interests and an algorithm that actively captures user interests through an interactive recommendation process. The key advantage of our algorithm is that it takes into account both exploitation (recommending items that belong to users’ core interest) and exploration (discovering potential interests of users). Extensive experiments using synthetic data and a user study show that our algorithm can quickly capture diversified user interests in an unobtrusive way, even when the user interests may drift along time.”

For more information about this article, follow this link:

Another reference for those who are really interested is
"Exploration/exploitation in adaptive recommender systems"

-5- Acceptance: Customer's reactions

Our main goal is to attract customer loyalty, and not offend them and drive them away. With this in mind, these questions are obvious to ask:
  • What does the customer like?
  • Can we treat multiple customer as one group?
  • What scares them?
  • How far can we push?

Also be careful to keep in mind about the alternative of no customization and no personalization. How will the world look like then?

Psychology of customization: the individual customer?

The main component of customization is the involvement of customers. Of course this is vital because the customers are the ones controlling (at least certain aspects of) the end product. In modern era, internet commerce allows convenient access to customization and enhancing it in many ways:
  • Ease of access from home
  • Massive options generated on a computer
  • Readily available and instantly updated product information
  • Absence of "intrusion" from salesperson

Further readings:
  • "Online Consumer Psychology", C.P. Haugtvedt, K.A. Machleit, R. Yalch, 2005, ISBN:0805851550
  • "Service Customization Through Employee Adaptiveness", Gwinner et al., Journal of Service Research, Vol 8, No. 2, 131-148 (2005)

Example: what drives people to get special ringtones?

People are much more robust when they customize something than having someone else personalize for them. Consider your mobile ringtone, obviously you want to express yourself, but in what way?
  • Customization: appeals largely to those who want to express the uniqueness of themselves
  • Personalization: customers are told how to be themselves; idols are created for the public to imitate

Psychology of personalization: pressure of the public?

Are you being "personalized" by the product?

It has been said in class that one reason to consider using Facebook may be that one's friends all migrated together to Facebook. Is the public's decision the appropriate one to follow after? Certainly it would be more appealing if instead of the general public, one's close friends and family want the same thing.

Example: why are you hooked to Facebook?

John Kirriemuir provides six points on why Facebook has such popularity:

  1. Collecting: putting together your social networks Personalization
  2. The paranoia of "not being invited to the party": "Look! I have friends on Facebook" Customization
  3. The voyeurism: digging data from everyone's profile you can see Personalization
  4. Diverse demographics: meeting different people, and they can be very different Customization
  5. The autobiography: relationships write your history for you Personalization
  6. Customizable applications: all these things to play with Customization__

Refer to the original for more details:

Pay-off matrices: individual differences in false-positives, false negatives

The product designer needs to understand where the consumer's mind set. A decent pay off matrix requires the following factors at least:
  • Negative performance effect factors
    • How much does the consumer get annoyed if we show something totally irrelevant to him/her?
    • How much does the consumer get annoyed if we missed something extremely attractive to him/her?
  • Positive performance effect factors
    • How does this particular individual's loyalty grow with each relevant and satisfactory recommendation?
    • What is the expected error the consumer may passively allow?

Privacy: how do we mine information?

To accurately propose solution to the consumer, a device needs to know certain information:
  • Geolocation
  • Age
  • Gender
  • Past preferences
  • Much more

What does the customer consider personal or private, as opposed to public? How does the device obtain these informations?
  • Blog: customized data; public
  • Anonymity: is this data useful at all? It is public, but we have no information to link it with any customer, thus private?
  • Pseudo-anonymity: are we really hidden when posting information online? Who has access to IP address traces?

Different people deal with private data very differently. Can we still treat a group of customers with one set of policy? Or do we necessarily need to look at individuals on his/her own?

See Lecture 8 on June 2, 2008 for more on privacy!