Dietmar Janetzko Nonreactive Data Collection on the Internet

1
Dietmar JanetzkoNon-reactive Data Collection on
the Internet

• Dr. Dietmar Janetzko
• National College of Ireland
• Mayor Street, IFSC, Dublin
• Telephone [phone]
• Fax [phone]
• Mobile [phone]
• E-Mail dietmarja_at_gmail.com

2
Outline

• The Concept of Non Reactive Data Collection
• The Technical Perspective
• The Methodological Perspective
• Thin and Rich Descriptions
• Two ways to deal with thin descriptions
  Horizontal and vertical enlargement of data sets
• Extensions Recent Developments
• The Enron Data Set (E-Mail)
• The AOL Data Set (Search Requests)
• Collection all Data about a Person
• Discussion

3
The Concept of Non Reactive Data Collection
4
The Concept of Non Reactive Data Collection

• Non reactive data collection is conducted in a
  naturalistic setting in such a way that persons
  studied are not aware of it.
• Thus, non-reactivity is not a characteristic of
  the data or the data collection procedure per se,
  but of the awareness of the persons (not)
  studied.
• Three kinds of non reactive data
• Environmental (PhysicaL) Traces
• Simple Observations
• Archival Sources (Frankfort-Nachmias Nachmias,
  2000).

5
Motivation for Non Reactive Data Collection on
the Internet

• A) Why non reactive?
• B) Why using the Internet/Internet technologies?
• C) What are the limits of using non reactive data
  
• collected on the Internet?

6
Why non reactive?

• Often, a reactive equivalent to NRD does not
  exist. It would be cumbersome to develop or it
  would severely interfere with the phenomena
  studied.
• The phenomenon of interest would be distorted or
  disappear if studied in a reactive way
• Example Studying dating on the Internet via
  reactive measures would defeat its purpose and/or
  would be open to criticism that an unsuitable
  method has been used.

7
Why using the Internet/Internet Technologies?

• Today, many social phenomena (e.g., communication
  in organisations) unfold especially or even
  exclusively via the Internet.
• NRD collected on the Internet highlight
  behavioral social phenomena, but is also
  indispensable for organising online research.
• Example Using cookies, IP-Addresses or time
  stamps to control if persons participate several
  times in an online study
• Data collection
• is relatively simple
• not limited to a fixed area/time
• may yield Data may be collected in large
  quantities
• can be done in an automated and objective way
• may cover sub-symbolic information (e.g.,
  hesitations to make a decision, Hofmann, Reed,
  Holz, 2006)

8
Limits of Non Reactive Data Collection on the
Internet

• Many techniques used for NRDCI have not been
  designed for online research studies in the first
  place.
• Example Log files have been devised to allow
  technical staff to control the proper working of
  systems like web server
• NRDC techniques facilitate studying a very small
  part of the spectrum of behavioral or social
  phenomena.
• Person characteristics like appearance, height
  and weight, attire, gender, age, ethnic group,
  facial expressions, eye contact, body language,
  gestures and emotive responses are filtered away
  (Dholakia Zhang, 2000).

9
Perspectives on Non Reactive Data Collection on
the Internet

• Non reactive data collection on the Internet can
  be
• viewed from different vantage points, e.g.,
• Technological Perspective
• Privacy Perspective
• Methodological Perspective
• Data Mining Perspective (Web Usage Mining)
• Commercial Perspective (e.g., marketing)

10
Technological Perspective
11
Techniques used for Non-Reactive Data Collection
on the Internet
12
Client Server
Client
Client
Web-Server
Client
Client
13
Client Server (cont.)
Browser asks (requests) a document from a server
available via a URL (Unique Resource Locator)
Server allocates values to environment variables
Server retrieves the document (usually HTML
code), generates a header, sends it to the client
Server writes entries into its log-files
Client presents the HTML code in a readable
way (renders the HTML code)
Browser may also call and thus cause to execute a
program (-gt CGI)
If the access fails, the server will generate
error-reports instead
-
Client-Server Interaction follows HTTP (Hypertext
Transfer Protocol)
14
Environment Variables

• To pass data about the information request from
  the server to the script, the server uses
  environment variables as well as the standard
  input and output streams of a CGI-script.
• Environment variables are set when the server
  executes the gateway program. There are some
  environment variables set for request-specific
  and some of set for all requests.
• Selection of some environment variables
• SERVER_SOFTWARE
• SERVER_NAME
• REQUEST_METHOD
• REMOTE_ADDR

15
Environment Variables (cont.)

• The Common Gateway Interface (CGI) is a standard
  for external gateway programs to interface with
  information servers such as HTTP servers. A HTTP
  server usually supports all environment variables
  of the CGI-Version to which it complies.
• The current version is still CGI/1.1. The
  CGI/1.2 ("Next Generation") Specification is
  still in the limbo.
• http//www.w3.org/CGI/
• http//hoohoo.ncsa.uiuc.edu/cgi/intro.html

16
Log Files

• Servers (HTTP Server, Web server) or clients keep
  a track of system or user activities by
  generating so-called log files. Log files provide
  valuable information, e.g., on the security of
  the server or the activities of the user
• There are different types of log files
• Standard log files that follow a specific format
• Vendor-specific Log Files
• Client-Side or Server-Side Log-Files
• Specifically Tailored Log Files

17
Log Files

• There are different Types of Standard Log Files
  generated by a HTTP Server (Web server)
• Access/Transfer Log information about who visited
  a site
• Error Log information on the errors that
  occurred while accessing the server
• Referer Log information on the source that
  referred a visitor has visited before
• Agent Log information on the clients
  browser and operating system

18
Log-Files / Entries (Tokens) of Log-Files
(Selection)
AG Type of Browser used (Agent) B Number of
Bytes transferred from Server to
Client BR Number of Bytes transferred from
Client to Server D Data/Time of the
Request S Service Requested H Clients domain
Name or IP-AddressI Identification of the User
on the Client Side NTSC Status Code (Win
NT)O Operation carried out (e.g., GET) P Files
(including Path) requested SA IP-Address of the
ServerSC Status CodeSN Name of the
ServerREF URL of the Site where the Client has
been immediately before
19
Standard Access Log Formats
20
IP Addresses

• Every computer connected to the Internet has a 32
  Bit IP or Internet Protocol address. It consits
  of 4 octets (Bytes) separated by dots (e.g.,
  192.168.1.1).
• IANA (Internet Assigned Names Authority,
  http//www.iana.org/) is responsible for the
  world wide administration of IP-Addresses
• An IP-Address is unique, but one computer may
  have several IP-
• Addresses one for each connection to the
  Internet. Vice versa, one
• IP Address can be used by several computers to
  access the Internet.
• Behind one computer there may be a complete
  network. This is the idea of a gateway. The
  gateway has a address that is visible from the
  outside. Thus, other computers within the network
  are not visible.

21
Cookies Introduction

• What is a Cookie?
• A cookie is an element of data that a Web site
  can send to your browser, which may then store
  it on your system. You can set your browser to
  notify you when you receive a cookie, giving you
  the chance to decide whether to accept it.
  Source http//www.w3.org/2001/10/glance/doc/priv
  acy.html
• Why are Cookies so popular?
• Not the kind of information per se that is
  managed by cookies makes them interesting. This
  means, cookies do not give youa privileged
  access to some pieces of information that you
  cant access via other techniques. What makes
  cookies interesting is the kind of information
  management they allow.

22
Non-Persisten Cookies (Session IDs)
Session identification URIs permit HTTP
transactions to be linked within a limited
domain. This allows a content provider to track
activities within sites on their network but does
not permit data from different sites to be
correlated without specific user authorization in
advance. Example of a session-idhttp//www.sun
.com/2000-1121/wlc/sessionidAY2D5XQAAB42RAMTA1L
U45Q http//subscriptions.sun.com/optin?id7289675
917258240725 http//www.amazon.com/exec/obidos/sub
st/home/home.html/103-6371678-0789449 You will
often see session-IDs as a string of numbers in
the browser address bar. These numbers will track
you via cookies and serve pages specific to your
"session". A session can be any time limit and
then it expires. Sites use these sessions to
serve custom content, defeat browser caching, and
to direct the flow of visitors through the
website.. http//www.webmasterworld.com/glossary/s
ession_id.htm http//www.w3.org/TR/WD-session-id.
html
23
Generation of Cookies

• How are cookies generated?
• Cookies can be sent by the server with a
  HTTP-responseor they can be set by a server-side
  (CGI) or client-side (JavaScript) program.
  JavaScript can also be used to read cookies - in
  accordance to the limitation of cookie usage.
• There are different types of Cookies
• Persistent Cookies vs. Non-persistent cookies.

24
Methodological Perspective
25
Thin Descriptions Rich Descriptions

• Usage of NRD leads to thin descriptions.
• Like all behavioral data, NRD gives no access to
  internal states.
• Example E-mail logs might indicate the
  intensity of a relationship between two
  communication persons. In itself, however, they
  do not indicate why they communicate in the first
  place. Likewise they do not reveal the content of
  the communication.

26
Two ways to deal with thin descriptions
Enlargement of a 2-dimensional Data Set
Horizontal Enlargement(data enrichment)
Vertical Enlargement
27
Two ways to deal with thin descriptions (cont.)

• Vertical Enlargment
• Merging same-format data of different sources
• Horizontal Enlargement
• Merging data / Triangualation (e.g., Webb et al.,
  2000)
• Inferring attributes
• Example Horizontal enlargement may violate the
  users privacy, e.g,
• when click-stream information is linked
  registration information. In this
• way data become personally identifiable

28
Combining two ways to deal with thin descriptions

• Example How can one predict the size of a
  household on the basis of
• the web sites viewed?
• Record the web sites viewed, the time spend etc.
  (non-reactive)
• Find out the size of the household (reactive)
• Record the web sites viewed, the time spend etc.
  (non-reactive)
• Predict the size of the household

29
Two ways to deal with thin descriptions (cont.)

• Vertical Enlargment
• Merging same-format data of different sources
• Buying addresses
• Horizontal Enlargement
• Merging data (online online, e.g., registration
  information online offline, e.g., operational
  data)
• Infering attributes
• GOFAST, e.g., regression analysis,
• data mining, e.g., probabilistic techniques,
• others, e.g., affinity scoring

30
Extensions and Recent Developments
31
The Enron Data Set (E-Mails)

• In December 2001, the Enron Corporation, an
  American energy company based in Houston Texas,
  collapsed and had to declare bankruptcy.
• Originally made public by the Federal Energy
  Regulatory Commission as part of the legal
  proceedings against the Enron Corporation.
• The data cover a huge collection of real e-mail
  messages sent and received by employees of the
  Enron corporation.
• The data set was purchased by Leslie Kaelbling of
  MIT, who discovered that it had integrity
  problems.
• People at CMU, led by Melinda Gervasio corrected
  these problems and deleted too sensitive/personal
  e-mails.
• Distributed in its present form by William Cohen.
• http//www.cs.cmu.edu/enron/
• The Enron Data Set has become a kind of
  Drosophila for data mining researchers who want
  to use non reactive data.

32
The AOL Data Set (Search Requests)

• In August 2006, AOL (America Online) published a
  huge data set of search requests of 650,000
  subscribers. Making this data set public was
  motivated partly in compliance to requests by US
  state authorities, partly due to errors by
  employees (Wray, 2006).
• The data have been sorted by anonymous user IDs.
  But soon it
• became obvious that it is possible that the
  queries in the data set can be traced back to the
  persons that entered them (Barbaro Zeller,
  2006).
• As a consequence, AOL quickly closed down the Web
  site where the data has been published.
• In the meantime, the data set has been downloaded
  several hundred times. A number of mirror sites
  have been set up such that the data is in fact
  available.
• The AOL data set provoked a debate among the
  privacy The number of things it reveals about
  individual people seems much too much. In
  general, you dont want to do research on tainted
  data. (Hafner, 2006)

33
Collecting all Data about a Person

• A number of projects work towards tracking a
  persons entire
• existence
• DARPAs LifeLog Project (2003-2004)
• Cancelled for an unknown reason. It is possible,
  however, that LifeLog is still, but clandestinely
  still in development.
• Microsofts MyLifeBit Project
• ACM Workshop on Continuous Archival Retrieval
  of Personal Experiences (CARPE)

34
Vannevar Bushs Memex Vision (1945)

• A memex is a device in which an individual
  stores all his
• books, records, and communications, and which is
  mechanized
• so that it may be consulted with exceeding speed
  and
• flexibility
• Full-text search, text audio annotations, and
  hyperlinks

35
A Personal Transaction Processing System for
Everything Inspired by Memexwww.MyLifeBits.com
36
MyLifeBits Software
MyLifeBits store
database
37
Discussion

• Often, when studying social phenomena on the
  Internet, there is hardly any alternative to non
  reactive data.
• Non reactive data may shed light on new social
  phenomena and facilitates studying the inner life
  of institutions
• There are, however, many challenges
• Turning thin data into rich and meaningful data
  (horizontal/vertical enlargement of data sets,
  usage of data mining techniques)
• Addressing privacy issues carefully.
• Data catastrophes (Enron, AOL) may give insight
  into social processes and/or institutions but
  raise severe ethical questions.

38
References

• Dholakia, N., Zhang, D. (2000). Online
  Qualitative Research in the Age of E-Commerce
• Data Sources and Approaches. Forum Qualitative
  Sozialforschung / Forum Qualitative Social
  Research, 5 (2). Retrieved September 4, 2006 from
  
• http//www.qualitative-research.net/fqs-texte/2-0
  4/2-04dholakiazhang-e.htm.
• Frankfort-Nachmias, C., Nachmias, D. (2000).
  Research methods in the social
• sciences (6th ed.).New York, NY Wadsworth.
  Hofmann, K., Reed, C., Holz, H. (2006).
• Unobtrusive Data Collection for Web-Based Social
  Navigation.In Workshop on the Social
• Navigation and Community-Based Adaptation
  Technologiesin Conjunction with Adaptive
• Hypermedia and Adaptive Web-Based Systems
  (AH06) June 10 20th, 2006, Dublin, Ireland.
• Hafner, K. (2006). Researchers Yearn to Use AOL
  Logs, but They Hesitate. New York Times, August,
  23.
• Webb, E. J., Campbell, D. T., Schwartz, R. D. D.,
  Sechrest, L. (2000). Unobtrusive
• measures. Thousand Oaks, CA Sage.
• The Enron E-Mail Data Set
• http//www.cs.cmu.edu/enron/
• Environment Variables on HTTP Servers
• http//publib.boulder.ibm.com/infocenter/iseries/v
  5r3/index.jsp?topic/rzaie/rzaieenvvar.htm