Information filtering in the Internet
 |
Written: 17/7/97
updated: 27/7/97
Version: 1.3
Author: Boaz Chen |
|
168 Kb |
| |
Background
The Internet, since its inception, has been global and "anarchic",
promoting a wide range of activities and freedom of expression
and information. On the other hand, as it has expanded and millions
of new subscribers have been added, the problematical issues associated
with this policy have surfaced. This is expressed by the type
of sites, discussion groups and IRC channels that include hard
pornography or incitements to violence, racism, terror and others.
Attempts to pass laws against the criminal use of the Internet
(based on existing laws for other types of media) have met with
strong opposition as well as problems of enforcement.
The issue has become the topic of community discussion due to
the increasing numbers of children and teenagers who use the Internet
and with the intention of protecting them from information of
this nature.
The Problem
Attempts to bridge the gap between freedom of expression and conservation
of the values we cherish have resulted in a search for solutions
that will enable filtering of information.
The requirements for filtering information are different and varied,
for example:
- Protecting children and youth from pornography, violence,
racism
- Prohibiting entry to amusement sites from the workplace
- Prohibiting entry to types of services or sites during peak
hours
and so on.
Mechanized Filtering
The original solutions put forward were based on analysis of information
during attempts to enter the site. This refers to programs that
check the information before it is displayed to the user, to ensure
it meets certain prescribed standards. For example: programs which
scan the text to ensure the non-appearance of specific words or
programs which check for the amount of uncovered flesh before
displaying pictures.
Solutions such as those have a number of difficulties:
The solutions are specific to particular criteria and will not
be compatible with information transferred or presented in a different
format. For example, undesirable text that is displayed as a picture,
program or even in a different language, will not be effectively
filtered.
Mechanical filtering is lacking because of the difficulty in properly
defining cross-sections and the overall context. Here are a number
of examples:
Programs that prevent the entry to a site containing the word
SEX will also prevent academic information on the subject and
even sites which contain sexual advice for teenagers. On the other
hand, sites that only contain pornographic pictures will not be
filtered out.
Attempts to prohibit entry to racially prejudiced sites will often
inhibit attempts to enter anti-racially sites as well.
Programs that filter pictures will also filter out artistic ../images
as well as medical, instructional pictures etc.
Checking of information at the time of loading slows down the
data acquisition process.
The natural difficulty in defining filtering thresholds and the
intense difficulties in translating them into programs, are made
harder due to the difference in approaches and cultural backgrounds
of the peoples in the virtual world.
The Classification and Labeling Solution
An innovative approach to this problem, which we will discuss
in detail in this survey, is based on human classification and
evaluation of information.
The intention is, in order to ease the filtering process, to check
the information, grading and labeling it based on predefined criteria
and a format which enables mechanized use of these labels.
This is similar to the classification methods already in use in
the fields of movies and videocassettes, newspapers or books,
except that the programs that will base their filtering criteria
on labeling will enable monitoring and enforcement while making
the filtering suited to the user.
In general terms, there are three elements involved in the process:
the distributor of the information, the evaluator responsible
for classification based on standards and the user whose filtering
program checks the information's labeling and decides whether
or not to allow its retrieval.
This is shown in the diagram:
This approach, which solves many of the problems associated with
the automatic analytical tools, creates new dilemmas and even
the suspicion that it is creating censorship. The major questions
it raises are:
- Who is responsible for the classifications?
- According to what categories will the information be classified?
- What will be the criteria used for grading within each level?
- Who will set the filtering parameters?
These questions and others made the World Wide Web Consortium1
to set new standard called PICS2 (Platform
for Internet Content Selection).
The standard defines two things:
Standard label structure - technical structure that defines
the format of the label.
A standard structure for the labeling server - definition
of how the labeling service provider defines the categories upon
which labels are based, the various levels of division within
categories and the interface for presentation of grading criteria.
These definitions are completely technical and as such enable
disconnection between the world of content of the classification
and the technical realization of it. (Similar to what the HTML
standard has done for www pages).
This definition, enables different authorities to make classifications
without the dependence on filtering programs, and vice versa.
Software developers will be able to develop filtering mechanisms
without having to comply with a specific classification authority
or standard.
In this way, the standard planners hope to maintain the atmosphere
of democracy in the Internet, so every user will be able to choose
the software, classification authorities, sorting category and
levels in each category he wishes to use.
For example:
A user will be able to connect with an authority specializing
in the fight against racism and choose which level in each category
constitutes a limitation, while making use of the criteria distributed
by that authority. Such a choice might look like the following:
| Category |
Level |
Criteria |
| Type of publisher
|
3 |
Institutions
and organizations, no private information |
| History |
5 |
Show all historical
fact |
| Language |
2 |
Enable condemnatory
nicknames and jokes, but stop at that. |
The information labeled by this authority will be displayed to
the user only if it meets the standards (it is possible to define
how to deal with totally unlabeled information).
Labels
The label is a formatted information, that is to say, information
that is readable by a computer and therefore usable by a filtering
program. The standard does not define the types of information
in the label and, by this, it enables labeling according to different
categories. For an example of a technical application of a label,
see Appendix A.
Labels may be used in two main ways:
Embedded in the information page - the label may be included as
part of the page header. This is applicable to those who wish,
to cooperate with the classification authority (a good reason
for doing so would be the fact that in most filtering programs
it is possible to disable access to all information which has
no classification).
Separated - It is also possible to keep groups of labels, which
relate to information all across the Internet, in different servers
separate from the information. The technical implementation of
this comes when, at the time of accessing the information, the
filtering program also checks the server containing the labels
and searches to confirm the existence of the appropriate label.
In this way, classification of information whose distributors
do not want to cooperate can also be achieved.
The advantage of the first method is that the classification function
can be carried out as a self-initiated or even independent basis
and changing the location of the information does not influence
it. The authenticity of the labecan be confirmed by use of a digitized
signature on the label, by its creator.
The second method has two advantages:
Independence between the publisher of the information and its
various evaluators. The publisher of the information does not
have to do anything for the information to be labeled. (Or even
to know that it has been done).
Prevention of increasing in size of all files containing information.
This is especially important when there are a large number of
evaluators. That is to say, rather than adding a large number
of labels to the information, each user can take the labels he
requires from the appropriate classifying server. The ability
to manage this procedure can be increased by concentrating all
these labels together and copying them all into the local server
(caching). This has the added advantage of enabling storage of
local labels that are sensitive to the language and norms of the
user's location.
Every label is relates to a URL and therefore labels can refer
to information in sites and HTML pages, FTP sites and specific
files, CHAT servers, rooms and notices posted on them and the
like.
Classification
Classification can be done by the information's distributor (by
independently adding a label to the document) or by an external
authority.
In addition to defining the label format, the PICS also defines
the method of transfer of data between classification servers
and the programs that use them. The server contains a vocabulary
that defines the types of categories that are used to label sites
(every "word" constitutes an evaluation dimension for the site
being labeled). For a technical definition of the classification
element, see Appendix A.
In this manner, every program will be able to work with every
server, by defining the categories of external labeling in it.
For example: The Recreational
Software Advisory Council which classifies computer games,
decided to classify them according to five levels of each of the
following dimensions: violence, nudity, sex and foul language.
The classifiers are generally experts in a particular field (software,
games, databases, etc.) or specific types of dangers (racism,
violence, fraud, etc.). It is reasonable to expect that authorities
that are already doing similar type of classification in different
environments will probably provide some of the classification
services.
The two types of classifier can be distinguished by:
Professionals - companies or organizations who employ subject
specialists, e.g: SafeSurf
Democrats - authorities that classify based on evaluations
from other users. This method enables the receipt of evaluation
of information based on what similar thinking people have said
about it (based on similarities with parallel evaluations).
Filtering
The filtering programs are those which, upon requesting information,
analyze the related labels and prevent the viewing of information
which has not been authorized (according to the user's preferences).
There are three main methods of filtering:
Filtering by Browser
The Microsoft
Internet Explorer Ver. 3.02 already supports filtering using
PICS and it is anticipated that the Netscape Communicator will
also support it in its forthcoming version.
In one of the option screens, it is possible to determine, based
on the type of categories of different authorities, the threshold
levels beyond which the browser will not allow to display information.
This is an inexpensive solution (free of charge) and does not
require special installation. It is though limited:
1. The filtering of the information is relating only to that that
is brought by the browser, and there is no protection against
information that arrives via IRC, FTP, e-mail or discussion groups.
2. The restrictions can be override by installing a different
browser.
Designated Filtering Programs
Comprehensive software package that monitors the information flow
in and out of a computer. These programs enable control of other
types of information as well, by filtering the computer's communications
channels (accordance to the user's preferences).
The programs also support individualized definitions for the enabling
access filtering to the different sources of information according
to different user profiles.
These packages generally include additional safety devices, for
example, CyberPatrol3 (today's market
leader) which enables the blocking of transmission of one's personal
particulars (name, address, credit card number, etc.), even during
the course of IRC conversation, definition of the permitted hours
of activity per user, locking of local computer applications (accounting
programs, for example) and protection against attempts to cancel
or erase programs.
Software programs of this nature are the best solution for the
home and private businesses.
Filtering by the Server
Software programs similar to those mentioned above can be purchased
as server versions. These programs connect to the proxy server
and enable filtering for everyone using the network. This capability
is very important for large organizations who wish to prevent
improper use of the network during working hours, or to enable
such use only during restricted hours (breaks, night shifts, etc.).
The advantage of such an installation is the ease of maintenance
since it eliminates the need to install and update versions in
end-user stations.
Large organizations can also form their own independent evaluation
authorities to sort and label, in similar manner to the general
authorities, using categories and criteria which are suited to
their organization. This makes it possible, by using standard,
universal tools, to sectionalize information even within the organization.
Another advantage in the use of filtering mechanisms of this type
is the adding of additional defenses against viruses or other
programs that are found on the Internet.
It should be remembered that the addition of programs is likely
to somewhat slow the connection to the network because it serves
as the bottleneck for the flow of information.
Additional Uses
The motivation for the labeling mechanism was primarily designed
to avoid accessing negative information but in the near future,
it is expected that the use of labels (in PICS standard) will
develop for other purposes in the fields of sorting and organizing
information on the Internet.
This expectation arises from the fact that the more the information
and the number of users of the Internet increase, the problem
of searching for information on the Internet becomes one of qualitative
location of relevant information.
Examples of uses on the horizon:
Qualitative assessments of sites and information on different
subjects - similar to what exists at the moment in a number of
catalogs. It will be possible to obtain in future, information
about the quality of the site and the kind of information it contains.
This will be done by organizations who create labels with this
information without their dependence on specific sources for their
information.
The addition of the information contained in the labels will enable
the search engines to work with greater abilities. This ability
will enable more efficient searches integrating the mechanized
abilities for locating of information with human evaluation.
The use of labels for classifying users in discussion groups or
messages posted from them. In addition to the filtering which
is mentioned above, it will be possible to evaluate the message
writers based on their messages. Filtering categories such as
professional interest level, style, level of succinctness and
more will enable the user to filter the messages by definition
of thresholds.
And so on.
The Fly in the Ointment
The process of labeling, and the filtering abilities described
above constitute a degree of danger to two major principles of
the Internet:
Censorship - in spite of the standardization of the labels, and
intent to establish a free market of different evaluators and
classifiers, there are those who claim that a number of large
organizations will control this field and establish a kind of
elitism that controls the accessibility of the information on
a global basis.
Accessability- resources required to carry out human labeling
of the information and updating of this labeling, will mean that
not every site could be labeled. It is expected that the use of
the possibility to prohibit all unlabeled information from being
accessed will create a situation wherein access to minority group
sites will be barred. The same goes for non-institutionalized
sites, whether this was the original intentional or not. This
will affect the special character of the Internet, in which every
site is accessible worldwide.
Summary - Filtering as a Service to Government
Offices
The major advantages of the Internet arise from the great accessibility
of information in it to users worldwide. In our process of evaluation
of this working tool, the advantage is also shown to be a disadvantage
inasmuch as the existing organization of most institutions means
that they are unable to monitor the use of this tool or to enforce
regulations concerning its use.
Like all organizations, the various Government offices can use
the filtering possibilities that have been described herein for
the purpose of solving this problem, should it exist.
The Government Internet Committee decided neither to issue policy
guidelines nor to enforce use of this nature. This is because
the decision to use this tool and the method of filtering by it
are issues that should be determined by the individual offices.
In the event that a Government office does decide to use the Internet
in this way, we recommend that it should be done by purchase of
specialized filtering software, which supports PICS, on the office
server. Use of this nature will enable central regulation and
bring about the greatest degree of efficiency.
Resources
We recommend the sites motioned in the footers and in addition:
Article in the Scientific American by Prof. Paul Resnick, Chairman,
PICS
www.sciam.com/0397issue/0397resnick.html
Article in Hot Wires, by Simson Garfinkel
www.hotwired.com/packet/garfinkel/nc_today.html
PICS newsletter on the organization W3C
www.w3.org/PICS
www.w3.org/pub/WWW/PICS
Appendix A. - Technical
Definitions
The technical definitions of the PICS standard are based on text
with the intention being to make them easy to construct and update.
As we have already mentioned, the standard defines two elements:
· The authority that classifies definition of vocabulary or categories
based on which its labels are defined.
The label itself.
For demonstration purposes, we will characterize definitions and
type based on the American standards for movie classification
(MPAA).
Classification Factor
In the URL address which defines the classification should be
part of the page which includes the text:
| The Text |
The Meaning
|
| ((PICS-version
1.1) |
|
| (rating-system
"http://MPAAscale.org/Ratings/Description/") |
Site address
of access |
| (rating-service
"http://MPAAscale.org/v1.0") |
Organization
site address |
| (icon "icons/MPAAscale.gif")
|
Organization
Symbol |
| (name "The
MPAA's Movie-rating Service") |
Organization
Name |
| (description
"A rating service based on the MPAA's movie-rating scale")
|
Organization
Description |
| |
|
| (category |
|
| (transmit-as
"r") |
Definition
of Category (word) |
| (name "Rating")
|
|
| (label (name
"G") (value 0) (icon "icons/G.gif")) |
For each (numerical) level of the category the name is defined
(or the literal description of the criteria) and a symbol
which shows it for graphics purposes in the filtering programs
|
| (label (name
"PG") (value 1) (icon "icons/PG.gif")) |
| (label (name
"PG-13") (value 2) (icon "icons/PG-13.gif")) |
| (label (name
"R") (value 3) (icon "icons/R.gif")) |
| (label (name
"NC-17") (value 4) (icon "icons/NC-17.gif")))) |
The Labels
The following text can be included in the headers of a page of
information or separately.
| The Text |
The Meaning
|
| (PICS-1.1 |
|
| "http://MPAAscale.org/v1.0"labels
|
Type of label
(organization of filtering which will define it) |
| no "1996.11.05T08:15-0500"
|
Date it became
valid |
| until "1997.12.31T23:59-0000"
|
Date it became
invalid |
| for "http://www.company.com/filmname.htm"
|
Information
referred to (enables separate placement) |
| by "John Doe"
|
Evaluator |
| ratings (r
2)) |
Rating based
upon each category (one in this case) |
1. The World Wide Web Consortium (W3C for short)
is a joint venture of industry and universities in the USA, Europe
and Asia, for development of protocols
2. For details see www.w3.org/pics
or www.w3.org/pub/www/pics
3.The program includes a list of recommended and
forbidden sites, and thereby is acts as a labeling authority as
well - www.cyberpatrol.com
|
| : |
|
