`Collaborative Help
`
`Mark S. Ackerman
`David W. McDonald
`of Information
`and Computer
`University
`of California,
`Irvine
`Irvine,
`CA 92717
`dmcdonal} @its.uci.edu
`.edu/CORPS/ackerman
`
`Science
`
`,html
`
`Department
`
`{ackerman,
`http://www.ics.uci
`
`ABSTRACT
`to a
`solution
`a collaborative
`examines
`This
`research
`common
`problem,
`that of providing
`help to distributed
`users.
`The Answer Garden 2 system provides
`a second-
`generation
`architecture
`for organizational
`and community
`memory applications. After describing the need for Answer
`Garden 2’s functionality, we describe the architecture
`of the
`system
`and
`two
`underlying
`systems,
`the
`Cafe
`ConstructionKit
`and Collaborative
`Refinery.
`We also
`present detailed descriptions
`of the collaborative
`help and
`collaborative
`refining
`facilities
`in the Answer Garden
`2
`system.
`
`KEY WORDS:
`work,
`cooperative
`computer-supported
`memory,
`community memory,
`corporate
`organizational
`memory, group memory,
`information
`refining,
`information
`retrieval,
`information
`access,
`information
`systems, CMC,
`computer-mediated
`communications,
`help,
`collaborative
`help, CSCW
`
`INTRODUCTION
`have a problem with delivering
`Many user communities
`help and general assistance. Unfortunately,
`the user is often
`left
`to sift
`through reams of documentation,
`find his way
`through mail archives, or pursue answers
`through trial and
`error.
`Normally,
`one
`attempts
`examine
`the
`to
`or other help sources,
`and then wanders out
`documentation
`into a hallway in search of friendly colleagues.
`
`in distributed
`however,
`acute,
`The problem becomes
`the astrophysics
`communities. We take for our example
`community,
`although this problem exists in most scientific
`communities.
`In the astrophysics
`community,
`the users
`may be spread
`across
`the world,
`they may work
`in
`isolation,
`and they may have need of relatively
`specialized
`help. What we would like is a surrogate
`for this hallway
`talk. Such a solution must avoid the broadcast
`uroblem of
`flooding everyone’s
`electronic mail basket wit~ thousands
`
`Permission to make digital/hard copies of all or part of this material for
`personal or classroom use is granted without fee provided that the copies
`are not made or distributed for profit or commercial advantage, the copy-
`right notice, the title of the publication and its date appear, and notice is
`given that copyright is by permission of the ACM, Inc. To copy otherwise,
`to republish, to post on servers or to redistribute to lists, requires specific
`permission and/or fee.
`Computer Supported Cooperative Work ’96, Cambridge MA USA
`@ 1996 ACM 0-89791-765-0/96/11
`..$3.50
`
`this work reports
`Instead,
`of questions.
`to the appropriate
`a question
`narrow-cast
`those others are experts or colleagues.
`
`on a system to
`others, whether
`
`solution to a
`a collaborative
`then, examines
`This research,
`common problem.
`Earlier work, a system called Answer
`Garden,
`allowed
`organizations
`to develop
`databases
`of
`commonly
`asked questions
`that grow “organically”
`as new
`questions
`arise and are answered.
`The subsequent Answer
`Garden 2, the focus of this paper, continues
`this work.
`It is
`a second-generation
`architecture
`for
`the
`same
`design
`problem,
`investigating
`some of the issues encountered
`in
`field studies of the original
`system.
`The new architecture
`provides
`a customizable
`and adaptable
`set of software
`components
`that
`allow a variety
`of organizational
`and
`informational
`configurations.
`Furthermore,
`it offers
`a
`generalized
`solution to the problem of finding help for any
`information
`system. We report here on the new architecture
`and its responses
`to the context and authoring issues.
`
`to the help and
`introduction
`The paper begins with a brief
`as a brief overview of
`the
`memory
`problems,
`as well
`original Answer Garden
`application
`and its field study
`results.
`Answer Garden
`2 is then introduced.
`After an
`explanation
`of
`its architecture,
`the paper
`analyzes
`two
`particular
`features of Ans wer Garden 2, These two features,
`collaborative
`help and collaborative
`refining,
`are explained
`at
`length.
`Collaborative
`help mechanisms
`provide
`the
`necessary context
`for information,
`and collaborative
`refining
`mechanisms
`provide
`support
`for authoring.
`The paper
`concludes with a survey
`of related CSCW systems
`and
`some conclusions
`about
`these design considerations.
`
`PROBLEM
`FRED’S
`at the Harvard-
`is an astrophysicist
`Fred (not his real name)
`He,
`like many
`Smithsonian Astrophysical
`Observatory.
`about software
`scientists,
`does not want
`to know anything
`He wants
`to do his scientific
`systems
`or his hardware.
`work, free of the multitude of computer problems
`that seem
`to get
`in the way,
`
`sat around in a common room,
`In the “old days,” everyone
`using their computer
`consoles with the mini-computer.
`If
`Fred had a question, he could ask one of the half-dozen
`to
`dozen colleagues
`and programmers
`sitting
`in the room,
`
`97
`
`Meta Platforms, Inc.
`Exhibit 1022
`Page 001
`
`
`
`so the community
`the answer,
`Everyone had to hear
`from the problems of each individual.
`
`learnt
`
`near his
`Now, Fred sits in his office with his workstation
`desk.
`It
`is quieter,
`but much more isolated.
`If he has a
`problem
`or
`a question,
`he
`can
`look
`through
`the
`documentation
`or send electronic mail for help.
`If he sends
`electronic
`mail,
`he may
`not get
`an answer
`from the
`programmers
`for some unknown period of time, or he may
`be given a response
`that makes him feel
`stupid for not
`knowing the answer. Often he resorts to wandering through
`the hallways,
`looking
`for people who might
`know the
`answer. He then tries various possibilities
`until he finds a
`solution or he gives up.
`
`or organization
`institution,
`Any community,
`often has a problem with answering
`questions
`manner. Yet, solving problems
`and completing
`often dependent
`on obtaining
`timely answers
`questions.
`
`of any size
`in a timely
`tasks are
`to specific
`
`Fred’s problem is the dual problem of help and of collective
`memory. We will use the term collective memory to
`denote
`the
`common
`attributes
`of
`organizational,
`institutional,
`and community memory.
`(The term has a
`but
`related,
`slightly
`different
`meaning
`in
`the
`historiographical
`and critical
`literatures,
`but
`there
`is no
`better term to denote memory in a range of collectivities.)
`
`individuals’
`community,
`or
`an organization
`Within
`finding the right part of the
`information
`seeking requires
`collective memories
`include
`collective memory. Typically,
`information
`databases,
`filing
`information
`repositories
`(e.g.,
`cabinets, documents).
`It can also include people (e.g., other
`organizational
`personnel)
`[25]. The collective memory to
`which Fred has access includes
`at least
`the documentation,
`the system programmers,
`and his colleagues. However, he
`may have great
`trouble
`finding
`the right piece
`of
`the
`collective memory that has the answer he needs.
`In other
`words,
`his access
`to the collective memory
`should
`be
`augmented.
`
`problem
`and Fred’s
`Answer Garden
`one way
`Previous work,
`reported in [4] and [2], considered
`around a
`of doing this augmentation.
`This work revolved
`of its use
`system called Answer Garden.
`Field studies
`uncovered
`a number
`of important
`problems
`in providing
`collective memory and help to users such as Fred. Before
`discussing
`these
`problems,
`and
`our
`subsequent
`investigations,
`it will be useful
`to briefly describe Answer
`Garden.
`This application
`still plays an important
`part
`in
`our current work.
`
`in two
`memory
`organizational
`supports
`Answer Garden
`and by
`retrievable
`knowledge
`ways:
`by making recorded
`In the
`accessible.
`making
`individuals
`with knowledge
`seek
`standard
`configuration
`of Answer Garden,
`users
`answers
`to commonly
`asked questions
`through
`a set of
`diagnostic
`questions
`or
`other
`information
`retrieval
`mechanisms.
`Figures
`1 and 2 show Answer Garden
`reimplemented
`in the World Wide Web.
`(Other,
`third-party
`
`and in Lotus Notes.)
`[22]
`in the Web
`exist
`versions
`the user
`through Web pages.
`questions
`guide
`Diagnostic
`the
`user may
`use
`a number
`of other
`Alternatively,
`retrieval mechanisms
`to find the pages
`that
`information
`may contain the answer.
`
`is
`answer
`the
`or
`an answer
`find
`cannot
`the user
`If
`the
`through
`the user may ask the question
`incomplete,
`system.
`(This
`is the result of the user pressing
`the “I’m
`Unhappy”
`link in Figure
`1.)
`In the original Answer
`Garden,
`the system would then route
`the question
`to an
`appropriate
`human
`expert.
`(This has been changed
`in
`Answer Garden 2 as will be discussed below.)
`
`then
`the expert would
`In the original Answer Garden,
`answer
`the user
`through
`electronic mail.
`If the question
`was a common one,
`the expert could insert
`the question and
`its answer back into the information
`database.
`Thus, users
`were not
`limited to the information
`in the system;
`if the
`information
`was
`not
`present,
`they
`could
`tap
`the
`organization’s
`experts. As a result,
`the organization would
`gain a corpus of information,
`an organizational memory.
`Users
`could
`obtain
`expert
`advice
`without
`a high
`organizational
`cost. Other
`interesting
`properties
`of
`the
`system are discussed in [2].
`
`issues
`research
`Open
`Field studies of Answer Garden’s use ([2], [3]) uncovered a
`number
`of issues. While
`the system was held to have
`worked,
`two issues were uncovered
`that are critical
`to the
`success of similar memory or help systems:
`
`Q
`
`into the system in a more
`Tying the social network
`natural manner. Answer Garden’s dichotomy
`between
`experts
`and users was problematic. While there was
`nothing
`in the underlying
`technology
`to force
`this
`dichotomy,
`it was a simplifying
`assumption
`in the
`field study to have separate
`user and expert groups.
`Real collectivities
`do not
`function
`this way. Most
`people range in their expertise
`among many different
`skills
`and fields of knowledge.
`Fred knows
`things
`about
`systems
`and his tasks, even though he may not
`be able to answer specific questions. We would like to
`allow everyone
`to contribute
`as they can, promoting
`both individual and collective
`learning.
`
`to
`person
`to allow each
`mechanisms
`However,
`contribute must not overwhelm the other people who
`use
`the
`system.
`For
`example,
`broadcasting
`each
`question
`to every
`person
`in an organization
`or
`community will fail. AG2 offers
`several mechanisms
`to ameliorate
`the overload problem while allowing and
`providing for a range of expertise.
`
`Ct
`
`thus
`of answers,
`the contextualization
`for
`Providing
`In
`of an answer.
`providing for the user’s understanding
`the Answer Garden field study, most users either did
`not need contextualized
`information
`or were able to
`contextualized
`it themselves.
`However,
`a significant
`portion of the participants
`did need more context.
`
`98
`
`Meta Platforms, Inc.
`Exhibit 1022
`Page 002
`
`
`
`W’ Do you have a problem with remote access?
`
`How can I make 10CSJpages public or private?
`
`1. Howdo I make someof my pages public?
`
`to to avoid problems with
`There are two answers to this. We have a firewall
`hackers and crackers, You can make any of your pages available on the internal
`server by putting the file in your public_html directory snd setting the
`permissions correctly (see below). If you want a page to be added on the external
`web server (e.g. to allow someone to grab your data) you need to submit
`the page
`to C. Stoll (x5-7135) and he will take it from there.
`
`2, Howdo 1 set permissions on my files?
`
`Assuming you want to make the pages accessible to others through the internal
`web server, First copy the pages into your public_html directory in your account
`using the ‘cp’ or’ mv’ command. Then use the ‘chmod’ command to chsnge the
`permissiorx. For example:
`
`david@saturn:
`david@saturn:
`david@saturn:
`
`cp nev~age.
`cd -{publ
`chwd
`a+r
`
`htrnl
`ic_html
`* . html
`
`-/publ
`
`ic_html
`
`Figure
`
`2: An Answer Garden answer page
`
`99
`
`Meta Platforms, Inc.
`Exhibit 1022
`Page 003
`
`
`
`to a question may be present
`the answer
`In Fred’s case,
`However,
`he may lack the
`in the documentation.
`required expertise
`to infer an answer or to even use an
`without
`additional
`situational
`explicit
`answer
`information.
`
`unfortunately,
`is,
`context
`proper
`the
`Providing
`of
`We will
`return
`below to one way
`difficult.
`providing
`this context
`at
`low cost. Our
`potentially
`mechanism also ameliorates
`the problem of providing
`answers at the right
`level and length of explanation.
`
`the
`To obtain answers,
`burden.
`Cl Easing the authoring
`cost of authoring must be minimized.
`Furthermore,
`authoring
`answers,
`as
`an
`individual
`activity,
`promulgates
`the distinction
`between
`experts
`and
`else.
`everyone
`The composing
`content
`of answers
`takes as long as any writing takes, but we may be able
`to ease the mechanics of the process.
`
`increasingly
`to become
`issues
`these
`expect
`One might
`or
`becomes
`non-technical
`as the information
`problematic
`For
`the users become
`less
`sophisticated
`in the domain.
`example, only astrophysicists
`can understand
`the scientific
`analysis
`tasks
`that create
`their questions
`about
`software
`systems.
`Astrophysicists
`will vary in their
`computer
`expertise,
`but few wish to spend time inferring the answer
`from substantial
`system documentation
`before continuing
`with their analysis
`tasks. And,
`the programmers who must
`currently compose the answers may not even understand the
`domain or its tasks.
`
`of
`a number
`uncovered
`studies
`the field
`Additionally,
`technical
`issues, such as the need to use varying “front-end”
`systems
`such as the Web or Notes,
`to consider
`additional
`methods
`of
`finding
`experts,
`and to find better ways of
`maintaining
`the information
`database.
`These
`technical
`issues and the above social
`issues
`led us to reconsider
`the
`architectural design.
`
`2 (AG2)
`GARDEN
`ANSWER
`of a second generation
`consists
`Answer Garden 2 (AG2)
`organizational
`memory
`and
`system architecture
`for
`collaborative
`help support. There are several advantages
`to
`this architecture.
`
`First,
`
`the design cleanly separates
`
`the front-end
`
`of Answer
`
`needs. More
`from back-end
`the user client)
`Garden (i.e.,
`it also
`decomposes
`the Answer Garden
`importantly,
`into a set of distributed
`software
`services.
`functionality
`a high level of organizational
`flexibility;
`the
`This provides
`services
`can be mixed and matched
`in order
`to provide
`additional
`flexibility.
`For
`example,
`by attaching
`an
`users
`anonymity
`service,
`of
`the system can send their
`By attaching
`an anonymity
`questions
`anonymously.
`service at another point
`in the distributed
`architecture,
`the
`experts
`answering
`the questions
`can also be anonymous.
`Or by not having an anonymity
`service at all, all users and
`experts can be known to one another.
`
`the change in architecture makes much of the help
`Finally,
`possible from any information system. This
`functionality
`work,
`then,
`is generalizable
`to any information
`system.
`
`components
`System
`AG2 is built upon two underlying
`systems, both of which
`provide
`a set of
`services.
`These
`services
`create
`the
`collaborative
`help and collective memory
`functionality.
`The two underlying systems are:
`
`l
`
`.
`
`CafeCK is a
`(CafeCK).
`The Cafe ConstructionKit
`sociality and information
`CSCW toolkit
`for supporting
`use
`in collaborative
`environments
`[6].
`CafeCK
`provides
`a set of reusable objects
`that
`include message
`transport
`for
`asynchronous
`and
`synchronous
`communication
`(including
`a Zephyr-like
`system,
`NetNews,
`and email),
`parsing
`for a variety of semi-
`structured
`protocols,
`private
`and public
`channels
`for
`narrowcast
`communication,
`message
`filters,
`and
`message
`retrieval
`by a variety
`of
`semi-structured
`methods.
`By selecting
`from the
`set of available
`a
`components
`(or by extending
`it) and by writing
`simple Tcl program,
`an application writer can create a
`set of distributed
`processes
`to handle
`information
`information
`retrieval,
`access,
`or
`electronic
`communications.
`CafeCK is implemented
`in C++,
`Tel, and Tk.
`
`Co-Refinery
`(Co-Refinery).
`Refinery
`Collaborative
`provides mechanisms
`for handling individual
`and joint
`information
`spaces.
`Central
`to Co-Refinery
`is the
`ability
`to individually
`and collaboratively
`view and
`manipulate
`Answer Gardens
`and other
`information
`
`. .. ..................................................................................
`Web client
`!
`and pages
`
`CafeCK collaborative
`help “back-end”
`
`\
`
`client WIjy
`
`“raw” information
`input
`(partially
`CafeCK)
`
`-i
`
`—;~
`
`\
`
`w(-~e-
`
`?
`
`information
`database
`
`‘front-end”
`
`Collaborative Refinery
`
`Figure 3: Answer Garden 2 (AG2) architecture
`
`100
`
`Meta Platforms, Inc.
`Exhibit 1022
`Page 004
`
`
`
`.e$k+o-+$?5!users
`
`/0
`
`help, we believe we have found mechanisms
`collaborative
`for reducing the context problem.
`
`client
`
`escalation
`agent
`
`to get an answer goes to a
`(a) The user’s first attempt
`chat channel.
`
`Web AG2
`client
`
`escalation
`agent
`
`~-+1””-J
`QA
`tracker
`
`help
`desk
`
`to get an answer gets
`(b) The user’s jth attempt
`escalated to a help desk.
`
`Figure4:
`
`Twopossible
`
`escalations
`
`foraquestion
`
`in situations where
`useful
`It is especially
`collections.
`one wants to refine and distill collections
`of materials
`as shared artifacts.
`It will be described
`extensively
`below.
`
`for managing
`include objects
`components
`Co-Refinery
`archive
`of materials,
`constructing
`and
`a collection
`a database
`of
`relationships
`for
`those
`maintaining
`and generating
`a suitable
`presentation,
`materials,
`Output
`from Co-Refinery’s
`presentation
`generator
`can
`be HTML, Notes documents,
`files, or e-mail.
`Co-
`Refinery is implemented
`in C++, and the Web portion
`relies
`upon
`HTML
`3.0
`and Netscape
`HTML
`extensions.
`
`as in Figure 3.
`are used together
`These two components
`into the collection archive through
`Raw information
`comes
`CafeCK processes
`(such
`as News
`filters),
`by being
`explicitly
`sent
`to the archive
`through
`e-mail,
`or through
`filtering agents.
`It may be partially processed,
`and then is
`moved into the information
`database.
`At snap-shots
`or
`upon explicit
`queries
`(depending
`on a site’s
`tailoring
`of
`AG2),
`the materials
`are built
`into Web
`pages, Notes
`documents,
`or flat
`files.
`In turn,
`the AG2 Web or Notes
`clients
`can send mail
`to CafeCK back-end
`processes
`that
`then handle
`the details of obtaining
`help. These CafeCK
`help processes will be described next.
`
`HELP
`COLLABORATIVE
`The problem as a duality
`AG2’s
`as a collective
`either
`“back end” can be viewed
`help system.
`(We use
`memory system or as a collaborative
`collaborative help to denote
`those help systems
`that use
`people
`as
`information
`sources,
`for
`example,
`through
`Computer-Mediated
`Communication
`systems.)
`Each of
`these views
`is the dual of the other. By duals, we invoke
`the language of linear programming, where two forms exist
`for each particular problem. Both forms are valid, and users
`are free to solve the form that provides
`them with the most
`analytical
`tractability.
`considering
`the
`“back-end”
`By
`organizational
`memory
`problem in terms
`of
`its dual,
`
`101
`
`it was noted that an open research issue was how to
`Above,
`the users’ need for contextualized
`information
`in
`alleviate
`solving their problems
`and finishing their tasks. This issue
`can be ameliorated
`by using
`collaborative
`help
`in a
`controlled manner.
`Collaborative
`help functionality
`also
`provides
`help to users at
`their own explanation
`level and
`potentially with iterative diagnosis.
`
`local
`Staying
`Providing
`on
`-- such as colleagues
`help from other people
`the same hall or other group members
`-- allows people to
`seek help first
`from the people most
`likely to know the
`local context.
`Colleagues
`can judge
`a person’s
`abilities,
`expertise,
`and situation,
`and can try to provide
`suitable
`information
`to solve
`the
`person’s
`problem.
`Local
`participants
`are also more
`likely
`to information,
`since
`personal
`social
`ties
`are key motivators
`in providing
`assistance
`[7, 19].
`
`problematic.
`is, however,
`asking one’s colleagues
`Always
`other
`people.
`AG2’s
`is
`still
`costly
`to ask
`it
`First,
`of previously-asked
`questions
`and frequently-
`repository
`information,
`however,
`attempts
`to reduce
`that
`required
`problem. More
`important
`y, one’s
`colleagues may not
`know the answer. While staying local
`is important,
`it can
`also be organizationally
`dysfunctional
`[ 10] when there is no
`In these
`situations,
`a means
`for
`local expert
`available.
`escalating answers past the local group is required.
`
`Escalation
`of CafeCK, we were able to simply
`Using the facilities
`construct
`an escalation
`agent
`for questions
`in AG2. This
`component
`allows
`the user
`to decide what
`to do if the
`It allows
`the user
`to consider
`question
`is not answered.
`whether
`to get answers
`from chat systems, bulletin boards,
`software agents, or other people.
`
`The typical way that we envision the system being used is
`to gracefully
`escalate
`the help request
`until
`it can be
`answered.
`Because
`the escalation
`agent
`is a CafeCK
`process,
`the escalation
`can be quite flexible.
`The agent
`is
`currently programmed
`to follow organizational
`rules on the
`order of escalation,
`although
`this is under user control.
`It
`would be a simple matter
`to change this to provide different
`organizational
`rules,
`complete
`user
`control,
`or even
`heuristics
`(such as avoiding the chat facility when no other
`users
`are logged
`into their machines).
`No doubt other
`mechanisms
`could be found;
`this
`is a potential
`research
`question.
`
`his
`through
`a question
`the user poses
`In our prototype,
`is an
`application.
`In the example
`of Fred,
`the user client
`AG2 front-end,
`but
`it can be any application
`that has
`asynchronous
`or synchronous
`communication
`capabilities.
`The
`application merely
`connects
`to a CafeCK process
`through,
`for example,
`e-mail.
`This CafeCK process,
`the
`escalation
`agent,
`is
`semi-autonomous,
`since
`it can be
`triggered either by the user or automatically.
`
`Meta Platforms, Inc.
`Exhibit 1022
`Page 005
`
`
`
`As an example,
`(Note that
`imagine the following scenario.
`components
`for AG2 provide
`an enormous
`the underlying
`flexibility,
`so users’ actual practices
`can vary widely from
`this.) The user, Fred again, has a question
`about his data
`analysis
`package,
`and he would
`like
`to know how to
`correctly massage
`his data.
`He first
`looks
`through
`the
`existing questions
`and answers, either in a stand-alone AG2
`information
`database
`or in an AG2 component
`of his data
`analysis
`application.
`Assuming
`that
`the answer
`is not
`there, or
`that he does not understand
`how to apply the
`information
`that
`is there, he composes
`a question and mails
`it off through his Web browser.
`
`Instead of the question going to an expert, as it would have
`in the original Answer Garden,
`the question
`goes
`to his
`escalation
`agent. This is, of course,
`invisible
`to Fred. The
`question is first sent
`to a synchronous
`chat system (Figure
`4a). We
`envision
`the chat
`system being
`set up with
`channels
`or subchannels
`for each work group, hallway, or
`other social grouping.
`If someone on the chat system can
`answer Fred’s question,
`and is inclined to do so, Fred gets
`his answer
`immediately.
`As mentioned
`above,
`nearby
`colleagues
`(as measured
`by geographical,
`social,
`or
`intellectual
`distance)
`are most
`likely to answer his question
`with the correct and sufficient context.
`
`the system pops up a
`after 5 minutes,
`In our prototype,
`window on the screen.
`In our scenario,
`the dialog box asks
`Fred whether he got an answer
`to his question,
`and if not,
`whether he would like to continue
`(Figure 5).
`If Fred says
`to continue,
`the system routes
`the question
`to a NetNews
`bulletin board.
`(It is also conceivable
`that
`it would route it
`to a chat channel with a wider distribution,
`but the point
`is
`the same.) After another period of time, perhaps 24 hours,
`the agent again pops up a window on Fred’s
`screen, asking
`whether he has received an answer. The process continues,
`perhaps
`routing the question to an expertise engine to find a
`suitable
`human expert,
`to a help desk (Figure 4b), or to
`agents
`that
`search
`for
`information
`on the Web
`or
`in
`proprietary
`information
`sources
`(such as Dialog or Nexus).
`One can even imagine agents that hire outside consultants
`if
`the need is great enough.
`
`is more assured of
`In this manner, Fred or any other user
`receiving
`a usable
`answer.
`Staying local
`lowers
`the cost,
`since
`organizational-level
`experts
`need
`not
`be used
`immediately;
`increases
`the chance
`of getting
`an answer,
`since colleagues may be more motivated
`to answer; and is
`more likely to provide
`context,
`since colleagues
`know the
`local situation.
`To be sure,
`this approach is not a panacea.
`While it does help provide the proper amount of contextual
`-..
`.............
`
`[~ Escda~on agerit’’’###
`....... ...-—.
`.
`. ...
`
`...
`
`I At 10:37am on 3/19AJ6 (today) ymuaskd
`
`....... ..—.4
`
`121
`
`I
`
`>Xubjecti Can I add a Web page to someone else’s site?
`
`~
`
`Figure 5: The escalation agent
`
`102
`
`and while
`the answer meaningful,
`to make
`information
`there is a greater
`likelihood
`that
`the answer will be at the
`right explanation
`level and length,
`there are difficult
`issues
`surrounding
`the social organization
`of channel
`groupings
`and the like. Colleagues’
`time is hardly free.
`
`thinking global) does allow
`(but
`staying local
`Nonetheless,
`group members
`to help one another, while preserving
`the
`capability
`to ask
`larger
`groups
`as well
`as
`experts.
`Furthermore,
`the dichotomy
`between
`experts
`and users is
`largely broken down.
`
`services
`help
`collaborative
`Other
`AG2 requires
`In
`of other CafeCK services.
`a number
`(chat,
`communication
`services
`addition
`to the
`basic
`NetNews,
`e-mail) and to the escalation agent, AG2 requires
`services
`to find experts,
`to provide basic statistical
`services,
`to make users anonymous,
`and to track users’ questions.
`The capability to find a suitable expert
`is required, and AG2
`currently uses a rudimentary
`rule-based finding mechanism.
`This
`is clearly
`a bottle-neck
`for
`real
`use,
`but other
`researchers
`are developing
`better mechanisms
`for handling
`this problem (e.g.,
`[17]).
`The anonymity
`service
`allows
`users
`to ask questions
`anonymously.
`Organizations
`or
`communities might not want
`this service,
`in which case the
`The
`statistics
`service
`notes
`service
`is merely
`omitted.
`which communication mechanisms
`are used, and also tracks
`the use of pre-existing
`answers.
`In a production
`system,
`users’ questions
`should be tracked; otherwise, questions can
`slip away.
`
`to be
`components
`these
`of CafeCK allows
`The design
`mixed and matched in a building block manner. Different
`organizational
`arrangements
`can be created
`through
`the
`architecture of the software components.
`Furthermore,
`each
`service can be tailored through its internal Tcl programs.
`
`and anonymity
`engine
`expertise
`only a simple
`Currently,
`service have been implemented.
`The others are planned.
`
`viewing the problem as one of collaborative
`In summary,
`one
`to
`remove
`the
`requirement
`that
`help
`allows
`memory merely
`be a set of
`information
`organizational
`Staying
`local, with
`the
`possibility
`of
`repositories.
`provides
`for a range of help from the people
`escalation,
`However,
`AG2
`also
`information
`seeker.
`around
`the
`includes
`stronger
`support
`for
`building
`information
`repositories
`as well. This support will be described next.
`
`REPOSITORY
`A MEMORY
`REFINING
`tons
`of
`unwinnowed
`shelves
`were
`On my
`shape it was of little use
`....
`In the present
`material
`Facts were too scattered;
`to me or to the world.
`indeed, mingled
`and hidden as they were in huge
`masses
`of debris,
`the more one had of them the
`worse.
`...To
`find
`a way
`to the gold
`of
`this
`amalgam... was the first
`thing to be done.
`(Bancroft
`[8], 1891, p. 135)
`
`is iteratively
`application
`to the Answer Garden
`Central
`answers
`and
`building a repository
`of commonly
`requested
`other
`information.
`If
`this
`is to be accomplished,
`low
`
`Meta Platforms, Inc.
`Exhibit 1022
`Page 006
`
`
`
`overhead
`members.
`
`is required
`
`for organizational
`
`or community
`
`The original Answer Garden design assumed that building
`such
`a memory
`repository
`would
`occur
`through
`the
`everyday
`interaction
`of users asking questions
`and experts
`providing
`answers.
`However,
`authoring
`was
`still
`a
`significant
`task.
`The effort of writing
`explanations
`and
`formulating
`answers
`cannot be minimized,
`Nonetheless,
`Answer Garden
`provides
`mechanisms
`for
`iteratively
`developing
`an
`answer.
`AG2
`provides
`additional
`mechanisms
`for refining answers
`from very raw information
`sources
`as well as removing
`unnecessary
`context. We
`developed Co-Refinery
`to provide
`these mechanisms.
`The
`goal
`is to enable groups of people to collaborate
`in jointly
`or
`individually
`building
`answers
`and
`information
`repositories
`over time.
`
`refining
`Collaborative
`AG2,
`supports
`system,
`through the underlying Co-Refinery
`activities:
`an authoring
`process
`that
`includes
`four general
`collecting,
`culling, organizing,
`and distilling. We assume
`that any of these activities,
`as well as authoring, may be
`done iteratively
`or in any order.
`Each activity
`is clearly
`important,
`although the major
`research contribution
`here is
`the support
`for collaborative
`distilling:
`
`is the phase
`Collecting
`is
`information
`in which
`collecting
`can be
`automatic
`In Co-Refinery,
`gathered.
`set up for certain
`types of
`information
`streams
`that
`occur
`in AG2,
`such as NetNews,
`synchronous
`chat
`channels,
`or distribution
`lists.
`In addition, manual
`collecting
`allows
`individual
`items
`to be submitted
`through the system directly
`or by e-mail. Collecting
`places items into the archive.
`
`the
`cull
`one must
`the material,
`collecting
`After
`interesting
`material,
`and the
`lesser
`collection
`for
`Culling is a
`material must be discarded
`or ignored.
`selection mechanism,
`identifying themes or threads that
`occur within
`a collection.
`A sizable
`reduction
`of
`material
`may
`be
`possible
`through
`culling
`the
`collection, making subsequent
`organizing and distilling
`easier. Culling reduces
`the apparent
`size of the archive,
`although
`in our current
`implementation,
`items
`are
`unreferenced rather than deleted.
`
`Organizing allows one to group materials
`according to
`some
`classification
`schemes
`so to enhance
`their
`Our
`current
`retrievability
`and
`understandability.
`prototype
`relies
`heavily
`on outlining,
`user-defined
`and
`keyword
`indexing,
`indexing,
`but
`other
`classification mechanisms
`are clearly possible.
`In Co-
`Refinery,
`retrievability
`is enhanced
`by making
`the
`culled
`subset
`a fully
`identifiable
`element
`in the
`In this way,
`organizing
`results
`in an
`collection.
`addition to the collection.
`
`is distilling --
`part of refining
`important
`The most
`boiling
`down the existing
`(and culled) materials
`in
`order
`to uncover
`the answers or knowledge.
`As with
`chemical
`or liquor distilling,
`the results
`should be a
`
`the original
`form of
`or concise
`concentrated
`more
`information.
`Creating
`or editing
`a summary
`or
`synopsis,
`for example,
`removes much of the tedious
`work
`of wading
`through
`extraneous
`or erroneous
`information.
`
`is a distillate.
`of distilling
`The result
`is
`Support
`for authoring a
`provided for generating the raw material
`distillate,
`but
`it assumed
`that only users
`can fully
`distill and refine the material.
`Co-Refinery
`currently
`supports a number of distillates.
`For example, a useful
`intermediate
`distillate consists of merely concatenating
`selected
`NetNews
`messages.
`This
`allows
`an
`author/editor
`to further prune the selected information
`into one final distillate
`consisting
`of an authoritative
`answer.
`This behavior
`is very similar
`to what people
`currently
`do when they compile
`a FAQ.
`Another
`useful distillate
`is temporally
`based;
`items in it vanish
`after a short period of time. Technical
`hotlines often
`have runs of questions,
`and this distillate
`solves
`the
`problem of communicating
`the temporarily
`needed
`answers.
`
`that users move fluidly among
`As mentioned, we assume
`these activities. We also assume
`that
`the refining is done
`iteratively
`and incrementally.
`In doing
`so,
`the system
`allows
`groups
`of users
`to interact
`in creating
`shared
`information
`artifacts
`and a common
`information
`space.
`This
`system represents
`an alternative
`to many current
`attempts
`to completely
`automate
`the refining process.
`
`the results of refining in AG2, and can be
`Figure 6 shows
`considered
`as a snapshot of an iterative process.
`In Figure
`6, the leaves are raw information,
`perhaps NetNews
`or e-
`mail messages. Authors and editors have created distillates
`for five of the threads, winnowing
`the material
`into answers
`for frequently-asked
`questions.
`One of these answers was
`shown
`in Figure
`2 above.
`After
`being finished,
`some
`distillates
`can then take the place of the raw material
`in the
`archive,
`Note
`that
`some
`of
`these
`distillates
`could
`be
`intermediate;
`i.e., not shown to the public because they are
`Additionally,
`all distillates
`can be
`under
`construction.
`iteratively revised.
`
`of distilling
`economics
`The
`There
`to authoring.
`Refining
`is not a complete
`solution
`will always be effort
`required
`to compose
`explanations
`of
`complex
`technologies
`and tasks.
`Refining
`reduces
`the
`overhead
`for
`that
`task,
`and
`simultaneously
`reduces
`information overload.
`
`It does
`We do not believe that all materials will be refined.
`through
`not make
`sense
`in all
`cases
`to move
`data
`information
`to organizational
`knowledge.
`For example,
`information
`that has a short-shelf
`life will not be refined,
`especially
`if the information
`also has a high throughput
`velocity.
`The cost would be prohibitive.
`Therefore, we
`have attempted
`to define some distillates
`that are suitable
`for temporally limited information.
`These distillates do not
`boil down
`the material,
`but
`they do make
`findin