Context Framework - an Open Approach to Enhance Organisational Memory Systems with Context Modelling Techniques Roland Klemke GMD - German National Research Center for Information Technology GmbH Institute for Applied Information Technology Schloß Birlinghoven, D-53754 Sankt Augustin Roland.Klemke@gmd.de expertise) [Rac 1997]. Some philosophers even deny the existence of a context-independent meaning of concepts Abstract [Hei 1962]. Even though context is recognised as being very Researchers from various fields (among which important, research concerning context (especially in the are Artificial Intelligence, Computer Supported area of organisational memories) is in its very early Co-operative Work, Information Retrieval, stages. It is not yet agreed in the scientific community Software Engineering, and Knowledge what context is and which elements of context are Engineering) try to address issues in important within organisational settings. It is still an open Organisational Memories (OM). Many of these field how to represent contextual information and how to approaches also identify the meaning of context use contextual information for reasoning purposes. for information captured inside an OM. Up to This paper gives an overview on the state of the art in now context has not been modelled explicitly as organisational memory research with a special focus on part of an OM. This paper gives an overview of the concept of context and develops a framework on how different approaches in OM research and to recognise, represent and use contextual information proposes research for recognising, modelling, and within an organisational memory application. retrieving contextual information as part of the The main question behind the presented approach is OM. The main question behind the presented whether knowledge about the creation or usage context of approach is whether knowledge about the any piece of information within the organisational creation or usage context of any piece of memory and knowledge about the current context of any information within the organisational memory organisational member may be used to effectively enhance and knowledge about the current context of any the individual's access to organisational information. organisational member may be used to effectively Gathered contextual knowledge may be used to enhance the individual's access to organisational automatically offer information related to the current information. context (by identifying similar contexts and information created within). Retrieval techniques may be enhanced by 1 Introduction extending queries with explicit information on past contexts. Context has been recognised by a wide range of researchers as being an important concept to consider 2 State of the Art in Organisational when looking at the meaning of information. Memory & Context Research Psychologists perform memory tests to analyse the effect of context for the remembrance of words [Sri 1997], Generally, an organisational memory (OM) comprises the Researchers in the machine learning area investigate the complete knowledge of an organisation collected over the effects of context on the automatic learning of concepts time of its existence. It consists of personal memories of and deliver promising results [MK 1996], Organisational people working in the organisation (i. e. their knowledge, Research people use communication models to investigate experiences, expertise), document archives (both the role of context in information product evaluation [Mur electronic and paper-based), and all further relevant 1996], and cognitive scientists stress the importance of pieces of knowledge that are important for organisational context for human expertise (and consequently machine success. In this paper we will use the term organisational memory in a more restricted form: OM is seen The copyright of this paper belongs to the paper’s authors. Per mission to copy synonymously to computerised organisational memory without fee all or part of this material is granted provided that the copies are not applications. The goal of such applications is to capture made or distributed for direct commercial advantage. knowledge or information within an organisation and Proc. of the Third Int. Conf. on Practical Aspects of distribute it to the workers who need it. It is the overall Knowledge Management (PAKM2000) goal of organisational memory systems to improve the Basel, Switzerland, 30-31 Oct. 2000, (U. Reimer, ed.) competitiveness of an organisation by improving the way in which it manages its knowledge [HSK 1996]. http://sunsite.informatik.rwth-aachen.de/Publications/CEUR-WS/Vol-34/ In the following sections we will review several R. Klemke 14-1 approaches to OM stemming from very different research information describing the message content, thus allowing directions and following different goals. We review these contextual organisation of messages. approaches with respect to their notion of context. Table 1 A second approach (Project Compendium, also summarises these reviews defining the underlying reported in [ZS 1997]) is aimed to capture formal contextual features and models. documents (e.g. project reports). Project Compendium creates/maintains a hypertext space of documents of 2.1 Help Systems different formats/origins allowing to associate documents with each other. This hypertext space serves as a basis for One of the first published OM systems was Answer ”Conversational Modelling”, a technique that is motivated Garden (Ackermann, [Ack 1994a], [Ack 1994b], [AM by the observation that modelling tasks performed by 1990], [AM 1996]) which aimed to provide a teams consist to a great extent of discussion, argument, continuously growing repository of hierarchically brainstorming, etc. From a different perspective these structured questions and answers including hyperlinks may also be seen as document context, as all communication means to route unanswered questions to links pointing to a single document can be seen as the domain experts. Goals of this approach were to make context in which this document is embedded. recorded knowledge retrievable and to make people with An approach that introduces explicitly contextualised knowledge accessible . Later versions of Answer Garden information to the CSCW domain is described in [Pri were expanded with regard to the use of different means 1993] (TOSCA). Here, an organisational information to get questions answered: browse through previously server that models organisational entities (people, answered questions, chat, news groups, help desk, etc. projects, departments, tasks, ...) as interrelated objects is The external communication means were used only to put described. Different views may be generated on questions there, they were not used to retrieve or archive organisational information and relations between previously answered similar questions. organisational objects may be followed. Presented Ackermann identifies contextual problems within OM information is always contextualised (creation context, by showing a trade-off between too much (non e.g. department or project) and communication means generalisable) and too little (non understandable) context allow contextualised discussion about & annotation of information. His idea is to ”strip away” contextual objects. As weak point of this approach one may see that information from documents stored within the OM to explicit organisational modelling is required a priori identify the general (= reusable) part of it and to provide which may lead to outdated organisational structures and explicit contextual information in a simple form (such as context models. Another problem is that the submission date and author). contextualised information can only be found by users The use of one (dynamically growing) categorisation who move to the appropriate context. Thus only retrieval hierarchy (i.e. a question hierarchy) classifying questions by ”matching context” (as opposed to the explicit retrieval and answers makes retrieval using Answer Garden of documents in a certain context) is supported. difficult as it does not allow different views on the categorised information. Every user, regardless of his/her 2.3 Artificial Intelligence, Knowledge Engineering, context, expertise, interest, etc. viewed the same answers Knowledge Management to the same questions using the same hierarchy (that, needless to say, grows a bit unmanageable in time). The An approach to an OM for knowledge workers that tries question & answer based approach makes Answer Garden to capture the history of decision processes is presented in a tool to be used in helpdesk applications rather than in [Buc 1997]. The author characterises knowledge work general OM applications. using a definition of tame and wicked problems and offers an approach for argumentation visualisation. The history 2.2 Computer Supported Co-operative Work leading to a decision provides the context in which this decision is made. A drawback of this approach is that the Some approaches to OM have been reported from CSCW visualisation of even a simple decision may look quite research. OM in these areas often is called group memory complex. This problem increases with complex decisions, underlining the informal character of supported user where many people are involved. It also requires groups. One approach that archives e-mail discussions (and consequently decisions) to be explicitly communications (Knowledge Depot) is reported in [KZR documented using the presented approach, which leads to 1997] and [ZS 1997]. This work identifies the concept of additional effort and cognitive load. ”Project Awareness” which comprises the awareness of Van Heijst, van der Spek, & Kruizinga [HSK 1996] discussions, decisions, and changes during project work. define corporate memories as an explicit, disembodied, Knowledge Depot organises the group memory into persistent representation of the knowledge and dynamically refinable hierarchical sections (just like information in an organisation that should support the Answer Garden) and classifies incoming e-mails based on basic knowledge processes (develop new knowledge, subject-line keywords. Users may now browse through the secure new and existing knowledge, distribute knowledge, archive or trigger selected sections to be automatically combine available knowledge). They aim to develop a informed about incoming mails or search the archive knowledge pump , i.e. a corporate memory that allows using keywords. If the user community agrees on a subject active collection and distribution of knowledge. They naming policy, these subject lines may contain contextual propose the use of knowledge profiles for every user as to R. Klemke 14-2 identify relevant knowledge objects within the memory. grappling with problems of OM. [KO 1997] offers a These profiles which can be seen as simple context matrix-based information retrieval approach to OM models are manually constructed and maintained by the document retrieval and resource allocation using users themselves (which may be seen as the weak point of relevance ranking and associative retrieval. The idea this approach: the maintenance effort may be eschewed by behind this approach is to identify all relevant concepts the users). (terms) in a domain and set up a concept vector. Every In [ABHKS 1998], [AAST 1998] & [BHS 1998] OM document is then represented by a relevance vector d with is seen as an enterprise-internal application-independent |d| = #concepts resulting in a matrix. A single entry in this information and assistant system that integrates various matrix denotes the occurrence of concept i in document j. techniques and tools to support knowledge management . Matrix operations allow the calculation of ”similarity The presented approach is based on enterprise-, domain-, measures” between documents and the ranking of and information-ontologies used to classify the archived documents with respect to certain concepts. An interesting information where the enterprise-ontology classifies aspect of this approach is that the ”document similarity contextual information, the domain-ontology classifies measures” allows the ranking of documents as relevant information content and the information-ontology even when they don't contain the queried terms. The classifies structure. The enterprise ontology may be used calculated matrix implicitly relates all concepts and thus to generate a context model for classified information that provides a context for each possible search term. The a describes the organisational context in which the priori definition of relevant concepts may be seen as information has been created. As context modelling is not shortcoming of this approach which may lead to the main focus of this research only organisational context maintenance problems in dynamic environments. is regarded here, which itself is reduced to a process A further approach from IR research focusing on the oriented context view. use of context is described in [Gök 1999]. The proposed [Schwa 1998] proposes the use of user centric meta research utilises Machine Learning techniques to learn knowledge in organisational memories by enhancing plain user context by observing subsequent queries. An existing text e-mails with links to appropriate concepts within the ”ContextLearner” component (of which no further details OM. In this approach the OM is considered to comprise are provided) shall be used in this project. A major two parts: a knowledge base containing organisational difference between context approaches in OM and IR is knowledge and meta-knowledge used to process the that IR systems (as the nature of things demands it) only knowledge. Meta-knowledge is considered to be user regard the user context at retrieval time, while OM offers centric and is used to identify relevant concept the possibility to enhance the contained information with descriptions in the form of user-profiles and shared context. Another issue is that an IR system cannot make semantics. User-profiles are used as more or less static any assumptions about the users work environment while user information (regarding e.g. position, current & past an OM will usually be embedded into an organisation's projects, ...) while shared semantics are concept work environment which may provide rich context descriptions that a user can ascribe to or not. All users information. who ascribe to the same description of a concept are believed to share the same view of that concept. In this 2.5 Organisational Learning approach users are required to actively ascribe to concepts An approach that covers personalisation issues in OM which have to be defined a priori. Thus it is questionable research is presented in a research proposal [FOS 1997]. whether in an environment of ever increasing amounts of The authors try to investigate three main OM issues: how concepts users are willed to keep their concept views up to capture knowledge; how to sustain timeliness & utility; to date. and how to deliver actively and adaptively . Their research In [MSPK 2000] a knowledge management approach is aims to support software development groups and is based described that uses ontology-based domain modelling on results of a previous project [Lin 1996] where techniques. The used ontology allows inheritance and complexity in design is analysed (concerning the synthesis instance-of relations to be modelled. It is built on concepts of different perspectives, the increasing amount of and instances and allows attribute-based, concept-based information relevant to a design task and the and text-based queries. Documents are manually enriched understanding of previous design decisions). A framework (contextualised) with concepts from the domain model. In for a group memory feedback loop is presented that tries the presented modelling approach only instance-of and to tackle two disparate goals: support for the current inheritance relations are supported. Especially design work at hand & support to record information for containment and general association relations are missing. future reuse. GIMMe (Group Interactive Memory Furthermore the underlying definition of context is not Manager), an e-mail-based tool to capture, store, organise, clearly stated. The enrichment of documents with domain share and retrieve conversations is presented. Similar to concepts is simply called contextualisation. [KZR 1997] GIMMe organises e-mails according to their A comprehensive survey of knowledge engineering subject lines. approaches may be found in [SBF 1998]. 2.6 Software Engineering 2.4 Information Retrieval, Text Filtering The Software Engineering community also offers The information retrieval community has also been R. Klemke 14-3 approaches to KM and OM problems. Maurer & Dellen context to the notion of ”current workflow task”-context [MD 1998] present an approach for process oriented and ”reflection on workflow”-context. knowledge management where information need and Another approach of integrating OM and WMS is knowledge provision are dependent on the process presented in [KS 2000]. The central idea is the explicit context. Their approach is related to the ”experience representation of mnemonic processes (i.e. processes to factory” approach [BCR 1994] that tries to package create, use and maintain knowledge) as business software development experience. Maurer & Dellen processes. The underlying hypothesis is that business present a process modelling approach that connects processes involving people and technology form that part documents to processes instead of using formal of the OM promising best utilisation of resources. classification & retrieval methods. While the connection Consequently, capturing and accessing OM should of documents with process states offers interesting concentrate on these processes. The presented work retrieval capabilities it is also the weak point of this adopts Takeuchi and Nonaka's modes of knowledge approach: only the exactly matching process context will conversion (socialisation, externalisation, combination, provide the right information, no explicit context model is and internalisation) and outlines the following process: maintained that might allow similarity measures and no identify core business processes; identify corresponding context-free retrieval (e.g. using keywords) is supported. people and agents; get descriptions for processes by The idea behind this approach is quite similar to the process members; use mnemonic process knowledge already discussed approach in [Pri 1993], where creation to externalise process, agent, and tool organisational structures are used instead of software representations; empower people in training sessions to engineering process models to identify context. use the system; and finally run the system to build knowledge. The modelling approach in this work is based 2.7 Workflow on the identification of business process models as primary objects and the identification of knowledge [Wol 1997] offers an approach to use explicit enterprise creator, knowledge user, expert, and knowledge models to circumvent the drawbacks of standard administrator as knowledge agents. Context is not information filtering methods in the distribution of explicitly mentioned here but as business process models corporate information. This quite interesting approach to can be seen as context models for business process precise information distribution based on enterprise execution it seems clear, that explicitly but manually models (thus providing some usage context) is limited to created context models are maintained by this approach. organisations with explicitly modelled, stable and reliable communication structures and responsibilities. In this 2.8 Virtual Enterprise approach information items get distributed within an organisation based on the organisational roles that people A virtual enterprise (VE) is an organisation comprising have and their relations to the organisational process that different people of different (physical) organisations to created the information. reach a dedicated goal in a limited period of time. As such Another work focused on highly structured application a VE is comparable to a project consortium. After the goal domains (here: insurance companies) is [Rei 1998], where is reached, a VE stops to exist. Approaches that try to the author tries to combine (integrate) several knowledge support VE and research communities with OM bases using knowledge formalisms. Their understanding technology can e found in [DCGR 1998], [RM 1998], & of OM is based on the perception of two roles: (1) OM [GS 1997]. acts as a passive container for relevant organisational Based on a corporate memory typology offered in knowledge; (2) OM acts as an active distributor for [DCGR 1998], [RM 1998] offers an analysis of the CM information needed in the task at hand. To reach the need of a VE exemplified for the domain of concurrent second role the author states, that the OM needs to know engineering (CE). Two levels of tasks in concurrent what the user is currently doing. He thus proposes the engineering are identified: individual design and co- integration of OM with a WMS which provides process operative evaluation. To support these tasks a corporate context. memory designed for a VE should be composed of a An approach that tries to put dynamics into the context profession memory (capturing knowledge about people, based information distribution of OM is presented in expertise, professions), a project definition memory [WWT 1998]. The authors try to integrate OM with an (capturing requirements & results), and a project design evolutionary workflow management system. The WMS rationale memory (keeping components, conflicts, stores completed processes in a case base providing problems, solving methods, arguments). access to best (and worst) practices and lessons learned Some open issues remain unanswered (and even (inner feedback loop: learning how to optimise process unidentified) by the authors: Why should one set up an execution) and providing the possibility to reflect on OM for a limited period VE when the effort of creating process models and modify them (outer feedback loop: and maintaining an OM only pays off in the long run? learning how to improve process models). During Which members of the VE own the OM? The members of execution of processes and tasks the WMS gives access to a VE may have the same strategic goal but do they share task specific documents and information items. An ”out of the same interest? Do they want their expertise to be context” information need, that exists outside a modelled shared with other VE members? process is not supported by this approach. Also, it limits An approach that is oriented towards the support of R. Klemke 14-4 research communities is presented in [GS 1997]. Though characteristics) in order to evaluate its relevance in a research communities are no virtual enterprises they share given situation. some commonalties: distributed over the whole world, working in closely related areas, interested in fast and 3 Context Modelling efficient knowledge exchange. [GS 1997] proposes In this section we will show some theoretical backgrounds knowledge management through capturing of live events of our context modelling work and motivate the (such as conferences) in hypermedia (WWW, CD-ROM, underlying goals. We will then use the results from the ...). Papers presented should be enriched by video state of the art review to derive context modelling captures of presentations. Electronic conference requirements. proceedings could then benefit from the technological advantages of linking text documents with picture, sound, 3.1 Background and Goals and video material. Knowledge management is concerned with data, 2.9 Our Previous Work - Information Brokering information and knowledge. Following Alavi and Leidner and Organisational Memories [AL 1999], we define data as raw unstructured symbols (such as text and numbers). Information is defined as In a previous project (COBRA - Common Open processed, conceptualised and categorised data. Brokerage Architecture, [KK 1999], [SMDP 1998]) we Knowledge is information that is made actionable by aimed to support the work of professional information being contextualised and personalised. The three levels brokers with specific information systems. We built an and their transitions are shown in fig. 1. organisational memory system (called bizzyB) supporting them in their daily work by integrating customer, case and profile management with automatic information retrieval no ati tua & from various heterogeneous information sources. lis ep n nc atio Co rso The information objects a broker has to deal with Pe nte nal Co plic (namely customer data, case data, profile data, and xtu isat Knowledge Ex ali ion dossiers built from retrieved information) were organised sa tio along their context of use. A broker working with a tio n n& isa tio customer's profile could then for instance easily access all n or lisa automatically delivered information for this profile. teg ua Information Ca cept During the evaluation phase of the COBRA project this n context-based information organisation proved to be a Co & useful concept. The underlying model of context used in this approach was a priori modelled and based on our Data analysis of the working situation of an information broker. In bizzyB we did not use automatic context observation techniques. Rather the user had to explicitly „move“ to the desired context instead of the system recognising it. This Figure 1. Data, information, and knowledge again proved to be useful for our application domain, as the set of different contexts the user community covers is An organisational memory aims to support the efficient reasonably small. and effective sharing of organisational information and One of the disadvantages of not using explicit context knowledge. To address issues like information overload, it models was that the system could not perform any kind of is necessary to condense any information given to a reasoning on context similarity. The user had to remember certain person by personalising and contextualising it. herself, that she has already been in a similar context and This paper covers the problem of supporting move to it in order to reuse information already created. contextualisation and personalisation through explicit and Another problem was that the approach was not very comprehensive context modelling techniques. We strongly flexible: a change in the work environment had to be believe that context modelling techniques help to decrease reflected in software changes. Furthermore in this an individuals information overload by delivering a approach we only regarded process-oriented context decreased amount of information of higher quality (where information, while ignoring other probably important we define the quality of information as its usefulness contextual dimensions. within the current context). In a previous paper [Kle 1999] we already presented first ideas towards the design of a context enhanced 3.2 Requirements organisational memory. In the following we will build on Table 1 summarises the reviews from the previous these ideas and present latest results. sections with respect to the identified contextual features In an other paper (see [JKN 2001] ) we offer an and the underlying (implicit or explicit) model of context. information brokering centred view on knowledge Many approaches recognise context as being a concept of management. There we identify contextualisation as an major importance. But as no consensus on what context is important task that annotates information with contextual exists in the research community we can observe that information (domain knowledge and situational R. Klemke 14-5 Table 1. Context Features and Context Modelling. Work Feature taken as context How context is modelled [Ack 1994a] Simple features like submission date or author Manually provide simple meta-information [KZR 1997], [ZS Content descriptors provide context E-mail classification & hypertext for “conversational 1997] modelling” [Pri 1993] Creation context of entities (department, project) Annotation as contextualisation based on organisational structure [Buc 1997] History of decision processes Argumentation visualisation [HSK 1996] Employees knowledge descriptors Manually constructed knowledge profiles [ABHKS 1998] Organisational structure Enterprise ontology [Schwa 1998] User centric meta-knowledge User profiles & shared semantics [MSPK 2000] Domain concepts from domain ontology Domain ontology; manual concept selection [KO 1997] Concept relations Matrix-based relation calculation [Gök 1999] History of IR system usage Learned through machine learning [FOS 1997] Conceptualised e-mails Conversation modelling [MD 1998] Processes to which documents are linked Process modelling [Wol 1997] Organisational roles and process relations Enterprise modelling [Rei 1998] Workflow process context Integration of OM and WMS [WWT 1998] Workflow process context Evolutionary WMS [KS 2000] Knowledge creation & use processes Business process models [GS 1997] Captured live events related to papers Association of papers and multimedia data [KK 1999] Information Brokering Process Process Modelling and Process Context Visualisation concentrate on one or two of the contextual aspects Process (e.g. Workflow) presented there. We feel a strong need to look at context Organisational Structure (e.g. Enterprise more comprehensively, which leads us to: Ontolgy) Requirement 1: A context modelling framework Domain Ontology has to identify all relevant contextual dimensions. Domain/Content based What does requirement 1 mean? Context may be defined Knowledge Profiles as “any information that can be used to characterise the Context situation of an entity; an entity is a person, place, or User Profiles / User Models object that is considered relevant to the interaction Personal between a user and an application, including the user and Interest Profiles applications themselves” [DA 1999]. The amount of Location information that could possibly characterise a given Physical situation is obviously far too big to be handled. Time Additionally, some potentially relevant contextual dimensions are hard to identify automatically or even hard Figure 2. Context Typology to be explicated at all (e.g. the current personal mood or the intention behind a certain action). every approach focuses on a special aspect of context. Therefore we require a set of relevant contextual Based on the different aspects of context being dimensions which: modelled by different approaches we found in the are of relevance to characterise a situation (i.e. it literature and based on our own experience, we have successfully allows to distinguish different situations) defined a context typology for working contexts as can be explicated depicted in figure 2. Most of the reviewed works can be (semi-) automatically identified R. Klemke 14-6 are sufficiently small in number to allow efficient Some approaches that associate information with storage and retrieval organisational models, software engineering process allow the definition of a set or range of possible values models or general workflow process models have already of sufficient accuracy been discussed above (see [MD 1998], [Pri 1993], [WWT allow the measurement of the similarity of each pair of 1998], & [Wol 1997]). These approaches have shown that values for a given dimension. information may be retrieved context-based (i.e. a user who is in a similar context can view, browse or retrieve While in [AMGPS 1996] organisational context is defined the contextualised information). None of these approaches along three dimensions ( organisation dimension, process however, maintains an explicit context model that would dimension, space dimension ) each of which is further allow additional retrieval strategies (e.g. match query & hierarchically refined in [Len 1998] twelve dimensions for similar context, match query & complementary context, describing contexts are identified in the background of match query only, or match context only). This leads us modelling and reasoning within real world knowledge. to: We define a dimension as relevant if it allows to separate information into groups that literally “make Requirement 3: Context-based and content-based sense”. Consequently, “outside temperature” is probably retrieval of information have to be possible not a relevant organisational dimension while “time” and independent of each other as well as in “process” are. Additionally, a relevant dimension needs to combination. be easily explicable to be included in the set of While the a priori modelling of contexts (and the represented dimensions. This strongly relates to the notion corresponding implementation mechanisms to exploit of explicit and tacit knowledge [Non 1994], where context information in an information system) is the right externalisation (i.e. the transformation of tacit knowledge approach for a domain with clearly structured work into explicit knowledge) is an essential part in the process processes that remain stable over a long period of time of corporate knowledge creation. We believe, that only (like information brokering), we now feel a strong need some parts of existing tacit knowledge can easily be towards more flexible approaches for other domains. explicated. Also, we think that a useful system should automatically Approaches in the IR community that try to make use recognise the users current context, to be able to provide of context knowledge to improve retrieval results have possibly needed information created in similar contexts been discussed above. They vary from long term user immediately. Thus the fourth requirement is: interest profiles (created explicitly by the user) to regarding the users retrieval history (observed Requirement 4: Automatic recognition of context automatically by the retrieval system) and similar should be done as well as giving users the approaches. All of these have in common that they only possibility to explicitly provide context look at the consumption side of the information retrieval information (thus simulating a certain context). process to make use of context. The production / provision side is not considered in these approaches. For 4 Architecture general purpose IR systems an approach to Based on the requirements defined above we now contextualisation of information at provision time would describe our architectural approach in more detail. not be appropriate as producers and consumers of Therefore we outline possible contents of organisational information are separated groups which presumably context models, followed by a description of our makes their context incomparable. This situation changes architecture giving an overview over the ContextService when we look at OM, which can be seen as special kind of and ContextAgent components (see figure 3). IR systems. OM contain information produced and consumed by the same group of people: the members of 4.1 Content of context models the organisation. Thus they share the same range of possible contexts. This leads us to the next requirement: Based on requirement 1 and our definition of relevance of Requirement 2: In a context-enhanced a context dimension we will now identify basic organisational memory system, context dimensions of context that we think are important for knowledge has to be associated to information at organisational context: production and consumption time and has to be A person is uniquely identified by an ID and/or a used during information retrieval. name. A person's context is further characterised by This requirement is based on the idea that knowledge her position within the organisation, her roles, her about the current context of a user may be used for at least skills, her interests and experience. two purposes: A location a person works at is not only characterised by its co-ordinates ( absolute location ) but also by to enhance any information currently created, further characteristics as name (e.g. Room number) modified, published, or used by the current user and and function ( type of location, e.g. Office vs. Meeting to offer possibly useful information created, modified, room) or published in contexts similar to the current user's A point in time may simply be described as absolute one. time but further characteristics are important for its R. Klemke 14-7 contextual description: e.g. something happened on a Monday morning ( type of time ), or something User happened two hours before something else ( relative Environment time) An activity describes what someone is currently doing. ContextAgent This is defined by the task a person has to fulfil (e.g. embedded in a workflow process), by the tools used to fulfil the task, by files opened and further characteristics Organisational ContextService Memory Through ontological refinement and association these basic dimensions cover all identified contextual aspects from the identified context typology in figure 1. Each of the attributes that further define the basic context Domain CM/OM Context dimensions can be of different types: they either are Contents Bridge Models represented by primitive values (like a timestamp, an ID, or a name) or they may be represented using complex values (e.g. a categorisation hierarchy to classify Figure 3. Context Framework Architecture organisational roles or interests). We use a fully implemented ontology-based knowledge modelling tool associate document identifiers (URLs) with context (Broker’s Lounge, see [JKN 2001] for a detailed models. discussion) that offers a flexible and user friendly ContextService offers an API which can be used to approach to model complex knowledge structures. The store new context models, retrieve stored ones, associate main benefit of using Broker’s Lounge is, that it offers a new document identifiers with contexts and perform type-based ontology-modelling approach that allows to context-based document retrieval. In particular, the create multi-dimensional knowledge models. We use this following API functions are offered: approach to model the relevant contextual dimensions we similar: ContextModel -> {ContextModel 1, ..., identified. Additionally, Broker’s Lounge offers a ContextModeln}, delivers a set of ContextModels that separate knowledge structuring level (categorisation level) are similar to the given one which we use to express similarities among different getDoc: ContextModel -> {DocID 1, ..., DocID n}, elements of the ontology. It is out of the scope of this delivers the set of document identifiers being paper to provide further detail about the underlying associated with the given ContextModel approaches of the Broker’s Lounge environment. getContext: DocID -> {ContextModel 1, ..., ContextModeln}, delivers the set of ContextModels 4.2 The Context Framework Architecture being associated with the given document identifier It is our aim to provide a component-based system that addDoc: ContextModel, DocID -> Ø, associates a can be easily integrated with existing intranet-based ContextModel with a document identifier, i.e. stores information systems. Therefore we impose only simple the ContextModel and creates an association of requirements to the existing environment: documents have ContextModel and DocID in the CM/OM Bridge. The to be identifiable using URLs and these URLs have to CM/OM Bridge is required to maintain the remain stable throughout the document lifetime. A URL independence of ContextService from the chosen OM. does not necessarily point to a pure HTML document, any Especially the API function “similar” is of importance: it other kind of document format is supported as well (as must be possible to retrieve similar context models from well as dynamic query URLs). the potentially huge collection in an efficient way. The In the following we will describe two central retrieval of similar context models is complicated by the components of the Context Framework: ContextService complex nature of the models. As we have seen in the and ContextAgent. ContextService is a background previous section, context models are multi-dimensional component that manages all existing context models and each dimension may have a hierarchic (topological) within the organisation and offers an API for retrieval and structure, that makes the design of similarity measures a storage of context models while ContextAgent is the main non-trivial task. In our current implementation we use a component for handling user interaction, automatic combination of a weighted distance measure for the final context observation and interaction with the user's distance calculation and separate distance measures for environment. each dimension. The similarity measure for the time dimension is a 4.2.1 ContextService combination of an absolute distance measure and a type of The ContextService component stores all known context time similarity measure. The type of time measure tries to models in a database. It is responsible for maintaining the find structural commonalties within two points in time history of context models for every user within the (e.g. both values represent a Monday morning but within organisation. Furthermore it offers the possibility to different weeks). Location similarity is calculated as combination of absolute spatial distance and type of place R. Klemke 14-8 similarity. Type of place similarity calculates the semantic organisational models. Further organisational information distance of two places (assuming that a location's semantic sources (e.g. organisational people database) offer more is its role e.g. as office or meeting room). The type of or less stable data about the user, e.g. information about place similarity measure is based on a taxonomic her position & roles may be collected. Information about description of all available types of places within an the highly dynamic current context is more difficult to organisation. Similarity measures for persons and extract, as reliable, quality controlled entries in databases activities are based on semantic distance calculation of are no useful sources here. Sources of information are the their respective taxonomic description. user herself (explicitly providing contextual information), While the time and activity similarity measures are the set of tools currently used (e.g. gathered through independent of other dimensions, similarity measures for interaction with the task manager) and additional location and person have a temporal aspect (e.g. a meeting information from some organisational database about the room becomes an office as an organisation grows and new purpose of each tool used within the organisation, or members arrive or the position of an organisational information gathered through interaction with a set of member changes during time). This requires to take the specially designed tools (e.g. workflow management history of persons and locations into account when systems, information systems, organisational memory, IR measuring their similarity. systems, or even the query history of ContextAgent itself). To improve the retrieval performance of our current implementation we evaluate several techniques from the 4.2.3 Integration case-based reasoning community (e.g. [DLTBP 1996], The ContextService component is designed to be [RS 1998], [Scha 1996]). integrated with our Broker’s Lounge knowledge By combining the API functions it is possible to create management environment [JKN 2001]. This allows to complex retrieval scenarios as e.g. document-based combine context-based retrieval with all retrieval retrieval of documents created in similar contexts as the techniques offered by Broker’s Lounge (full-text, concept- given one. To allow a greater retrieval flexibility further based, category-based, domain-relevance-based) to reach API functions are defined, that allow the manipulation of a flexible and comprehensive set of retrieval capabilities. threshold values and similarity weights. Additionally, it is also possible to integrate ContextService with any kind of intranet-based 4.2.2 ContextAgent information management solution, as long as it allows the The ContextAgent component is the main point of user identification of documents with URLs. The integration interaction with ContextService. It serves as intermediary with these tools will be twofold: between the user and ContextService, offering the Firstly, when documents get submitted to the following kinds of interaction: traditional KM tool ContextService needs to know their ContextAgent may automatically observe the user's identifier and the valid ContextModel. The process of current context and recognise context shifts. To recognise adding a document has to be changed slightly therefore. the user's context ContextAgent observes the set of tools Rather than adding a document to the KM tool directly it used by the user, interacts with a set of specifically will be “added” to ContextService. ContextService in turn designed tools like workflow management tools, forwards the add operation to the KM tool and simply information management systems, organisational memory stores the identifier and the associated ContextModel. systems, information retrieval systems, and observes This does not require an changes to the API of the KM names and locations of files currently worked with. tool, just the corresponding ContextService wrapper has Instead of relying on the automatic context recognition a to be provided. user may also explicitly provide information on his Secondly, queries to the traditional KM tool will also current context (or any other virtual context). be handled by the ContextService, in order to extend or When ContextAgent recognises a context shift it reduce the number of hits given by the KM engine. interacts with ContextService to retrieve relevant Therefore queries will have to be send to both systems information from contexts similar to the current one. and the results will have to be combined. The only thing Results of this operation are proposed to the user in a that has to be done to provide this, is to write a query none-disruptive manner. The user may look at the wrapper, that forwards queries to ContextService and the information proposed or ignore it and simply continue her existing KM tool and combine the results. This integration daily work. On user demand ContextAgent performs the is straightforward. retrieval operation explicitly, either using the automatically recognised context or the explicitly user 5 Conclusion & Future Work defined one. We have shown the state of the art in organisational ContextAgent makes use of different information memory systems with a special focus on the notion of sources to build the complete model of the user's context. context. Based on our previous experience on context- By using location aware components (e.g. the based information access in the domain of information ContextToolkit, [DA 1999]) and time observation precise brokering we presented our requirements towards context- data about the user's temporal and geographical context is based organisational memories. Our approach is based on gathered. Knowledge about location types (e.g. office or ontological context modelling, automatic context meeting room) may be further inferred from R. Klemke 14-9 observation and similarity measurement of different context. context models. We have presented the Context Despite the limited implementation we have Framework architecture, which realises the presented encouraging experiences from first system uses. In the ideas. near future we mainly plan to extend the current version in By offering a completely new range of information two directions: firstly, we want to improve the retrieval & information filtering methods that take the ContextService performance and context model working situation of employees into account complexity. Secondly, we want to extend ContextAgent ContextService aims to significantly improve the access of with further automatic observation functionality which individuals to organisational memory systems and through will then also offer the possibility of active context-based extending information with context-based meta- information provision by ContextAgent. information it improves the way organisational information is organised. 6 Acknowledgements The aim of the presented approach is to provide help to Special thanks go to Dawit Yimam and Achim Nick for workers by providing needed information at the right time helpful comments on earlier versions of this paper. I and at the right place. It is not our aim to control or would also like to thank the research group at GMD.FIT supervise workers within an organisation. As this for fruitful discussions and insights. Parts of this work approach depends on the motivation of people to have been funded by the EU project COBRA. participate in knowledge sharing processes, it is important that users trust in the system. Therefore means of security have to be offered to users. It has to be assured that References information is only published to others if the user [ABHKS 1998] A. Abecker, A. Bernardi, K. Hinkelmann, explicitly agrees and is kept private otherwise. O. Kühn, M. Sintek . ”Toward a Technology for Furthermore organisational agreements are important that Organisational Memories”, IEEE Intelligent Systems & guarantee the privacy of automatically gathered Their Applications, May/June 1998. information and that clearly define the possible range of uses of this information. [AAST 1998] A. Abecker, S. Aitken, F. Schmalhofer, B. We believe that if the security and privacy concerns of Taitschian. ” KARATEKIT: Tools for the knowledge- individuals are respected sufficiently the approach creating company ”, 11 th Workshop on Knowledge presented can efficiently improve the information Acquisition, Modelling and Management KAW’98, distribution within organisations. Banff, 1998 By providing a service oriented open framework, the ContextService can be integrated with any kind of KM [Ack 1994a] M. S. Ackermann. ” Definitional and tool, that allows the external identification of documents Contextual Issues in Organizational and Group with identifiers (URLs). This allows to integrate Memories”, Proc. of 27 th Hawaii Int’l Conf. on System ContextService with a wide range of Intranet-based Sciences (HICSS’94), 1994 information management systems. Currently, the implementation of ContextService and [Ack 1994b] M.S. Ackermann. ” Augmenting the ContextAgent is in a prototypical stage. A first Organizational Memory: A Field Study in Answer ContextService prototype exists that offers the defined Garden”, Proc. of ACM Conf. on Computer Supported API functionality as described above. However, the range Cooperative Work (CSCW’94), 1994 of context models (i.e. the set of values for each of the [AM 1990] M. S. Ackermann, T. W. Malone. ” Answer defined dimensions) is limited in the current version. The Garden: A Tool for Growing Organizational Memory ”, retrieval of similar context models is not very efficiently Proc. of ACM Conf. on Office Information Systems, done within the current version. We expect major improvements for the retrieval efficiency from case-based Cambridge, 1990 reasoning approaches. [AM 1996] M.S. Ackermann, D.W. McDonald. ” Answer The Broker’s Lounge system, which is the basic Garden 2: Merging Organizational Memory with ontology-based knowledge management toolkit [JKN Collaborative Help”, Proc. of ACM Conf. on Computer 2001] for developing ContextService and ContextAgent supported Cooperative Work (CSCW’96), 1996 exists as a fully implemented prototype that has already successfully been used for further applications [NKS [AMGPS 1996] A. Agostini, G. de Michelis, M.A. 1998]. Grasso, W. Prinz, A. Syri: “ Contexts, Work Processes, The implementation of ContextAgent is in a and Workspaces ”, Computer Supported Cooperative preliminary stage. Currently, the automatic context Work: J. of Collaborative Computing 5: 223-250, 1996 observation is restricted to the time and person dimensions of the user’s context. We experiment with two [AL 1999] M. Alavi, D. Leidner. “ Knowledge components to observe location (Context Toolkit, [DA Management Systems: Emerging Views and Practices 1999]) and activity (Envoy) and plan to integrate these from the Field”, Proc. of 32 nd Hawaii Int’l Conf. on within our Context Framework, but for now, the user has System Sciences, Hawaii, 1999. to define the values for these dimensions of her current R. Klemke 14-10 [BCR 1994] V.R. Basili, G. Caldiera, H.D. Rombach. [KO 1997] S. O. Kimbrough, J. R. Oliver. ” On Relevance ”Experience Factory ”, J. J. Marciniak: Encyclopedia of and Two Aspects of the Organizational Memory Software Engineering, vol. 1, John Wiley Sons, 1994. Problem”, Proc. of the 4 th Workshop on Information Technology and Systems, 1994 [BHS 1998] A. Bernardi, K. Hinkelmann, M. Sintek: ”Information Systems in Knowledge Management - An [KS 2000] R. Klamma, S. Schlaphof (2000). ” Rapid Application Example ”, Proc. of 1 st Int’l Conf. on Knowledge Deployment in an Organizational-Memory- Practical Applications in Knowledge Management Based Workflow Environment ”, 8th Europ. Conf. on (PAKeM’98), London, 1998 Information Systems, ECIS 2000, Vienna, July 2000. [Buc 1997] S. Buckingham Shum. ” Negotiating the [KK 1999] R. Klemke, J. Koenemann. “ Supporting Construction and Reconstruction of Organisational Information Brokers with an Organisational Memory ”, Memories”, Universal Computer Science, 3(8), in: XPS-99, 5 th German Conf. on Knowledge-Based Springer, 1997 Systems, Workshop on Knowledge Management, Organizational Memory and Reuse, Internal Report, [DLTBP 1996] J. Daengdej, D. Lukose, E. Tsui, P. Würzburg University, March 1999 Beinat, L. Prophet. ” Dynamically Creating Indices for Two Million Cases: A Real World Problem ”, Advances [Kle 1999] R. Klemke. “ The Notion of Context in in Case-Based Reasoning, LNAI 1168, Springer, 1996 Organisational Memories ”, CONTEXT'99, Modeling and Using Context, LNAI 1688, Trento, Italy, [DA 1999] A. K. Dey, G. D. Abowd. “Towards a Better September 1999, Springer Understanding of Context and Context-Awareness ”, Technical Report GIT-GVU-99-32, College of [KZR 1997] M. Kantor, B. Zimmermann, D. Redmiles. Computing, Georgia Institute of Technology, 1999 ”From Group Memory to Project Awareness through use of the Knowledge Depot ”, Proc. of the California [DCGR 1998] R. Dieng, O. Corby, A. Giboin, M. Ribière: Software Symposium (CSS’97), California, 1997 ”Methods and Tools for Corporate Knowledge Management”, 11 th Workshop on Knowledge [Len 1998] D. Lenat: ” The Dimensions of Context- Acquisition, Modelling and Management KAW’98, Space”, at: http://www.cyc.com/publications.html, 1998 Banff, 1998 [Lin 1996] S.N. Lindstaedt. ” Towards Organizational [FOS 1997] G. Fischer, J. Ostwald, G. Stahl: ”Conceptual Learning: Growing Memories in the Workplace ”, Proc. Frameworks and Computational Support for of Int’l Conf. on Computer Human Interaction Organizational Memories and Organizational (CHI’96), ACM, 1996 Learning”, Project Proposal, 1997 [MSPK 2000] M. Mach, T. Sabol, J. Paralic, R. Kende. [GS 1997] B.R. Gaines, M.L.G. Shaw. ” Knowledge ”Knowledge Modelling in Support of Knowledge Management for Research Communities ”, Proc. of Management”, Proc. of Conf. on Research Information AAAI Spring Symp. Artificial Intelligence in Systems, CRIS 2000 Knowledge Management, Stanford, 1997 [MD 1998] F. Maurer, B. Dellen: ” A Concept for an [Gök 1999] A. Göker. ” User Context Learning for Internet-based Process-oriented Knowledge Intelligent Information Retrieval ”, project proposal at Management Environment ”, 11 th Workshop on http://www.scms.rgu.ac.uk/staff/asga, Robert Gordon Knowledge Acquisition, Modelling and Management University, Aberdeen, 1999. KAW’98, Banff, 1998 [Hei 1962] M. Heidegger. ” Being and Time ”, Harper & [MK 1996] S. Matwin, M. Kubat. ”The role of Context in Row, New York, 1962. Concept Learning ”, proc. of the 13 th int’l conf. on machine learning, Bari, Italy, 1996. [HSK 1996] G. van Heijst, R. van der Spek, E. Kruizinga. ”Organising Corporate Memories ”, Proc. of 10 th [Mur 1996] L. D. Murphy. ” Information Product Knowledge Acquisition for Knowledge-Based Systems Evaluation as Asynchronous Communication in Workshop, 1996. Context: A Model for Organizational Research ”, Proc. of the 1st ACM int’l conf. on Digital Libraries, 1996. [JKN 2001] M. Jarke, R. Klemke, A. Nick. “ Broker’s Lounge - An Environment for Multi-Dimensional User- [NKS 1998] A. Nick, J. Koenemann, E. Schalueck. Adaptive Knowledge Management ”, accepted for 34th ”ELFI: Information Brokering for the domain of Hawaii Int’l Conf. on System Sciences, Hawaii, 2001. research fundingt”, Int’l Journal of Computer Networks and ISDN systems, Oct. 1998, pp. 1491-1500 R. Klemke 14-11 [Non 1994] I. Nonaka. “A dynamic theory of [Schwa 1998] D.G. Schwartz. ” Towards the use of user- organizational knowledge creation ”, Organizational centric meta-knowledge in applying organizational Science, Vol. 5, No. 1. memory to email communications ”, Proc. of Europ. Conf. on Artificial Intelligence (ECAI’98), [Oar 1997] D.W. Oard. ” State of the Art in Text Interdisciplinary Workshop on Building, Maintaining Filtering”, in User Modeling and User-Adapted and Using Organizational Memories (OM-98), 1998 Interaction 7: 1997 [SMDP 1998] R. M. Strens, M. Martin, J. E. Dobson, S. [Pri 1993] W. Prinz. ”TOSCA – Providing organisational Plagemann. “Business and Market Models of Brokerage information to CSCW Applications”, in: Proceedings of in Network-Based Commerce “, Proc. 5 th Int. Conf. on Int’l Conference on Computer supported Cooperative Intelligence in Services and Networks (IS&N’98), May Work (ECSCW’93) Kluwer Press, 1993 1998, Antwerp, Belgium. LNCS, Springer, 1998. [Rac 1997] P. Y. Raccah. ” Science, Language, and [SBF 1998] R. Studer, V.R. Benjamins, D. Fensel: Situation”, in: proc. of the 2 nd Europ. conf. on Cognitive ”Knowledge Engineering - Principles and Methods ”, Science, workshop on context, Manchester, UK, 1997. Data & Knowledge Engineering 25 (1-2), March 1998 [Rei 1998] U. Reimer: ” Knowledge Integration for [Sri 1997] K. Srinivas: ” How is context represented in Building Organizational Memories ”, Proc. of 11 th implicit and explicit memory ”, Proc. of the Europ. Conf. Workshop on Knowledge Acquisition, Modelling and on Cognitive Science, Workshop on Context, 1997. Management KAW’98, Banff, 1998 [WWT 1998] C. Wargitsch, T. Wewers, F. Theisinger. [RM 1998] M. Ribière, N. Matta. ”Virtual Enterprise and ”An Organisational-Memory-Based Approach for an Corporate Memory”, Proc. of Europ. Conf. on Artificial Evolutionary Workflow Management System – Concepts Intelligence (ECAI’98), Workshop on Building, and Implementation ”, Proc. of the 31 st Annual Conf. on Maintaining and Using Organizational Memories, 1998 System Sciences, Vol.1, Los Alamitos 1998. [RS 1998] F. Ricci, L. Senter. “ Structured Cases, Trees th [Wol 1997] M. Wolverton: ”Exploiting Enterprise Models and Efficient Retrieval ”, Proc. of the 4 Europ. for the Automatic Distribution of Corporate Workshop of Case-Based Reasoning [EWCBR-98], Information”, Proc. 6 th Int’l Conf. on Information and Dublin, Ireland, Sep. 1998. Knowledge Management (CIKM’97), Las Vegas, 1997 [Scha 1996] J.W. Schaaf. “Fish and Shrink: A Next Step [ZS 1997] B. Zimmermann, A.M. Selvin. ”A Framework Towards Efficient Case Retrieval in Large-Scale Case for Assessing Group Memory Approaches for Software Bases”, Advances in Case-Based Reasoning, LNAI Design Projects”, Proc. of DIS’97 1168, Springer, 1996 R. Klemke 14-12