2 Recommended References General literature on knowledge management:J. Liebowitz (ed.): Knowledge Management - Handbook. CRC Press 1999.Harvard Business Review on Knowledge ManagementReference for change management:B. Dellen: Change Impact Analysis Support for Software Development Processes. Shaker Verlag 1999.Abstraction and Cases:R.Bergmann: Effizientes Problemlösen durch Wiederverwendung von Fällen auf verschiedenen Abstraktionsebenen, DIKI 138, infix Verlag 1996Processes and information:Boris Kötting, Michael M. Richter, Sigrid Goldmann: Flexible Workflow Management in Software Engineering Processes
3 KM and Suppliers Utility The overall guiding line for knowledge management activities is provided by the preferences and utilities of the supplier who has, however, to take care of customer needs.This is mainly reflected in the strategic model of the supplier.Here this will not be discussed, we instead look at the consequences for the formal models and actions of the supplier. They will be discussed on a general level and have to be instantiated for specific applications. Each application has its own characteristics.An orientation is given by the sales cycle and its refinements in chapter 1, augmented by the suppliers view.
4 The Flow of Knowledge Data Knowledge Base Data Bases Information restructureKnowledge BaseData BasesInformationmake explicituseFlow from external sourcesKnowledgeActions
5 The General SzenarioWe assume a general agent scenario (see chapter 14B).The agents (humans or machines) areOne or more actors who carry out certain actionsA knowledge manager KM, who has access to information sources, and who has to structure, maintain and apply them;An external environment which can generate events.In this szenario a communication goes on, and any of the agents can take the initiative.The knowledge manager has to interact with all other management activities because they all need the knowledgeThe KM can act on demand and on his own: „pro-active“
6 Active and Passive Active and passive are roles of agents Passive means in a context that the action an agent performs is a reaction on some other action which contains a demand.Active means that the action is not determined by a demand but the agent sees a necessity from an overall point of view. The action is usually triggered by some event The action usually asks for a further reaction.The switch from the role “passive” to “active” is called to be pro-active.
7 Fully Automatic Systems In the past knowledge based systems worked fully automatic:They contained a correct and complete knowledge base.When they obtained an input they derived the output via reasoning (using an inference component) applied to the knowledge base.Often the desired the output contained besides the problem solution some explanation.The problems connected with fully automatic systems wereto achieve and to maintain such a knowledge base (the “knowledge acquisition bottleneck)partial solutions were useless because they are reported as failures.
8 Assistant Systems (1)The idea of an assistant system is to operate only partially automatic and to employ humans too.A consequence is that assistant systems usually do not perform long chains of inferences.Advantages of assistant systems are:The work can employ knowledge and abilities of humans.To shift tasks between human and machine: If a task is fully understood and all knowledge for it is available it can be transferred to the machine, i.e. automized. This can be done incrementally.Humans can take over the responsibility for decisions.
9 Assistant Systems (2) Assistant systems and knowledge: The humans use and need knowledgeKnowledge helps the humanA knowledge based system to support humans has to have the character of an assistant system. Consequences:Knowledge and its use has to be integrated in the general structure of the organizationThe division of labor between human (e.g. decision maker) and machine (automatic presentation of knowledge) has to be well understood
10 Assistant Systems (3)Assistant systems are a special kind of knowledge based systems. Two types of agents cooperate:Human agentsSoftware agentsHence assistant systems are examples of socio-technical systems.The human agent is dominating:Sets the goalsIs responsibleThe human agent is creative.The human agent cannot deal well with large data sets, complex computations etc.
11 Assistant Systems (4)The software agent (the assistant) is subordinate:The decisions are of limited characterThe space of freedom for actions is limited and precisely defined.The assistant knows :The range of the decisions, i.e. what to do on his own with which degree of freedomWhom to inform about the decisions.Often the assistant has only to provide information:in a very effective waynot to much and not lessat the right time.
12 Assistant Systems (5)A major problem is the interface between human and software agents which is the bases for communication:The software agent acts as a formal system and requires formal input in a specific representation formThe human agent has limited memoryThe human wants information in a form suitable for human understanding and reasoning.Therefore the interface has to perform a non-trivial transformation. The use of interchange formats like XML may be helpful but is by no means sufficient.The interface has also to reflect that the human has the responsibility for descisions.
13 Classification of Knowledge Management Tasks 1) Searching for knowledge and receiving knowledge2) Restructuring the knowledge3) Making knowledge explicit4) Associating the knowledge with the actions described in the process model5) Making knowledge available for actions which need it and delivering it to the right agents in the right moment6) Updating knowledge and change management7) Quality management
14 Knowledge Management and General Management All these tasks cannot be separated from the general management activities.Knowledge is used when actions are performed and actions are organized by the management.Actions on the other hand change the knowledge, e.g.organizational changeschange in the employeeschange of the context (new products, customers etc.)Therefore knowledge management is a central element of management.
15 Knowledge Management: Technical Aspects Knowledge is electronically stored in data bases on computers.These locations have arisen historically and are often not compatible with each other.A consequence is that important knowledge cannot be found when necessary.The first step for the management is to define a knowledge structure in order to know “where is what”.The second step is to organize communications, e.g. by introducing adequate client-server structures.
16 Task (1): Searching and Receiving Knowledge Data, information and knowledge does not come from itselfSome sources of knowledge are known, others have to be foundKnowledge sources do not continuously have new or interesting knowledgeThe management task isGet an overview over sources and organize the search for themDetermine the times (or periods) when sources have new knowledgeOrganize the access to and the flow from the sourcesReceive the demanded knowledge properlyClassify and receive the knowledge which came in but not on demandIt is important that these activities are standardizedTechniques of document analysis are important
17 Document Oriented Knowledge Structure The knowledge is in documents and the content is clear from the document description; the type of reaction to the content is known from the type of the document, e.g.:BillsNew pricelistNew product listChange of addressAccess to the knowledge inside of the documents does therefore not require to study the document itself.This is like the access in data bases where one has only to know the key of the stored data.The flow of knowledge therefore reduces to the flow of documents and the search for knowledge is the search for documents.
18 TablesOften knowledge is organized in tables with a number of columns and rows.Such tables are presented in a certain layout.It is not always easy to reconstruct the original table from the presented layout:lenght of entries may varyentries may contain several linesDistances between rows or columns may varyThis puts restrictions for extracting the content of the table from the table document.
19 Content Oriented Knowledge Structures (1) It is not sufficient to know the title or the key of the document in order to react properly.It is rather necessary to study the document itself.Examples:Complaints from customersSpecial regulations for special purposesScientific documentsLegal documentsThe access to the documents should be simplified, e.g. using abstracts, extracting key words etc.
20 Content Oriented Knowledge Structures (2) The knowledge management should structure the knowledge and simplify the access.This is an area where the similarity concept plays an important role because no exact key matches are possible but often inexact matches with document descriptions are applied.Linguistic tools, Thesauri etc. are useful.In most situations, content oriented structures are still handled by humans. But the humans need support.
21 Task (2): Restructuring Knowledge (1) The incoming data, information and knowledge are usually not structured in the form required from the applications, e.g.Wrong formatRedundantIn an inadequate contextNot applicable etc.The task of the knowledge management is to organizeRestructuringPointing out weaknesses and getting other sourcesAgain, this should be standardized
22 Task (2): Restructuring Knowledge (2) Restructuring has to aspects:Restructuring of a single input documentEmbed ijn or distribute the input over the whole knowledge structure.The whole knowledge structure is determined by the general structure of the company. Therefore the proper handling of input knowledge is connected with the general structure and strategy.It may be necessary to duplicate knowledge used by different agents.Different agents may need knowledge pieces in different forms or formats.
23 Task (3): Making Knowledge Explicit (1) Knowledge is often implicitly contained in data or texts.It is the purpose of data mining techniques to make knowledge in data bases explicit.The knowledge management has to organize this:Where are weak points ?Which information can be helpful for improvement ?How to obtain the information ?Data mining activities are long term activities, they are costly and need careful planning (see chapter 13).The knowledge managers decides which data mining activities have to be carried out.
24 Task (3): Making Knowledge Explicit (2) Knowledge in texts can at least partially be made explicit byExtracting key wordsextracting phrasesextracting abstracts.The key words, phrases or form of the abstract has to be determined according to the needs of the users. Problems arise if different user types are present.Key words are often insufficient or even misleading.Such techniques have been developed in information retrieval and use e.g. liguistic tools.
25 Task(4): Which Knowledge for What ? The use of knowledge in business is not for fun butIs oriented on business processesInfluences partially the general structure of the processesHas to allow a fast and optimal representation of the knowledge in actual contextsIf no actions are involved the knowledge is silent !If actions are performed without knowledge they are useless !
26 Knowledge and Processes general processneedsneedsactual data andinformationgeneral knowledgeinstanceactual process
27 Types of Processes (Examples) Sales offer: A dialogue has to be startedDesign and planning processes:Process models are instantiatedExecution processes:Correct information of participants(A problem if changes occur: Who hasto be informed about what ?)Fault diagnosis: Reasons for failure ? Which data areneeded for the diagnosis ? (Help desk problem!)Logistic chains:Transportation and delivery over several stepsProcesses may deal with physical objects or pieces of information.
28 Knowledge Support for Processes (1) Step in the general process (process model)InstantiationDetailsCorresponding steps in the actual processThe knowledge manager organizes the necessary sourcesKnowledge SourceKnowledge SourceKnowledge Source
29 Knowledge Support for Processes (2) The process model can be described in various levels of abstraction. On each level the preconditions and effects of an action are described in an appropriate abstraction.On abstract levels the types of knowledge dominate. Each type is associated with a knowledge source.The main tasks of the knowledge manager include:Structuring the knowledge sources according to the process modelDistributing the knowledge correctlyEstablishing the links between the actions in the process model and the knowledge sources dynamically (i.e. observing the time schedule)
30 Information GoalsThe actual information needed for a process is usually incompleteThe information is available from internal or external sourcesCosts are involved in order to obtain the informationThe information has some value for the actions chosen in the processConsequence for knowledge management:Define information goals and a plan to achieve them in order to have optimal effects
31 Cost/Benefit Consideration Cost aspect: Obtaining information has costs (direct payment, time of employees etc.)The value of the information is determined by the actions performed:Performance of actions is more costly if done without the right knowledge. It is important do quantify this properly !Executed actions lead to new situations which have now costs or gains as a consequence. Knowledge can make predictions.The value of a piece of knowledge is the difference of costs connected with the action when performed with or without the knowledge.This has an individual and a statistical interpretation.
32 Standardized and Non-standardized Processes Standardized processes occur regularly in the same way although each instance has different data inputs.Non-standardized processes also occur often but only the principal task of the process is known and each time the process has its own appearance. A general process model may be too abstract to be useful.Non-standardized process should not be mixed up with completely new and surprising actions which react on unforeseen events and which have no process model at all.
33 Standardized Processes Because the structure of the process is known it is also known which type of knowledge is needed in order to perform them properly.The task of the knowledge management is to provide such knowledge and data structures such that knowledge support is simple: Here the support is document oriented. For this purpose the KM has to watch the process.Example: Because employees have their regular vacations a corresponding list is advisable and persons on vacation can be replaced properly. This task is more difficult when persons become sick or leave the company.
34 Non-standardized Processes Because of the somewhat irregular type of these processes it is often not possible to provide standard documents with the knowledge needed.On the other hand the type of knowledge is known and it is the task of the knowledge management to make access to this knowledge possible.Example: It cannot be foreseen which employee will be sick and a list of all possible replacements for every sick person is usually impossible. But the knowledge structure should it make possible to find out who can replace a person in an actual situation.
35 Task (5): Organizing the Use of Knowledge Knowledge has for eachtask to be accessiblefor the right personsat the right timeat the right placein the needed formatMissing Knowledgecreates errorsToo much knowledge confusesThis task is very complex and uses different techniques.Some will be discussed here.
36 Task (6): Change Management Knowledge is not invariant but undergoes continuous changes. There are external reasons for this (the context changes) as well as internal reasons (e.g. organizational changes).These changes have to be reported at the right time to those agents who need it.The report can be given on demand as well as pro-active.The change management organizes this in a systematic way.
37 Task (7): Quality Management (1) The aspects of quality are a consequence of the utility concepts.Quality decreases over time due to changes (external as well as internal) if no reaction takes place.The quality of the processes has to be controlled continuously:Oberservation of the environment dataObservation of the processInterpretation of observed data on the basis of quality models.The results of the control are transformed into actions which re-establish the quality.
38 Quality Management (2) Quality conditions are defined as constraints Hard constraints: Have to be satisfied in any caseWeak constraints: should be satisfied but not under all circumstances.Weak constraints have degrees:in hierarchical orderingsby point valuationsin fuzzy degreesThis leads to an optimization task: Weak constraints should be satisfied in an optimal way. This should optimize the intended quality.
39 Quality Management (3)For each violation of a constraint a maintenance operation has to be defined.The degree of weakness of each constraint is transformed to a degree of importance of the maintenance operation.There have observable events to be defined which can easily be checked and which indicate (possible) constraint violations.On this basis a practical system has to be built in order perform maintenance efficient and economically.The knowledge manager has to ensure the quality of the knowledge and has in particular to deal with knowledge gaps (see chapter 2).
40 Maintenance The maintenance operations are structured in two ways: importanceevents which trigger the operationsThe trigger is usually an eventAn analysis can show that the events take place in certain periods of time: Then time points can take over the role of triggering.Operations which are dependent on similareventspoints of timeobjects to maintaincan be grouped into packages of maintenance operations.
41 Formal NotionsIn order to support the KM all activities and objects of interest have to be formally represented.We refer to chapter 4 with respect to the formal representation techniques.We distinguish between actions which occur in the planning phase of the manager and actions which the manager really executes. These actions will change the real world (e.g. sending a message).The knowledge manager has an own knowledge base which governs the management actions and is about the other knowledge bases.
42 Actions on Knowledge Bases A knowledge manager has to maintain the knowledge bases and this requires actions which change these bases.The formal notion of such an action is defined in chapter 4.A particular type of action is important in this content: Actions which are generated because of the change of the context (the outer world).We call such changes events.The formal representations are the ECA- Rules (Event-Condition-Action-Rules:IF Event AND Conditions THEN ActionThese rules specify the preconditions in the way that they distinguish between external events and internal conditions.
43 Changes and Dependencies Entities for a process are concepts which organize the knowledge for the process (e.g. catalogue or an internal price list).An entity E has a change impact on an entity F if a change of E results in a change of F. E.g., the internal price list has a change impact on the catalogue.An entity E is change dependent on an entity F if a change of E may result in a change of F.Two entities are change dependent if one of them is dependent on the other one.E.g., the buying price and the sales price of a product are change dependent.
44 Change KnowledgeThe change knowledge describes all aspects of changes:The source and the initiator of the changeThe description about what is changedThe reasons of the changeThe dependencies and the dependent entitiesThe rationals for the dependenciesThe impact on the dependent entitiesThe agent who executes the change
45 Change OperatorsThe actual change is described by operators. These operators are defined on information units, e.g. on attributes, formulas etc.)We distinguish (as usual) three types of operators:ADD operators: Adding an information unit.REMOVE operators: Removing an information unit.MODIFY operators: Replaces some information unit by another one. This can be defined as a macro operator in terms of ADD and REMOVE.Operators are defined on a general level and can be instantiated.
46 Information Dependency An action A is strongly information dependent on an information unit IU if A cannot be properly executed without the knowledge of IU.An action A is performance dependent on an information unit IU if A can be better executed with the knowledge of IU.In both cases the execution of A gives rise to the information goal IU.If the agent who executes A is active he sends a query to some agent (possibly the KM) or knowledge source.If the KM is active he sends the information to the agent (pro-active).
47 Pro-active Actions, Trigger and ECA-Rules Pro-active actions have to occur with a goal: Actions have to be carried out where and when it is necessary but not unnecessary or randomly.In order that such actions do not take place in an arbitrary way they have to be activated by a trigger.The most important form of triggering is again represented by ECA-rules:Event-Condition-Action RulesEvent: Something (usually external) that starts the ruleCondition: Description of special circumstances necccessaryAction: Action (e.g. giving certain information to some agent)
48 ECA-Rules Actions on demand are also described by ECA-Rules: Event: A demand or queryCondition: Description of when action is expectedAction: Action (e.g. giving certain information or help to some agent or to perform something)ECA-Rules contain important knowledgeECA-Rules have to be structuredThe structure should reflect the tasks of the KM and the process model
49 Generating and Using Rules Rule patterns are defined at compile time; they represent general knowledge. The generality is contained in the variables which can be instantiated in their domain.Patterns use typed variables (e.g. for products, documents, agents, ...).Instances of rules are generated at run time and this is again triggered by events.The time of generation of rules is not necessarily identical with the time when they are used. This time is determined by the event which is part of the precondition of the rule.
50 Example: ECA-Rules for Change Management (1) Change rule pattern:Event: ADD, REMOVE or MODIFY operation (with variables also for documents).Condition: A formula (with variables also for documents).Action: Two types of actions:1) Change actions: As in the event part2) Notify actions: An operation of the formNOTIFY(recipient, cop, rat, reason, init)where recipient is an attribute which applies to agents or an operation of the form responsible(d) where d is of type document, cop stands for the change operation in the event, rat is of type rationale for the change dependency, reason is of type reason for the change, init is of type agent or of type responsible(d); d of type document (the agent responsible for the event).
51 Example: ECA-Rules for Change Management (2) Example: Changes in the internal price list (document pl) cause a change in the catalogue (document c). Short form of the rule pattern:Event:REPLACE(pl, product prod, price x, price y)Condition:product prod in pl and in cAction:NOTIFY(responsible(c) , “event”, price of prod, new parts, price manager)An instance of the pattern is e.g. given bypl = pl1, c = c1, prod = TV, x = 750, y = 770, responsible(c1) = Hans, new parts = video, price manager = Peter.
52 Discussion (1)In the example the only action is notification. Further support would provide information how the catalogue is changed, e.g. additional information on the improved product. Here this is left to the creativity of the catalogue agent.Important knowledge is contained in the relation that product prod is in both, the price list p and the catalogue c. It is therefore useful that the agent responsible for c has for each product an entry which specifies the source of the description details.
53 Discussion (2) The KM has to define the rule patterns and the instancesgroup the rules in packagesdetermine when the rules are appliedorganize the network through which the communication takes place and how the actions are executeddetermine how the notification is received (e.g. : is a receive notification demanded ?).These are organizational aspects responsible forsafeefficientflow of information.
54 A Change Management Architecture (1) Agents and their relations:The knowledge manager KM: Takes care on the knowledge base.The generic rule pattern manager: Stores the generic change rule patterns and has to listen to knowledge base updates which may result in new instances of the patterns (any addition/removal of an entity may result in addition or deletion of a change rule).A domain rule pattern manager: Same as above, but dealing with domain dependent change rule patterns.Change manager: Manages the change dependencies. Changes in the knowledge base are notified by this agent and may cause the change rules to fire which in turn cause further knowledge in the knowledge base.
56 Example: ECA-Rules for Information Management (1) Suppose agent ag is responsible for action A.Information query rule pattern:Event: Action A has to executedCondition: A formula F (with variables) in the preconditions of AAction: Prolog-type query to the KM: For which values of the variables in F will f evaluate to true?QUERY(recipient, ag, F, list of variables, A)where recipient is an attribute which applies to agents or knowledge sources.Example: QUERY(storage manager, sales person, delivery time(prod), offer(prod)).
57 Example: ECA-Rules for Information Management (2) Suppose agent ag is responsible for action A.Pro-active information rule pattern:Event: Agent ag will now execute action A (this event is a result of the observation of ag by the KM)Conditions: A formula Improve(A, IU) (information unit IU improves performance of A); a formula notinformed(ag, IU) (agent ag does not have information IU)Action: NOTIFY(ag, A, IU, reason)Example: NOTIFY(ag, consulting customer C on product type P, special information on P, C has asked for details on P last time)Such rules are also used during a dialogue
58 Summary A general knowledge management with agents was described. Knowledge management and process models are strongly connectedKnowledge gaps are often difficult to discoverDifferent tasks of the KM were discussed, in particular change and information managementECA-rules were introduced as a general formalism for triggering actions.