Chapter 4. Knowledge Support based on Case-based Reasoning and Data Mining
4.3. Knowledge support for problem-solving
4.3.2. Knowledge recommendation
The problem-solving process module employs CBR to identify the current situation or retrieve similar situation-cases according to the similarity measures. The knowledge rec-ommendation module then suggests relevant documents according to the situation profile of the current situation or similar cases, as shown in Fig. 7. The system also recommends relevant action documents (e.g., operating procedures and guidelines) according to the ac-tion profile. Note that the top-N relevant documents are recommended according to the co-sine measure of the term vectors of the documents and the situation/action profiles, as de-scribed in Section 4.2.
Moreover, the system suggests possible actions for handling the current situation ac-cording to the decision-making knowledge patterns. Note that the actions in the deci-sion-making patterns (i.e., situation => action) whose left-hand side match the current situation are suggested and ranked according to the confidence values of the rules. De-pendency knowledge patterns are also suggested to help workers predict a possible chain reaction across different stages and develop appropriate action plans.
Fig. 7: Situation profile and relevant documents
4.4. System implementation
We developed a prototype system to demonstrate the effectiveness of the proposed knowledge support system for problem-solving. The implementation is conducted using several software tools, including the Java(TM) 2 Platform Standard Edition Runtime En-vironment Version 5.0, Java Server Page, and Macromedia Dreamweaver MX. A web and application server is setup on Apache Tomcat 5.5.7, and Microsoft SQL Server 2000 is used as the database system for storing data related to the problem-solving process and codified knowledge documents. The data mining tool Weka 3.4 is used to discover knowledge pat-terns in the historical problem-solving log.
The generic problem-solving process, situation/action profiles, decision-making and dependency knowledge patterns form the knowledge support network. The network pro-vides relevant knowledge documents, and suggests decision-making and dependency knowledge patterns. The problem-solving knowledge support system is integrated with the knowledge support network to provide more effective knowledge support for browsing problem-solving knowledge patterns. The interface of the problem-solving knowledge support system includes the system frames for user login, search engine, and user-guide. A worker Annie logs into the system and gets a problem list. Once she selects a generic problem-solving process to browse, the problem (e.g., water supply problem) can be browsed further in the system platform, as shown in Fig. 8.
Fig. 8: A generic water supply problem-solving process
Annie can choose a situation/action to get knowledge support. Fig. 9 shows an example where Annie chooses the situation “Controller Temperature abnormal issue” in the normal management stage of the water supply problem-solving process. The system presents the decision-making knowledge patterns: “Controller Temperature abnormal situation → Monitoring the output action” in the knowledge support network. The relevant documents for the situation “Controller Temperature abnormal issue” are shown below the page. The system also displays the key terms of the profile for the situation, including Controller Temperature Abnormal and Controller operation status. The key terms give workers an overview of the current situation. By reading the relevant knowledge documents, AF0001C0F25 and A9600400762, Annie can understand the situation, identify its causes, and take appropriate action. Moreover, the suggested dependency knowledge pattern can help Annie realize a possible chain reaction across different stages. Accordingly, workers can develop appropriate action plans across different problem-solving stages.
Fig. 9: Decision-making knowledge patterns for the water supply problem
Chapter 5. Knowledge Support based on Case-based Reasoning, Data Mining, and Rule Inference
In this chapter, we describe the knowledge support based on CBR, data mining, and rule inference techniques, including context-based knowledge support framework for problem-solving, discovery of context-based problem-solving knowledge, the prototype system implementation, discussions, and comparisons.
5.1. Context-based knowledge support framework for problem-solving
The adapted system framework comprises a problem-solving process, context-based inference rule discovery, context-based situation profile discovery, context-based knowl-edge pattern discovery, and knowlknowl-edge recommendation modules, as illustrated in Fig. 10.Problem-solving module Context-based inference rule discovery module Historical log of intranet portal Run-time information collection
Context feature modeling and processing for knowledge discovery
Retrieval of situaiton/action relevant documents
Knowledge document recommentation
Context-based inference rule discovery
Data processing for knowledge discovery
Historical log of intranet portal
Context-based situation profile discovery
Inferred situation/action feature collection
Context-based inference rules
Historical log of intranet portal Context-based knowledge patterns and
inferred knowledge Data processing for knowledge
discovery
Context-based knowledge pattern discovery
Context-based knowledge pattern discovery module Context-based situation profiles Context-based knowledge patterns
Fig. 10: The adapted system framework of context-based knowledge support
Problem-solving module. This module gathers production line run-time information, such as problem situation features. Collected and inferred situation/action features help CBR to retrieve similar situation/action cases. Workers can then execute a specific problem-solving process and obtain relevant knowledge documents from the knowledge recommendation module. The problem-solving steps, including the situations, actions, and corresponding knowledge documents, are recorded in the historical log. This is described in Section 5.2.1.
Context-based inference rule discovery module. This module gets situation description, attributes, features and relevant context information from problem-solving process module.
The framework uses constraint-based association rule mining to discover the association between situation features and relevant context characteristics. The discovered con-text-based inference rules are used to infer more relevant problem features in order to assist CBR identify problem situation encountered. The module is described in Section 5.2.1.
Context-based situation profile discovery module. This module analyzes the historical log file to discover context-based situation profiles. For specific situations/actions in certain context, relevant information (documents) accessed by workers is recorded in the prob-lem-solving log. Historical codified knowledge (textual documents) can also provide valuable knowledge for solving the target problem. Information Retrieval (Automatic In-dexing) techniques are used to extract the key terms of relevant documents of a specific situation for certain context. The extracted key terms form the context-based situation pro-file, which is used to model the information needs of the workers in certain context. The knowledge support system then uses the profile to gather relevant information and help workers solve the target problem. Further details are presented in Section 5.2.2.
Context-based knowledge pattern discovery module. This module discovers context-based knowledge patterns for situation in certain context. Based on collected attributes, context characteristics, and relevant context-based inference rules, the system continually infers relevant situation features and actions as clues to form the context-based knowledge patterns, including decision-making and dependency knowledge. We design a scoring mechanism to represent the action importance of discovered context-based decision-making knowledge patterns. The context-based knowledge pattern with its score helps workers take reasonable actions in certain context of problem-solving process. The details of this module are de-scribed in Section 5.2.3.
Knowledge recommendation module. This module recommends context-based knowledge patterns, inferred knowledge, and relevant situation documents as context-based knowledge support. The context-based knowledge patterns and inferred knowledge assist workers take appropriate actions for solving situation or realize dependency of situations/actions in cer-tain context. As noted previously, the context-based situation profiles are used to gather existing and new relevant knowledge documents of a specific situation for certain context.
The relevant documents provide practical knowledge support to help workers solve prob-lems. Further details are presented in Section 5.2.4.
5.2. Discovery of context-based problem-solving knowledge
This section describes the procedure of discovering context-based knowledge from historical problem-solving logs. To illustrate the proposed approach, we use data from the log file of a semiconductor foundry’s intranet portal, which contains the problem-solving log for handling problems on the production line. The company operates wafer manufacturing fabs to provide the industry with leading-edge foundry services. The log file records the encountered situations and actions taken. The system also contains profiles accessed by workers for each situation of different context during the problem-solving process. The data fields of the problem-solving log include the situation features and context information.
5.2.1. Context-based situation identification and case-based reasoning
Each situation or action is a case that is characterized by a text description, situation features and a set of attribute values. The attribute values provide important information such as the symptoms of a situation to identify the situation case. Situation features are analyzed from previous problem situations/actions and can be predefined in system. Such situation features may be collected in run-time by the system or selected by the user. For undefined situation causes, users need to provide a text description of the situation. The text description can be used to extract identifying terms for the situation. Moreover, situation features collected by the system are usually partial and incomplete. Context-based inference can be initiated to infer more situation features. The text descriptions, situation features, attribute values contribute to similarity matching and situation identification. For the target situation/action, namely, the case workers are currently handling, the system identifies an existing case identifier or retrieves similar cases based on CBR.Extraction of identifying term vectors. The data stored in the Subject field of an existing
case is a text description of the situation. For example, Subject: “FAB8D Cu-BSC DI Water flow capacity insufficient issue” is the description of the situation - insufficient water flow capacity. The terms extracted from the subject field are used to identify the situation and attributes, e.g., situation name: insufficient water flow capacity; factory name: FAB8; de-partment identification: D; system type: DI; system status: water flow capacity insufficient.
The relevant context entity and feature include staff: Annie; role: DG; time: 20040502-PM;
location: Hsinchu, service name: DI water supply service, etc. Note that the terms are ex-tracted using term transformation steps, including case folding, stemming, and stop word removal. We simply extract the terms without considering the term frequency, since the subject field generally contains a short text description. The extracted terms form identifying terms to identify a situation case. Moreover, the user needs to provide a text description for the target case, namely, the situation or action which he/she is handling. Similarly, the identifying terms of the target case are extracted from the text description using the term transformation steps. Let Tj be the set of identifying terms extracted from the subject field of a situation case Cj. An identifying term vector Crj
is created to represent Cj. The weight of a term ti inCrj is defined by Equation 5. Equation 6 defines the similarity value simT
(Ck, Cj) of two situation cases Ck and Cj based on their text descriptions. The similarity value is derived by computing the cosine value of the identifying term vectors of Ck and Cj.
Similarity value by attribute. An attribute value may be nominal, binary, or numeric. For numeric attributes, a data discretization process is conducted to transform their values into value ranges or user-defined concept terms (such as low, middle or high). Equation 7 defines the similarity value simA(Ck (attrbx), Cj (attrbx)) of two situation cases Ck and Cj, derived according to their values of attribute x; value(Ck (attrbx)) denotes the transformed value of attribute x of Ck , which is calculated by the discretization process.
Context modeling. The context information is any information about an entity status. An entity can be the user, physical location, service, or service relevant object, etc. Due to the variety of context information, it is not easy to represent the complete context information of an entity. Therefore, based on the problem-solving environment, this work uses a modeling mechanism which composes with three levels to formalize the context information including Context entity level, Context feature level and Context association level.
z Context entity level. This level represents the conceptual abstraction of context entity spread in a problem-solving environment, includes physical, organization, process, staff, service, and document entities, etc.
z Context feature level. The context feature may be predefined by a domain expert that shows relevant information of a specific entity. A context entity may include one or more context features, for example, a physical entity covers the identification, time and location features; an organization entity may include the factory and department fea-tures; a process entity contains stage, task, and status feafea-tures; a staff entity has user, role, degree, and activity features; a service entity may involve with system, component, and parameter features; a document entity includes original, type, author, and score features, etc.
z Context association level. This level defines the association relationship between relevant features and attributes of the context entities. The association relationship is used to collect more relevant information of current problem-solving process based on context characteristics. We list some pre-defined association types as follows.
¾ The organization-staff association describes the relationship between organiza-tion and staff entity, e.g., Annie belongs to DG role in B department of Fab8 factory.
¾ The staff-process association describes the relationship that user-role carries out the specific process, e.g., DG-Annie carries out the water supply problem-solving process.
¾ The staff-service association describes the relationship that user-role uses the specific system service, e.g., DG-Annie uses the DI water supply system service.
¾ The process-service association shows the relationship between the process and service entity, e.g., the water supply process contains the DI water supply and pipe control system services.
¾ The process-document association describes the relationship that some documents support specific process, e.g., expert or experiential reports of specific situation.
¾ The service-document association shows the relationship that some documents belong to specific service, e.g., user guide or technical documents of specific system service.
Based on context modeling, the system proactively collects the relevant context entities and features of current situation. For example, when staff Annie suffers from the controller temperature abnormal situation, the relevant entities include physical time, location, or-ganization, Annie, water supply problem-solving process, DI water supply system, and relevant knowledge documents, etc. The system also gathers relevant features of context entities in a controller temperature abnormal situation, such as physical time: 20040502-PM 3:24; location: Hsinchu; factory: Fab8; department: B; user-role: DG- Annie; process: water supply problem-solving process; stage: normal management stage; situation: controller temperature abnormal situation; service: DI water supply system service; document:
AF0001C0F25; author: PTC; Score: 4; original: DIFF knowledge base, etc. The collected context entities and features of specific situation are stored in enterprise knowledge base for context-based inference rule discovery. Context entities and situation/action features are represented in some meta-rule format predefined by expert. The proposed system enforces the constraint-based association rule mining to discover the context-based inference rules from the problem-solving log.
Context-based inference rule mining. The context-based inference rules discovered from association rule mining represent the associations of situation features and context charac-teristics. The rule format is shown as Equation 10:
[featurep … and contextq …] → [featurer] [Support = s%, Confidence = c%] (10)
For example, for the controller temperature abnormal situation, the features of staff entity:
“Annie” and service entity: “DI water supply system service” are associated with the feature of DI water supply system service entity: “Parameter incorrect”. The context-based infer-ence rule is shown as follows.
[Staff(Annie) and DI water supply system service()] →
[DI water supply system service(Parameter: incorrect)] [Support =2%, Confidence =13%]
For specific situation, the collected context entities and features are used to discover relevant actions. The format of context-based inference rule is represented as Equation 11:
[featurep … and contextq …] → [Actionr] [Support = s%, Confidence = c%] (11) For example, for the controller temperature abnormal situation, the features of staff entity:
“Annie” and service entity: “DI water supply system service” are associated with the Action:
“Reporting the outcome”. The context-based inference rule is shown as follows.
[Staff(Annie) and DI water supply system service()] → [Reporting the outcome action()]
[Support =2%, Confidence =13%]
For specific problem-solving process, the collected context entities and features of specific situation are used to discover relevant situation features. Equation 12 shows the format of context-based inference rule that infers relevant action feature of specific situation; the format of context-based inference rule that infers relevant situation features of specific ac-tion is represented as Equaac-tion 13:
[featurep… and contextq …]Si *…and [featureu … and contextv …]Aj* →
[featurer] Ak [Support = s%, Confidence = c%] (12)
[featurep … and contextq …]Si *…and [featureu … and contextv …]Aj * →
[featurer] Sk [Support = s%, Confidence = c%]
(13)
The examples are illustrated as follows. The feature of context entity Staff: “Annie” in controller temperature abnormal situation of Normal Management stage and the feature of context entity Staff: “PTC” in consulting with the expert action of Engineering Improvement stage are associated with the feature of DI water supply system service entity: “Parameter:
increasing pressure” in modifying the configuration action of Exception Management stage.
The context-based inference rule is shown as follows.
[Staff(Annie)]NM_S7 and [Staff(PTC)]EI_A2 →
[DI water supply system service(Parameter: increasing water pressure )]EM_A1
[Support = 1%, Confidence = 14%]
The feature of context entity DI water supply system service: “Parameter: output value” in monitoring the output action of Normal Management stage and the feature of context entity Document: “A9600400762” in testing based on the SOP action of Engineering Improvement stage are associated with the feature of DI water supply system service entity: “Parameter:
water quantity” in supply quantity abnormal situation of Exception Management stage. The context-based inference rule is shown as follows.
[DI water supply system service(Parameter: output value)]NM_A5 and [Document(A9600400762)]EI_A1 →
[DI water supply system service(Parameter: water quantity )]EM_S2
[Support = 3%, Confidence = 11%]
Certainty Factor value of context-based inference rule. The certainty degree of system collected situation feature is set to 1. For inferred situation features, this work employs the method of Certainty Factor (CF) value (Shortliffe et al., 1975) to derive the certainty degree during the inference, as defined in Equation 3. The preceding set denotes run-time situation features and context characteristics; the succeeding set is the situation feature that we want to infer its certainty degree. For example, The CF value of the context-based inference rule:
[Staff(Annie)] → [DI water supply system service(Parameter: incorrect)] is 0.033. The de-tails of calculation are shown as follows.
[Staff(Annie)] → [DI water supply system service(Parameter: incorrect)]
[Support = 2%, Confidence = 13%]
S([DI water supply system service(Parameter: incorrect)]) = 10%
CF([Staff(Annie)] → [DI water supply system service(Parameter: incorrect)])
= (13%-10%)/(1-10%) =0.033
Inference for situation features. Based on the CF value of situation feature and con-text-based inference rule, the inference process follows the rules defined in Equation 4. An example is illustrated in Fig. 11. The details of inference process are shown as follows. The context-based inference rule: [Role(DG)] → [Staff(Annie)] indicates the feature: DG of context entity: Role inferring the feature: Annie of context entity: Staff. Its CF value is 0.7.
two context entities: [Staff(Annie)] and [DI water supply system service ()] have “AND”
relationship. Its output CF value is 0.5. The CF value of [Staff(Annie) and DI water supply system service ()] → [DI water supply system parameter(Incorrect)] is 0.3. The CF value of [Pipe system service()] → [DI water supply system parameter(Incorrect)] is 0.2. Finally, there is a “JOIN” relationship with two inference conditions. The CF value of [Staff(Annie) and DI water supply system service ()] → [DI water supply system parameter(Incorrect)], [Pipe system service()] → [DI water supply system parameter(Incorrect)] is 0.3. Inferred situation features with high ranking of CF value are considered as the Inferred knowledge to assist CBR in identifying situation encountered.
Fig. 11: An example of inference process
CF(Annie) = CF(DG) * CF(IF DG THEN Annie) = 1.0 * 0.7 = 0.7 CF(DI)= CF(Service) * CF(IF Service THEN DI) = 1.0 * 0.5 = 0.5 CF(Annie DI) = MIN(CF(Annie), CF(DI)) = 0.5 ∧
∧ DI) *CF(IF Annie ∧DI THEN Incorrect), CF(Incorrect) = MAX(CF(Annie
CF(Pipe) * CF(IF Pipe THEN Incorrect) ) = MAX(0.5 * 0.6, 1.0 * 0.2) = 0.3
Inferred situation features with high ranking of CF value are considered as the Inferred knowledge. Then the inferred knowledge assists CBR in situation identification. Let Fj be the set of situation features of Cj that are collected by the system or inferred by the con-text-based inference rules. A feature vector CrFj
is created to represent Cj. The weight of a feature fi inCrFj
is defined by Equation 14.
⎩⎨ cases Ck and Cj based on their situation features. The similarity value is derived by com-puting the cosine value of the feature vectors of Ck and Cj.
⎩⎨ cases Ck and Cj based on their situation features. The similarity value is derived by com-puting the cosine value of the feature vectors of Ck and Cj.