AI – CS289 Knowledge Acquisition Knowledge Acquisition 11th September 2006 Dr Bogdan L. Vrusias b.vrusias@surrey.ac.uk AI – CS289 Knowledge Acquisition Contents • • • • Defining Knowledge Acquisition Interviewing for Knowledge Acquisition Case Study Terminology 11th September 2006 Bogdan L. Vrusias © 2006 2 AI – CS289 Knowledge Acquisition Knowledge Acquisition • Knowledge acquisition can be regarded as a method by which a knowledge engineer gathers information mainly from experts, but also from text books, technical manuals, research papers and other authoritative sources for ultimate translation into a knowledge base, understandable by both machines and humans. • The person undertaking the knowledge acquisition, the knowledge engineer, must convert the acquired knowledge into an electronic format that a computer program can use. 11th September 2006 Bogdan L. Vrusias © 2006 3 AI – CS289 Knowledge Acquisition The Process of Knowledge Acquisition • In the process of Knowledge Acquisition for an Expert System Project, the knowledge engineer basically performs four major tasks in sequence: – First, the engineer ensures that he or she understands the aims and objectives of the proposed expert system to get a feeling for the potential scope of the project. – Second, the engineer develops a working knowledge of the problem domain by mastering it's terminology by looking up technical dictionaries and terminology data bases. For this task the key sources of knowledge are identified: textbooks, papers, technical reports, manuals, codes of practice, users and domain experts. 11th September 2006 Bogdan L. Vrusias © 2006 4 AI – CS289 Knowledge Acquisition The Process of Knowledge Acquisition – Third, the knowledge engineer interacts with experts via meetings or interviews to acquire, verify and validate their knowledge. – Fourth, the knowledge engineer produces a "document knowledge base"; a document or group of documents (nowadays in electronic format) which form an intermediate stage in the translation of knowledge from source to computer program. This comprises: • the interview transcripts, • the analysis of the information they contain • and a full description of the major domain entities (e.g. tasks, rules and objects). 11th September 2006 Bogdan L. Vrusias © 2006 5 AI – CS289 Knowledge Acquisition Interviewing Techniques for KA • The Informal or Overview Interview – To familiarise the knowledge engineer with the domain and the particular problem which the proposed expert system is intended to solve. • The Focused Interview – Focused interviews are similar to ordinary "chat show" conversations or discussions where the interviewer is interested in a topic of which the interviewee is knowledgeable. – It is normally conducted by following a pre-determined agenda. The interviewee is initially prompted with the first topic or question, but is given a great deal of freedom of expression thereafter. 11th September 2006 Bogdan L. Vrusias © 2006 6 AI – CS289 Knowledge Acquisition Interviewing Techniques for KA • The Structured Interview – Structured interviews normally occur well into the knowledge acquisition phase. – They are used when information is required in much greater depth and detail than the other techniques can offer and is more interrogative than conversational. • 'Think aloud' Protocols – A technique used by cognitive psychologists to study the strategies with which people solve problems. Case studies are advantageous because the end results are already known so the expert should repeat the strategy he used for that problem when describing his solution. 11th September 2006 Bogdan L. Vrusias © 2006 7 AI – CS289 Knowledge Acquisition Interviewing Techniques – Do's and Don'ts • It is essential to record and transcribe all the (video/audio-taped) interviews. • Transcripts should be clearly cross-referenced to (video/audio-tape) recorder counter numbers. • Include all the sketches, photocopies or reproductions of diagrams, tables or the like, that were referred to during the interview(s). • Once completed a copy should be sent to the interviewee for comments, corrections and criticism. There is always the possibility of misunderstanding by the knowledge engineer when interpreting a statement or explanation. • By involving the expert in validating his or her own transcript it reduces the chance of erroneous information appearing in the prototype's knowledge base. 11th September 2006 Bogdan L. Vrusias © 2006 8 AI – CS289 Knowledge Acquisition Knowledge Acquisition Tasks Discovery Phase Learn Objectives Revision Phase Domain Terminology Revise Salient domain features Outline Scope of the problem Tec hnique Used Overview Interview Consult Tex tbook s Constrain Knowledge Sourc es Focussed Interviews Literature Review Problem-solving tas ks Verify Specify Domain objects Structured Interview Paper Knowledge Base Validate Produc e Rule Animation Rules and Heuristic s 11th September 2006 Bogdan L. Vrusias © 2006 9 AI – CS289 Knowledge Acquisition A Case Study in Interview-based KA • Project PLAIM (Platform Lifetime Assessment through Analysis, Inspection and Maintenance) was sponsored by the European Union during 1988-89. • The project had two major objectives: first to collate, analyze and archive the inspection and maintenance related data. And, the second aim is to establish a computer program which will: – allow access, and guide the user to the appropriate data (or data files); – provide an 'intelligent' interface to mathematical models, industrystandard simulation programs and empirical equations – acquire, formalise and disseminate the experiential (and previously undocumented) knowledge of inspecting and maintaining off-shore structures. 11th September 2006 Bogdan L. Vrusias © 2006 10 AI – CS289 Knowledge Acquisition A Case Study – The Interviews • A total of three knowledge elicitation interviews were conducted lasting over 5 hours and covering a broad range of topics relevant to the target problem: – The first interview provided the overview. – The second being much more focused on domain description and terminology. – The third interview was the only formally conducted, structured interview. • Regular prototype revision meetings were conducted in a similar interrogative style inspired by a demonstration of the prototype and review of the current knowledge base. • All but one of the interviews were recorded using a video-cassette recorder; all were transcribed and, where considered useful, the transcripts were sent to or discussed with the interviewee. 11th September 2006 Bogdan L. Vrusias © 2006 11 AI – CS289 Knowledge Acquisition A Case Study – The Interviews Interviewee Departmental Manager, AME Ltd Departmental Manager, AME Ltd. Senior Structural Engineer, UK Offshore Operator 11th September 2006 Subjects Covered Interview Technique Overview and explanation of idea behind PLAIM Overview Interview How an expert system is expected to fit in and what it was expected to do. (followed by structured interviews at prototype demonstrations) General introduction to terminology, design practice; Focused design for fatigue, classification of members, nodes, joint types, construction, practice, welding and fabrication. Think aloud Current inspection, repair and maintenance. Assessment of AME proposed approach to IRM; Opinion of where expert systems would be useful generally and specifically to the operator practice. Structured Bogdan L. Vrusias © 2006 12 AI – CS289 Knowledge Acquisition A Case Study – Overview Interview • The overview interview requires the preparation of a well targeted set of questions. The interviewee, the PLAIM project manager, was video taped and a transcript of his interview was produced. – The interview began with a discussion of a 'flow chart' for conducting fatigue analysis of offshore structures. • The interviewer, who already had access to a variety of contract documents related to PLAIM, asked the expert to explain the 'flow chart'. This led to a set of well focused questions such as: – Please outline algorithms, data input and output, data requirements. – What sort of knowledge and expertise is expected to be included in this prototype? – Please give your view on judgments on accuracy and calibration with real data? – How do you tell from residual strength and reliability index the lifetime of the structure or cracked joint? i.e. how long before the crack causes failure? – Please suggest further information sources. 11th September 2006 Bogdan L. Vrusias © 2006 13 AI – CS289 Knowledge Acquisition A Case Study – Overview Interview • The result of the overview interview led to the identification of the broad scope of the project and in cataloguing important technical documentation as textual knowledge sources. • The preparation of the questionnaire for the interview helped the knowledge engineer to learn much about the expert's impression of the problem and his understanding of how an expert system could be applied. • Some key phrases of the domain terminology were also introduced and explained. • A number of knowledge sources were identified by the domain experts ranging from research papers in learned journals to textbooks and repair and maintenance manuals. 11th September 2006 Bogdan L. Vrusias © 2006 14 AI – CS289 Knowledge Acquisition A Case Study – Focused Interview • The purpose of this interview was to cover two broad topics. Firstly, to describe a typical oil production platform and secondly to outline fatigue damage design, analysis and repair practices. The help of a second domain expert, who has hands-on experience of designing such structures, was enlisted. His reply comprised the following topics (The numbers on the right are video-recorder counters): 000 063 125 140 157 170 222 254 273 313 357 11th September 2006 Major Components of a Typical Platform (Figure 1) A Barge Launched Jacket (Figure 2) Fatigue Problem Areas Pile Sleeves Nodes Importance of Various Members in a Jacket Scour problems Anodes and Corrosion Protection Defects Fatigue Analysis: Procedure and Calculations Wave Data Bogdan L. Vrusias © 2006 15 AI – CS289 Knowledge Acquisition A Case Study – Focused Interview • 000 Major components of a Typical rig (Diagram 1) – The diagram shows the topside, consisting of the cellar deck to support the drilling rig, accommodation module, helideck etc. Also shown are the flare boom and other crane booms. The jacket supports the topside fixing it securely to the sea-bed above the level of the highest waves likely to be encountered in the North Sea. Piles are driven through guides in the legs of the jacket into the bed rock to ensure the rig position is solid. As the jacket structure is a group of frames made up of tubular steel sections and linked together by other frames, a method of identifying individual members and nodes at which groups of members coincide is required. The convention used on engineering drawings to identify the frame structure in plan view at each level or staging is shown below. This particular jacket was lifted into place using a crane. 11th September 2006 Bogdan L. Vrusias © 2006 16 AI – CS289 Knowledge Acquisition A Case Study – Focused Interview GRID SYSTEM Topside Rows 1 2 3 Cellar Deck B Sea level Jacket Fac es A Sea bed Piles Diagram 1. Major components of a Typical rig 11th September 2006 Bogdan L. Vrusias © 2006 17 AI – CS289 Knowledge Acquisition A Case Study – Focused Interview • 015 A Barge Launched Jacket (Figure 2) – The isometric view of the same jacket shows in more detail aspects of the frame structure the type of loading experienced and typical trouble spots. The increasing diameter of the leg is so that it is strong enough to be able to take the increasing axial load at the lower levels. When a wave hits the platform it causes an overturning moment which in turn causes an axial load in the leg. This is resisted by the piles, but in this example the eccentricity of the load due to the leg shape causes flexure in the short stubby diagonal braces and causes fatigue problems in their corresponding node joints. Other crossed-diagonal members also experience fatigue due to this sort of flexure but not to the same degree. 11th September 2006 Bogdan L. Vrusias © 2006 18 AI – CS289 Knowledge Acquisition A Case Study – Focused Interview conduc tor frame WAVE increas ing leg di ameter Short s tubby members Diagram 2. A Barge Launched Jacket 11th September 2006 Bogdan L. Vrusias © 2006 19 AI – CS289 Knowledge Acquisition A Case Study – Analysing the Data OBJECT ATTRIBUTES rig or platform topsides part of platform cellar deck part of platform jacket part of platform sea-bed piles part of jacket attached to (leg 1, 2, 3, 4) ... number of guides sleeve type member part of jacket part of level frame type of (leg, brace, diagonal, horizontal ...) 11th September 2006 Bogdan L. Vrusias © 2006 20 AI – CS289 Knowledge Acquisition A Case Study – Analysing the Data • rule 1 if: then: jacket is barge launched jacket will have extra structural members included purely for transportation and launching which become redundant once it is placed on the sea bed. • rule 2 if: then: a wave strikes the jacket the diagonal members will take the load/shear force. • rule 3 if: then: a jacket has sloping legs any crossed diagonal members at the lowest level will flex and cause fatigue in their corresponding node joints. • rule … 11th September 2006 Bogdan L. Vrusias © 2006 21 AI – CS289 Knowledge Acquisition A Case Study – Transcript Corrections • The interview transcript was sent to the expert for comments and criticism and was duly returned with corrections. It is not easy to classify the comments, except that the expert imposed constraints on his statements or expanded on others. Some examples below are presented to highlight the point we have just made. The amendments are shown in italics: • 000 Major components of a Typical rig – The diagram shows the topsides, consisting of the cellar deck to support the drilling rig, accommodation module, helideck etc. Also shown is the flare boom and other crane booms. The jacket supports the topsides fixing it securely to the sea-bed above the level of the highest waves likely to be encountered in the North Sea at the site. Piles are driven through guides attached to the legs of the jacket into the sea be to ensure the rig position is … 11th September 2006 Bogdan L. Vrusias © 2006 22 AI – CS289 Knowledge Acquisition A Case Study – Structured Interview • Based on the data collected from the previous two interviews a third interview was prepared and then planned with the Senior Structural Engineer UK Offshore Operator. • The main areas of the third interview was focussed on the following: – Current inspection, repair and maintenance. – Assessment of AME proposed approach to IRM. – Opinion of where expert systems would be useful generally and specifically to the operator practice. 11th September 2006 Bogdan L. Vrusias © 2006 23 AI – CS289 Knowledge Acquisition Specialist Languages • Characteristics of Specialist languages: – Considered variants of natural language are restricted lexically, syntactically and semantically. – Have a preponderance of open class words. – Have single and compound noun phrases (NP). These phrases are used to name objects, events, actions and states. – Have few adjectives and adverbs. 11th September 2006 Bogdan L. Vrusias © 2006 24 AI – CS289 Knowledge Acquisition Terminology • Systematically organised collection of terms and their elaborations, including definitions, grammatical categories, and related term. • The system used is usually a conceptual one. The conceptual basis is that of the discipline and its potential application. For example: – physicists organise their subject discipline in terms of forces, energy and mass; – chemists focus on atoms and molecules; – biologists organise their subject in terms of kingdoms, families and species. 11th September 2006 Bogdan L. Vrusias © 2006 25 AI – CS289 Knowledge Acquisition Terminology • Usually nouns, few verbs, adjectives or adverbs. • Terminology can be documented in: – paper form: terminological dictionaries; glossaries; thesauri; hierarchy diagrams or ontology. – electronic form: term-bases; hierarchies; ontology. 11th September 2006 Bogdan L. Vrusias © 2006 26 AI – CS289 Knowledge Acquisition Terminology and Knowledge • The terminology of a specialist domain, and to some extent the details of the problem-solving heuristics and that of the meta rules, reflect the underlying structure of the domain. • This structure allows the members of the domain community to develop new ideas, to challenge existing wisdom, to disseminate and to learn from each other. In effect, the underlying structure provides a cohesive framework for the domain community to function as a whole. 11th September 2006 Bogdan L. Vrusias © 2006 27 AI – CS289 Knowledge Acquisition Closing • • • • Questions??? Remarks??? Comments!!! Evaluation! 11th September 2006 Bogdan L. Vrusias © 2006 28