Learning Outcomes

Knowledge Representation and Reasoning

Maria Ingold
AI generated with Bing
  1. Critique the need for formal approaches to knowledge representation and reasoning.
  2. Review critically properties of a knowledge-based system.
  3. Appraise critically modelling techniques for knowledge representation and reasoning.
  4. Examine and incorporate different modelling approaches to solving KRR problems.

Artefacts

Collaborative Discussion 1

Maria Ingold

Topic

Review Weststeijn's 2011 paper on hieroglyphs to universal characters. Is Knowledge Representation a recent phenomenon? And how useful is it without reasoning support?

Learning Outcomes

  • Outcome 1. Simply having symbols like Ancient Egyptian hieroglyphs or Chinese characters is not enough. The symbols need to be interpretable. By utilising the Rosetta Stone to translate from Ancient Greek, a formal language structure was reinvented to read the Ancient Egyptian symbols.
  • Outcome 2. Knowledge is a relationship between a proposition and a knower. Representation is a symbol that stands for something else. Reasoning is the manipulation of symbols to derive new knowledge.
  • Outcome 3. In this case, the map between the symbolic representation and reasoning was missing for Ancient Egyptian because it had been systematically destroyed in the 5th century, so a new model had to be created that interpreted both ideography and consonant-based alphabet.

Feedback

Ungraded. Peer response: "This post is insightful and interesting! I never thought about how we often interpret unknown symbols as "rituals" due to a lack of understanding." Concurs that knowledge representation is not a recent phenomenon tied to computing. Felt my Rosetta Stone example was a good illustration of reasoning based on knowledge of other languages to interpret a third. Raised a good example of libraries, storing knowledge without actively reasoning, and that data analysis uncovers unknown patterns, even without explicit reasoning. I found these insights useful and valuable, and they have broadened my understanding.

Artefacts

Collaborative Discussion 2

Maria Ingold

Topic

Based on Kalibatiene & Vasilecas (2011) posit that “An ontology is a formal, explicit specification of a shared conceptualization”, which language do you believe is the most useful to express ontologies that can be utilised by software agents on the WWW: KIF, OWL2, RDF or OWL-lite?

Learning Outcomes

  • Outcome 1. KIF, RDF, OWL Lite and OWL 2 present different formal approaches, each solving different problems. Each has different degrees of support for the formal, explicit, shared and conceptualisation aspects of an ontology. All but KIF are designed for the semantic web.
  • Outcome 3. KIF predates the WWW, RDF is designed for web and OWL 1 and 2 are built on RDF. OWL Lite was designed to be lightweight but was not carried forward due to computational complexity, and OWL 2 addresses issues but is itself complex.

Feedback

Ungraded. Peer response: "You have great points!" Reminded that OWL 2 might be overkill for simpler tasks. Concluded with a great point that "As the agent evolves, its language needs might too!" In general, we were in complete agreement, however, he believed OWL Lite was a bridge, while I believe it was left out of OWL 2 for a reason, including potential exponential complexity.

Artefacts

Case Study Review

Maria Ingold

Topic

Review Malik et al. (2015) report on ontology development for the agriculture domain. Identify two application areas within agriculture where the approaches can be applied. Discuss the business context and need for ontology in the application areas and how the approaches will be adapted for the other application areas.

Learning Outcomes

  • Outcome 1. Analysed the Noy and McGuinness methodology used, compared three applications of it and contrasted it with the Gruninger and Fox approach. Gruninger and Fox has more formal competency questions.
  • Outcome 2. Critiqued the requirement for formal competency questions, validation, evaluation, and evolution in the feasible delivery of an ontology.
  • Outcome 3. Critically reviewed two application approaches in agriculture--fertiliser and cash crop--which both used Noy and McGuinness. The fertiliser example built on the original paper, which created an overly broad agriculture ontology.
  • Outcome 4. Fertiliser example included reasoning with Pellet, validation with OOPS!, and competency question checks with SPARQL. Cash crop contrasted with a more robust and rigorous Gruninger and Fox approach and field trial validation used for Sri Lankan farmers.

Tutor Feedback

80% (Distinction). Described as "excellent review" with "ample evidence" for understanding an ontology and "comprehensive evaluation of development approaches". Requested further assessment and analysis of limitations and challenges.

Artefacts

Modelling Assignment

Maria Ingold

Topic

Local Community Library Ontology (LoCLOnt) is a prototype Protégé OWL 2 ontology for library books that introduces a novel way for patrons to find books that interest them at their reading level. Ontologies provide a backbone enabling Artificial Intelligence (AI) semantic search.

Learning Outcomes

  • Outcome 1. Online Public Access Catalogue (OPAC) systems provide syntactic search without context, while ontologies enable semantic search using concepts and relationships. These can be validated with First Order Logic (FOL) competency questions.
  • Outcome 2. Common library systems, like Spydus, are OPACs that use MAchine-Readable Cataloging (MARC 21) and a relational database. OPACs require exact matches to find search terms. "God" did not find a search for the book "American Gods". LoCLOnt uses an ontology, which is more flexible and has extensible classes (concepts). It finds "American Gods" from "God" and can be used for more complex queries like "Which book titles are suitable for Middle Grade children who like fantasy".
  • Outcome 3. Reasoning with Pellet is helpful to maintain validity. Noy and McGuinness is a straight-forward easy to follow approach which seems suitable for a prototype. However, they do not provide a formal method for competency questions. Gruninger and Fox use FOL to validate competency questions. This approach has been used here and was effective in creating questions that translated well into SPARQL queries.
  • Outcome 4. LoCLOnt followed the seven-step Noy and McGuinness methodology and used Grüninger and Fox competency questions, validated by FOL. Classes, properties and property restrictions enabled representation of concepts, relationships, and requirements. LoCLOnt evaluated seven semantic relationships using competency questions. While only in prototypes stage, competency questions showed semantic search. The ontology can be easily extended from prototype but reuse of existing ontologies seems difficult.

Tutor Feedback

90% (Distinction). Overall: "Overall this work is outstanding." Also: "Excellent explanation of context." Delved "very deep into the subject by examining the Spydus Library System and how ontology designed AI driven model could enhance its information retrieval system and capabilities."

Formative Activities

Ungraded formative activities / homework per unit.

Maria Ingold

Learning Outcome

  • Unit 1: Knowledge versus Information - Outcome 1
  • Unit 2: Sets, Set Theory, Truth Tables and Logic - Outcomes 1, 2
  • Unit 3: Reasoning from First Order Logic - Outcome 1, 2
  • Unit 4: Logic Programming - Outcomes 1, 2
  • Unit 5: Modelling - Outcomes 1, 2, 3
  • Unit 6: Data Acquisition Methods - Outcomes 1, 2, 3
  • Unit 8: Ontology Development - Outcome 4
  • Pizza Ontology: Protégé OWL 2 ontology - Outcome 4

Reflections

Critical reflection on learning.

Maria Ingold

SWOT Analysis

SWOT

Tutor Feedback

90% (Distinction). "Excellent analysis and evaluation of the challenges. Developed detailed understanding of a range of challenges and issues influenced by the course. Excellent evidence thinking and reasoning on the subject matter as a whole. This is supported by the consistent manner which you have examined ontology design and modelling."

Reflection available upon request.