Knowledge Graph知识图谱—9. Knowledge Modeling

9. Knowledge Modeling & Ontology Engineering

Modeling knowledge within a knowledge graph (KG) involves addressing several key aspects: defining the types of entities that exist, establishing the connections and relationships between these entities, identifying the constraints and rules that govern their interactions, and ultimately formalizing these fundamental questions within the ontology framework of the knowledge graph.

Up to now, we have developed ontologies. The process began with - Read the specifications - Select an initial point randomly - Engage with Protégé to experiment - Learned through trial and error.

That was rather “Ontology Hacking” than “Ontology Engineering”

How to build ontologies?
Methodologies

What are the best practices for building good ontologies?
– Best Practice Solutions
– Design Pattern Instances
– Anti-Pattern Avoidance Methods
– Fundamental Ontology Structures

SECI模型
两类知识
Tacit知识 具有直觉性且难以形式化
例如：骑自行车、即兴音乐创作
明确的知识 是经过系统化的知识表示方法
例如：运动学、音乐理论
Tacit知识的产生通常依赖于明确知识的支持，并相互促进；知识创造的过程通常是合作性的.

SECI模式也被称为知识转换模式它阐述了组织内部知识创造与转换的机制 SECI由四个核心环节组成：社会化（Socialization）、外部化（Externalization）、结合（Combination）以及内部化（Internalization）。
其中社会化（Socialization）这一过程主要依赖于个体间的直接互动与共同经验分享来传递难以形式化的隐性知识 这种传递往往涉及技能、洞察力及直觉等无法言状的因素 社会化强调通过导师制度学徒制度及共同实践等社交活动来进行学习与积累。
在外部化（Externalization）环节个人所获得的隐性知识被系统地转化为可传达的概念与模型 这种转化通常采用叙述法隐喻法或类比法将难以形式化的思想转化为清晰可理解的形式 从而促进知识的有效传播与应用。
结合（Combination）环节则聚焦于对明确知识体系的构建 通过对现有明确知识进行分类整理及优化重组以期在此基础上产生新的见解与发现 这种过程常通过创建数据库手册或其他结构化的存储系统来实现高效的知识管理。
最后在内部化（Internalization）环节个人所掌握的明确知识被重新转化为自身的隐性认知体系 这一过程强调通过实践操作及经验积累使得理论成果得以内化成为个人能力的一部分 内部化的最终目标是实现理论与实践的有效循环往复以达到知行合一的效果。

9.1 Methodologies

Methodology（方法论）涵盖了研究领域内为解决各种问题提供指导的一系列原则性文件与规范性文件，并指导相关的操作流程与技术手段以促进研究发展与实施工作的发展。它是一种有序的结构模式与执行框架，在特定领域或活动场景中引导相关参与者遵循既定程序与步骤以实现系统的运行与管理。

9.1.1 Grüninger & Fox’s Methodology

den Grüninger & Foxschen Methode ist eine für das Ontologieren eingesetzte Frame-work, der besonderen Augenmerk auf die formale Darstellung und die-symbolische Be-representation von Domainwissen legt.

9.1.2 Methontology

Proceed incrementally from informal to increasingly formal ontologies.
Returning to a simpler state is permitted.
Along the process, documentation is generated.
A glossary contains terms with their definitions, synonyms, and antonyms.
A taxonomy outlines hierarchical relationships between classes.
Ad hoc binary relations are known as ObjectProperties.
A concept dictionary includes terms with their definitions, relationships, and instances (if applicable).

9.1.3 OntoClean

Methodology section outlines a comprehensive set of analysis techniques and evaluation criteria. The approach systematically addresses various aspects of system evaluation. The methodology is structured around several key principles. Rule 1 defines the concept of rigidity within the framework. The OntoClean framework distinguishes between rigid and non-rigid classes. An entity's classification as belonging to a rigid class is absolute; once an entity ceases to be part of that class, it is no longer considered part of it. This principle does not apply to non-rigid classes. For instance, in rigid categories we include entities such as Person (人), mountain (山), and company (公司). In contrast, non-rigid categories encompass more dynamic entities like Student (学生), stock company (上市公司), town (城镇), Caterpillar（ caterpillar ）and butterfly（蝴蝶）。

OntoClean 规则：刚性类别不应成为非刚性类别的子类别。

Typical rigidity issues commonly encountered in many domains

Rule2: Identity

Consider examining several instances.
– :1h a :Duration.
–: 2h a :Duration. …
–: Mo10-11 a :Interval.
–: Mo11-12 a :Interval. …
Clearly, the number of Interval instances exceeds those of Duration [contradiction].

What methods do we use to ascertain whether two entities are identical? Certain categories possess identity standards. • Student ID numbers for students. • Tax identification numbers issued to citizens and businesses. • Nationality codes used internationally. …

Observation: The identity criteria are of the two classes are different

The OntoClean rule stipulates that whenever p is a subclass of q, p must not include any identity criteria that are exclusive to q.

Rule 3: Unity

除了整体现象类之外的所有类别都只有「统一性类别」构成

OntoClean规则：Unity classes can only have unity classes as their subcategories, while anti-unity classes can only have anti-unity classes as their subcategories.
In our example:
– OrganicMatter is classified as an Anti-unity class
– Animal is a Unity class

Summarizing OntoClean

9.2 Ontology Design Patterns

Origin of the term 'pattern design'
Architecture
– Common issues
– Standard solutions that offer varying degrees of flexibility

Example
– Problem: Rain falling into the structure
– Solution: Roof is the solution
• Degrees of freedom: Shedding, dome, and hip styles are available…

Types of Ontology Design Patterns

• Presentational Patterns, for instance, naming conventions.
  • Logical Patterns:不受领域限制的逻辑模式；始终针对特定语言设计（如OWL DL）。
  • Content_patterns:依赖于领域但与语言无关的内容模式。
  • Transformation_patterns(例如，在如何将一个ontologue从一种语言转换到另一种语言方面)。

9.2.1 Presentation Patterns

9.2.2 Logical Patterns

9.2.3 Content Pattern

9.3 Anti-Patterns

Things that should not be done. But are frequently lead to issues. Possible causes – Neglecting the potential consequences – Insufficient or incorrect knowledge of RDF/OWL principles

Rampant Classism

How to distinguish classes and instances?

Exclusivity

Currently, what is taking place here? The ontology was specifically designed for a particular domain, such as cities. It tends to fail when applied in different contexts, like countries.

Core Principles of the Semantic Web include AAA, which stands for Anybody can assert Any statement in Any context. i.e., statements must function across diverse contexts.

Another example:
Every person is married to at most one other person

9.4 Classification of Ontologies

9.4.1 Top Level Ontologies

Top Level Ontologies (Very general)

• Domain independent

• Task independent
  Goal

• Reuse existing resources

• Ensure semantic precision

• Model guidance emphasizing the avoidance of poor models

• Compatibility or system interoperability

Aristotle’s Ontological Square
One of the oldest top level ontologies

Basic Categories for Top Level Ontologies
Abstract versus concrete entities
Abstract things lack both temporal and spatial dimensions, for instance, Numbers and Units of measurement.
Concrete things, on the other hand, inherently possess at least a temporal dimension, specifically a time span during which they exist; spatial dimension is an optional attribute. Examples include Objects like books and tables, as well as Events such as lectures and tournaments.

Between 3D and 4D perspectives
Under a 3D view, things extend into space. At every moment of time, they are fully present

4D view : Things exist within temporal and spatial dimensions. At any given moment, they may be present at the same time.

Actual vs. possible entities
– Actualism : only existing entities are included in an ontology
– Possibilism : all possible entities are included in an ontology

Co-location
多个实体是否可以在同一位置存在？
这似乎并不困难……
– 3D视图：否
– 4D视图：是的，但并非同时存在
但这并非易事

Example: 例如一件雕塑及其制作所使用的泥土量 Do statues even exist? –难道雕塑真的存在吗？ –或者在这种情况下是否只有泥土？ –如果两者都存在的话，则它们应该属于同一类别吗？

Further example: A sample of Swiss cheese with a defect. Do defects even exist? Or are there only perforated objects?

John Sowa's Top Level Ontology: A foundational framework
A more foundational top-level ontology from the nineteen nineties.
The twelve basic categories are derived from three key distinctions.

Physical vs. Abstract

Continuant: Things that constitute an entire entity at every moment of time
Occurent: Only partially constituting things at each moment of time; for instance, a lecture is an example of this concept

独立性、相对性和中介性对比
独立性：能够单独存在的事物
相对性：依赖其他事物存在的事物
中介性：介于两者之间的第三类事物

DOLCE
描述性本体论用于语言与认知工程
DOLCE是最具知名度的顶层本体之一

Particulars, universals, and quantities are fundamental concepts in categorization. The concept of universals is that they represent categories which can have specific examples. For instance, "City" and "University" are classic examples of universal concepts. On the other hand, the concept of particularness refers to individual entities that cannot be further divided into subcategories. Examples include "Mannheim" as a city name and "Mannheim University" as an educational institution. Qualities describe specific attributes or properties that pertain to individual instances. For example, the color of a book or the height of a person illustrates qualities. Neither particulars nor universals / qualities are applicable to abstract concepts / nor can they exist without reference to specific instances.

The highest-level ontology for specific instances
For existing and potential entities (possibilistic perspective)

4D: Some entities may have a temporal dimension

Co-location
被允许的
限制：在同一时间和空间位置上不允许有两个同一类别的实体
Not：两个雕塑 But：一个雕塑和一定量的粘土

Top Hierarchy of DOLCE

对比：Endurants与Perdurants之间的区别在于它们存在于时间中的不同状态。思考：诸如人、书籍等具体事物……此外，也可能是非物理性的：如组织、信息片段等。在其存在期间的每个时间点上始终完整存在。

Perdurants occur in time.
Think: events and processes.
Only existing partially at each moment during their existence, such as past and future portions of the perdurant, might not (yet or anymore) exist at a particular moment.

Qualities are attached to endurants and perdurants

Abstracts : numbers, units of measure, etc.

Engaged parties participate in enduring processes
– Active parties include readers who engage in continuous activity.
– Passive parties such as book readers focus on passive engagement.
– DOLCE systematically defines various types of participation.

Endurants仅由endurants构成，perdurants仅由perdurants构成
– 书籍由pages、covers等构成
– 阅读涉及perceiving、turning pages等行为

Endurants in DOLCE

区分保持者
量与物理物体的区别
**量（Amount）**表现出"物相不变性"（mereological invariance），即：

• 量的部分（subportion）仍然是一个量（AoM）；例如：将"一些水"分成两份，则每份仍是"一些水"；
• 但容器等物理物体（Physical Object）的部分则未必是同一类物体；例如：一个杯子的组成部分可能不再是完整的杯子；
• 参考物象与非物象概念在OntoClean中的统一性与反统一性（cf. unity/anti-unity in OntoClean）

Features
Cannot exist without a physical endurant , e.g., holes, fringes

Perdurants in DOLCE

Distinguishing Perdurants

Qualities
Fundamental differentiation
The quality represents a characteristic inherent to each entity
The quality space constitutes all potential value configurations that a quality can assume.

qualities depend on entities. Generally speaking, any particular may possess qualities. Qualities exclusively exist when the entity exists.

9.4.2 Other Top Level Ontologies

SUMO: Proposed Integrated Ontology - Approximately 1,543 classes - Robust formalization using the Knowledge Interchange Format (KIF).
Cyc: Originated from Encyclopedia Cybernetica - Based on its own language (CycL) - Comprising a Top Level ontology as well as a deep general ontology - The OpenCyc project is no longer available but remains accessible for reference.
PROTO: PROTo ONtology - General Top Level plus Upper Levels, offering diverse domain-specific extensions - Providing approximately 354 classes and about 678 relationships.