Aristotle University Of Thessaloniki (AUTH)
Commonsense agent reasoning by exploiting knowledge graphs and computational argumentation
Abstract
dc:descriptionCommonsense agents are intelligent agents that use commonsense knowledge and commonsense reasoning in order to perform a task on behalf of or take a decision to help a human user, usually in open environments, which are dynamic, partially observable, and non-deterministic, where agents possess incomplete, imprecise, or even conflicting information. The aim of this thesis is to study commonsense agent reasoning and propose new models / methods for commonsense knowledge and commonsense reasoning. To do so, we (i) propose methods for commonsense agents to exploit commonsense knowledge from Semantic Web Knowledge Graphs (KGs), by allowing them to access domain-specific knowledge and use sophisticated reasoning techniques in order to answer commonsense questions, and (ii) propose new Argumentation Frameworks (AFs) that will allow commonsense agents to represent arguments with commonsense knowledge and provide the human users the reasons why a decision was made or a task was performed, by using commonsense knowledge, and therefore increase transparency and explainability.A Knowledge Graph is a knowledge base that uses a graph-structured data model or topology to integrate data. Knowledge graphs are often used to store interlinked descriptions of entities, objects, events, situations, or abstract concepts, while also encoding the semantics underlying the used terminology. Since the development of the Semantic Web, knowledge graphs are often associated with Linked Open Data projects, focusing on the connections between concepts and entities. A knowledge graph can include an ontology that allows both humans and machines to understand and reason about its contents and derive new knowledge.Argumentation theory, or argumentation, is the interdisciplinary study of how conclusions can be supported or undermined by premises through logical reasoning. Argumentation theory studies rules of inference, logic, and procedural rules in both artificial and real-world settings. Argumentation includes various forms of dialogue such as deliberation and negotiation which are concerned with collaborative decision-making procedures. Efforts have been made within the field of artificial intelligence to perform and analyze the act of argumentation with computers. Argumentation has been used to provide a proof-theoretic semantics for non-monotonic logic, starting with the influential work of Dung (1995).Computational argumentation systems have found particular application in domains where formal logic and classical decision theory fall short in capturing the richness of reasoning, domains such as law and medicine. Argumentation can formalise commonsense reasoning with empirical generalizations. In commonsense reasoning, people apply generalizations if nothing is known about exceptions, but they are prepared to retract a conclusion if further knowledge dictates that there is an exception. This kind of reasoning can be captured easily via argumentation.In the first part of the thesis, we present various novel methods we have developed for constructing commonsense KGs for cognitive agents that act in a household environment. These KGs are equipped with an information retrieval mechanism that allows any cognitive agent that uses theKGs as their knowledge base to answer sophisticated questions and achieve commonsense reasoning. Moreover, the information retrieval mechanism that accompanies the KGs uses external knowledge from SW KGs, such as ConceptNet, DBpedia, and WordNet, so that when it cannot answer a user’s questions, it will consult the external KGs and find an appropriate answer. Next, we propose a method on how the information noise found in KGs can be removed in order to provide commonsense knowledge more effectively and answer commonsense questions more accurately. Finally, we present a combination of KGs and learning through Argumentation, that allows a cognitive agent to use commonsense reasoning with the help of KGs in order toargue with a user over its decisions and, if the agent is proven wrong, to learn new knowledge from the user.In the second part of this thesis, we present various Argumentation Frameworks (AFs) that we proposed in order to represent and reason about commonsense knowledge than the already existing frameworks. We start by giving an initial notion of what can be considered a commonsense argument. We continue with an AF that classifies its attacks based on the application they can have over an argument, capturing the fact that not all attacks have the same affect in attacking or defending an argument, but rather there is a volume in their effect that cannot always be represented by a quantitative value. For instance, the testimony of an eyewitness who saw a crime in a dark room is not as strong as that of an eyewitness who saw a crime in daylight. Next, we present an AF where arguments have a domain and scope of application; with this framework, we capture the notion of context over the argument, meaning that an argument, though it might be correct, does not have an effect over some arguments because it isconsidered out of context. For example, an argument claiming that all objects will fall towards the center of the earth due to gravity, is context-dependent, as this argument is false on the international space station. Finally, we present a visualization tool for argumentation dialoguesthat displays the attacks, the defenders, and the acceptability of an argument.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Vasileiadis, Alexandros
- Βασιλειάδης, Αλέξανδρος
Subjects
dc:subject × 22- Γράφοι γνώσης
- Αναπαράσταση γνώσης
- Σημασιολογικός ιστός
- Εξωρυξη Δεδομενων
- Απεικονιση Δεδομενων
- Επιχειρηματολογία
- Συλλογιστικη Κοινης Λογικης
- Οντολογίες
- Knowledge Graph
- Knowledge representation
- Semantic web
- Information retrieval
- Data Mapping
- Argumentation
- Commonsense Reasoning
- Ontologies
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/54396