Institute for Web Science and Technologies · Universität Koblenz - Landau

Approximate Inference for Argumentation in AI

[go to master theses]

Open Master Thesis - Contact a supervisor for more details!


Argumentation in AI is an approach to common sense reasoning with an explicit focus on arguments and attacks between arguments [1]. Arguments typically consist of a set of premises and a conclusion, as well as a set of rules used to derive the conclusion from the premises. An argument A attacks an argument B if A is a counterargument for B. Reasoning in argumentation proceeds by constructing arguments and attacks on the basis of a knowledge base consisting of incomplete, inconsistent or uncertain information. Given these arguments and attacks, the next step is to determine which arguments are acceptable, which tells us which conclusions follow from the knowledge base.

The advent of "big data" requires methods to deal with ever growing amounts of information. However, algorithms in argumentation often do not scale well. One problem is that the number of arguments constructed on the basis of a knowledge base may grow exponentially with the number of rules in the knowledge base. Determining which arguments are acceptable can therefore be unacceptably expensive.

This master thesis topic deals with *approximate inference* for argumentation. It is often unnecessary to construct all arguments in order to answer a query. This means that we can obtain correct answers even if we construct only a fraction of the full set of arguments. The aim is then to find the best balance between performance and accuracy. Other approaches to speed up inference may also be considered.

We want to focus on particular on Assumption-Based Argumentation (or ABA, for short) [1]. ABA is one of the main structured argumentation systems an is closely related to Abstract Argumentation [2]. A full implementation of ABA is included in the Tweety library [3]. The main task of the thesis is to adapt this implementation to support approximate inference, and to evaluate the effectiveness of the approach using large real-world or randomly generated knowledge bases.

Literature
[1] Dung, Phan Minh, Robert A. Kowalski, and Francesca Toni. "Assumption-based argumentation." Argumentation in Artificial Intelligence. Springer, Boston, MA, 2009. 199-218.

[2] Dung, Phan Minh. "On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games." Artificial intelligence 77.2 (1995): 321-357.

[3] Thimm, Matthias. "The tweety library collection for logical aspects of artificial intelligence and knowledge representation." KI-Künstliche Intelligenz 31.1 (2017): 93-97.

Supervisors

  • staab@uni-koblenz.de
  • Professor
  • B 108
  • +49 261 287-2761
  • rienstra@uni-koblenz.de
  • Scientific Employee
  • B 218
  • +49 261 287-2779
  • thimm@uni-koblenz.de
  • Scientific Employee
  • B 112
  • +49 261 287-2715