Extracting qualitative information from graphs

Software	Java or C++ or Smalltalk or another object oriented language with good GUI facilities
Covers	Artificial intelligence knowledge representation and uncertainty reasoning, possibly GUI development
Skills Required	Programming, interest in AI, preferably interest in engineering, not afraid of maths
Challenge	Conceptual ??? Technical ?? Programming ????

Brief Description

One way in which people express information diagrammatically is with data graphs, or rough sketches of data graphs. Simple two variable graphs show how the value of one variable varies according to the value of another variable; more complex data graphs show more complex relations. Sometimes people use sketched graphs to indicate how some quantitative dependent variable varies with different quantitative or qualitative values of some independent variable. The information in the sketch may be intended to show approximate quantitative relationships, or just qualitative relationships with rough indication of magnitudes.

In order to use AI techniques to reason about the characteristics (of some entity) depicted in a graph, we may want to extract qualitative and approximate quantitative relationships from a graph and represent them in symbolic form.

The objective of this project is to develop algorithms for identifying qualitative relationships in data graphs and constructing symbolic representations of them. It includes developing appropriate formulations of the input to the relationship extraction algorithms, and identifying or devising appropriate formalisms for the symbolic relationships.

Extensions

If the task is to interpret experimental or observational data, or to interpret sketched graphs, the input data will be messy. The project would gain from the inclusion of mechanisms for dealing with this.

While the project could be completed satisfactorily with handcrafted data sets, it would gain from an interface that allows users to draw lines on a data graph template on screen, to express qualitative and quantitative relationships.

Back to