Linger
(machine learning, evolutionary computing)

- Linger: Linger Is Neural & Genetic Email Reader

Explore the issues of
- Email filtering
- Feature selection
- Hybrid approach using Genetic Algorithms (GA) and Neural Networks (NN)

Context
- Many email clients offer automatic email filtering using simple keyword spotting.
- The rules have to be manually crafted by the user.
- The rules become out-of-date

Approach
- Hybrid approach to classification
- Soft computing techniques
- Since GA is good at searching through a large space of possible parameters, it is used to isolate which features can be used to distinguish between mail types.
- Optimal feature selection is to be found quickly based on the variance of each word across al mailboxes
- NN is good at adapting to input and generalizing for unseen data, hence, it is used to classify the unknown mail based on what features it contains

Mining Financial News
(machine learning, textual information)

Explore the issues of
- Financial market prediction using textual and numerical data

Context
- AOI (Australian Ordinary Index) from Australian Stock Exchange

Approach
- Finding the most significant 50 keywords
- An increase in the AOI gives a value between 0.5 and 1, and a decrease in the AOI a value between 0.5 and 0. The closer a value is to 0 (or 1), the further the AOI dropped (or rose).
- Induction tree, neural networks and textual analysis (based on probability based) are experimented.

Student Modeling
(education, genetic algorithms)

Explore the issues of
- Student modeling
- Parallel hypothesis of student’s problems using Genetic Algorithms (GA)
- To what extent can this approach be generalized to other teaching domains
- Individualized instruction

Context
- Mathematics tutoring system
- A Prolog system driving the design of questions

Approach
- GA is used to implement a way of reasoning similar to human reasoning.
- Initial population is instantiated using prototypes.
- The approach of this program is to evolve several different hypothetical student models simultaneously, and constantly discard erroneous hypotheses and generate new hypotheses.

Research Interests

Machine Learning - Concept drift - Reinforcement Learning - Coevolution of feature selection, text case generation and learning - Particularly interested in text-related information - ML for User Modeling

Textual Information - Information extraction - Text mining - Financial market prediction using text - Speech interface

Education
- Application of workflow technology
- Just-in-time feedback
- Student Modeling
- Ontology

Evolutionary Computing
- Hybrid approach with other techniques, e.g. neural networks, nearest neighbor, HMM
- Coevolution
- GENIE architecture – an artificial life approach to reinforcement learning