Introduction 

Recently a number of LKL members have raised questions regarding the existing techniques in Artificial Intelligence in Education that are used to learn from data in order to either support their research or to directly allow them to design data-informed technology enhanced learning.   In a continuing effort to consolidate the knowledge existing at the LKL and to ensure that LKL’s research is supported, where appropriate, by the state of the art in the field of AIEd and computer science, we would like to offer a series of tutorials in the area of data mining and machine learning.  

These tutorials are intended as an introduction to the basic techniques, with the intended outcome including the ability of the interested parties to make informed decisions about the possible techniques that they may want to use for their research and to be informed about the resources available. 

The tutorials will last 1.5 hours and will consist of an introduction to the topic, explanation of the basic technique and whenever possible, a hands-on experience, using common tools and applying the techniques on data.  

Below is a brief summary of the topic and the basic techniques that we intend to cover during the tutorials.  We would appreciate if you could read through the summary and indicate what would you like to gain from the tutorials and which of those techniques you might be interested to understand the most (click here to contact us through a form).

This will help us in structuring the tutorials according to your needs.  We would appreciate a prompt response as we are in the process of preparing our first tutorial – to take place on the last Thursday of every month starting in January 2009. 

Many thanks 

Kaska and Manolis

(contact us) 

AIEd Tutorials Summary 

Machine learning and Data Mining are two closely related fields in Artificial Intelligence and Computer Science that are (broadly speaking) concerned with the automatic discovery of potentially useful information from data by computational and statistical methods. The boundaries of the two fields are blurred and the distinction is mostly philosophical (something which will not concern us during these tutorials). However, one of the main distinctions is that data mining deals mostly with descriptive models and patterns from existing (usually large) datasets, while machine learning focuses on deriving models that can allow future predictions.  

Machine learning and data mining techniques have been applied in a several fields from industry to research. In recent years, there has been an increased interest in their ability to contribute to computational modelling of learners’ cognitive, behavioural and affective characteristics in order to (1) better support learners through intelligent educational software and (2) to facilitate investigation of scientific questions within educational research in order to understand how learners learn either in traditional or technology- enhanced settings. 

Employing these techniques in the educational context requires an interdisciplinary approach that relies on input from educational theory, cognitive science, psychology, artificial intelligence and computer science. For example, a model can be developed via knowledge elicitation approaches (i.e. reasoning based on data or available theories) and then used to inform (or validate) the computational techniques employed. 

The specific techniques in question fall broadly into three general categories (although combinations of these techniques are also possible).

1. Prediction
   
   The goal of prediction methods is to develop a model that can infer a characteristic of the data (predicted variables) from some combination of other elements of the data (predictor variables). There are several methods for prediction and their effectiveness varies depending on the data used. The most popular ones include decision trees, linear and logistic regression, Bayesian classifiers and Bayesian belief networks, neural networks and support vector machines. 

2. Clustering
   
   The goal of clustering is to derive natural group that splits the dataset into a set of categories. Clusters can be created at several different possible grain-sizes. For example, data about different schools could be clustered together to investigate similarities and differences between them, data about learners’ actions could be clustered together to enable one to investigate patterns in their learning behaviour. Clustering is particularly useful when categories within the data set are not known in advance but it can also be used with prior hypotheses about clusters possibly generated in prior research. 

3. Relationship mining
   
   Relationship mining attempts to find which variables are most strongly associated with a single variable that might be of particular interest to the researcher, or which relationships between any two variables are the strongest. Broadly speaking, there are four types of relationship mining:
   
   1. association rule mining which produces if-then rules;
   2. correlation mining which finds correlations between variables
   3. sequential pattern mining which determines temporal associations between events and
   4. causal data mining for identifying causal relationships between events.