e-Learning with Kaggle in Class: Adapting the ALTA Shared Task 2013 to a Class Project

e-Learning with Kaggle in Class: Adapting the ALTA Shared Task 2013 to a Class Project Karin Verspoor1,2 and Jeremy Nicholson2 1National ICT Australia 2Department of Computing and Information Systems The University of Melbourne Melbourne VIC 3010 Australia karin.verspoor@nicta.com.au, nj@unimelb.edu.au Abstract The 2013 ALTA Shared Task was utilised as a class project for a subject taught at The Uni- versity of Melbourne in the second semester of 2013. This paper reviews the experience of us- ing an on-line, Kaggle in Class-based shared task for class work. Adoption of the shared task enables a blended learning paradigm that en- gages students in problem-based learning in a shared and open context. 1 Introduction As in recent years, the Australasian Language Tech- nology Association sponsored a shared task in 2013 to stimulate interest in language technology tasks among university students (Molla, 2013). This year’s task was primarily organised by Diego Molla of Macquarie University and addressed the restoration of normal case (capitalisation) and punctuation to a noisy text input not conforming to conventional use of case and punctuation. As described in (Molla, 2013) the task is framed as a simplification of the gen- eral task explored by (Baldwin and Joseph, 2009). Because the task is specifically aimed at university students with programming skills, and as it can be ap- proached as a classification task, it is appropriate to consider as a project for a university subject that ad- dresses machine learning algorithms. Furthermore, it provides an opportunity to make use of blended learning (Garrison and Kanuka, 2004) in the class- room; that is, integrating face-to-face learning with on-line asynchronous learning opportunities. Enrich- ment of the traditional classroom learning experience with on-line activities has been suggested to have positive benefits for student learning, in addition to student satisfaction and retention. The task was therefore selected as a project for ap- proximately 115 students registered for The Univer- sity of Melbourne’s Knowledge Technologies subject, a subject in the Department of Computing and Infor- mation Systems for which the stated objective is to “learn algorithms and data structures for extracting, retrieving and storing explicit knowledge from vari- ous data sources, and methods for data mining and machine learning with complex data.” The students were given the option to register for the shared task formally through the Kaggle In Class system. 2 Organisation To adapt the shared task to the classroom context, the project was split into several stages. 2.1 Data pre-processing and task familiarisation The students were introduced to the task through the ALTA shared task data, without an explicit reference to the shared task itself. They were given the context of the task in a project specification, provided with the training data and asked to write scripts to manip- ulate the data in various ways. In one subtask, they were asked to map the provided ALTA Shared Task data format to the ARFF (Attribute-Relation File For- mat) format which is used in several machine learn- ing frameworks. This was intended to get the students comfortable with …

Content truncated to 500 words