Towards automatic assessment of online writing on Wiki

Abstract

Wiki is a collaborative writing platform where all the users to the site can create new pages and modify the existing one. Due to the advantages of Wiki in the educational context, Wiki has been increasingly introduced into the classroom as the platform supporting students’ collaborative writing for group project. However, assessing writings on Wiki brings teachers the extra burden, especially considering the assessment of historical versions of writings kept on Wikis. This study aims at exploring the potential of various features in automating the assessment of the quality of Wiki writing. The Wiki writing to be studied has been extracted from one undergraduate course of two academic years at the University of Hong Kong (HKU) and analyzed quantitatively. Based on the literature of writing analytics, four types of textual features were proposed to measure quality of writing, including complexity, cohesion, reader and writer interaction, and content. Stepwise regression analysis and machine learning techniques were adopted to find whether the features can measure and predict the writing quality as reflected in assessment scores given by human instructors. In addition, all proposed features were calculated on all historical versions of the Wiki writings, and the study also investigates whether the changes of the features over the writing process can be used to measure and predict writing quality. Findings of the study will form the foundation towards automating the assessment of Wiki writings which can provide assistance to instructors and/or serve as timely feedback to students.