A dynamic semantic space modelling approach for short essay grading

The assessment of knowledge is considered as one of the most important aspect of the learning process. Despite its development over decades the underlying problem of `How best to assess learning?' still remains. In the field of education, essay questions are considered as the most appropriate question types for assessment compared to closed questions to evaluate the knowledge of the students. However, evaluation of answers of essay type questions consumes a long time, effort and includes unavoidable human errors. Therefore, the development of an automated essay assessment system is beneficial due to those reasons. The focus of this research is to present a novel approach for automated essay scoring (AES) using Vector Space Models (VSMs) and Natural Language Processing techniques. It employs model answer based evaluation for the scoring process. In order to handle variations in the students' essay answers, NLP techniques (lemmatization, tokenization, handling of spelling mistakes, relation of objects, upper and lower case of words, short term resolution) were used. Proposed approach does not need any pre-training prior to each essay questions compared to most of the existing systems. Importance of this approach is that it does not need any domain specific corpus; instead, a sematic space is built using same students' answers. The scores were derived by comparing the semantic similarity or deviation between the model answer and the student answers and finally provide an accurate and consistent score. Results suggested that there is a strong correlation between the system score and the average human score.