School of Science and Technology 科技學院
Computing Programmes 電腦學系
| Programme | Bachelor of Computing with Honours in Internet Technology |
| Supervisor | Dr. Andrew Lui |
| Areas | Teaching and Classroom Support |
| Year of Completion | 2017 |
| Award Received | First Runner-Up, IEEE Computational Intelligence Chapter FYP Competition 2017 |
The aim of the project is to develop a semi-automated grading short answer algorithm to reduce and make better use of instructors grading effort. In traditional way, short answer grading process need a lot of effort and time. We think that this process can be done in a more efficient way. Semi-automated grading is the combination of traditional grading and automated grouping of answer. We will design a framework for clustering the short answer in different group. The answers which are similar and share common feature are grouped in a cluster. We want to explore the ideal feature set so that the answers will successfully cluster the answers according to the similarity. Hence, instructors only grade representative answer of each group, the grading apply to all the answer in the cluster.
We also want to investigate the balance of the numbers of cluster needed to grade and the accuracy of clusters. We hope instructors can grade the fewest cluster and still have excellent grading performance.
To achieve the aim, the main of objective is to classify and cluster the student short answer and then grade the short answer automatically. The project has defined a number of sub-objectives as follows:
This project is a research based project which aims to cluster the short answers to amplify graders effort. Graders are likely to have less clusters to grade (i.e. the minimalized K value). However, with the lower clusters number, the harder to get pure cluster. Pure clusters mean that the group contains answers that deserve same grade.
To achieve this, there are some investigations on balancing these two dimensions as follows:
Semi-Automated grading is still a challenging problem. This research suggests the use of Typed Dependencies as features is advanced and worth to continue investigating further. Compared to fully-automated grading, semi-automating combines the human grading and computer-assisted, which provides higher flexibility and accuracy to the graders. Also, their effort would be amplified compared to traditional grading process. Our project aim is achieved. This research has mainly two contributions. In the following paragraphs, we will discuss them in the following sub-chapters.
There are some limitations in our project. Synonyms and antonyms cannot be identified in our algorithm. It is important to have it because students may use different words to present the same meaning. On the other hand, abbreviation cannot be detected in our algorithm. In some dataset, there is a frequently use of abbreviation to express their answers. For example, “us” can be literally understand as a pronoun and the abbreviation of “United State”. Handling them can benefit to the accuracy of the algorithm.
Also, we did not investigate other possibly use of clustering algorithm. We used K-Means in our experiment only. There is possibility that other clustering algorithm could have better performance then K-Means.
As the limitations stated above, we suggest that future work can aim at them. Synonyms, antonyms and abbreviation features can be investigated to check whether it is useful to identify the answers more accurately.
Clustering algorithm investigation is also one of the future work. The choice of algorithms can be a main issue to have different performance.
