Automatic assessment of descriptive answers in online examination system using semantic relational features

• Published in: The Journal of supercomputing Vol. 76; no. 6; pp. 4430 - 4448
• Main Authors: Nandini, V., Uma Maheswari, P.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.06.2020; Springer Nature B.V
• Subjects: Classification; Compilers; Computer Science; Evaluation; Feature extraction; Interpreters; Manpower; On-line systems; Processor Architectures; Programming Languages; Quality; Questions; Systems analysis; Descriptive answers; Question classification; Answer classification; Automatic assessment
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-018-2381-y

The revolution in technology reduces the effort of manpower in many of the areas. The boon of the technology and rapid advancements in education industry has provided a good learning environment. It offers qualification and credits at the desktop through online courses and evaluation. The prevailing system has its own pause in terms of volume, staffing, variation in the strategies of assessing. As of now, the objective-type questions alone can be practiced and assessed through online examinations. Researchers strive to build systems for evaluating descriptive answer as it is challenging and could not take up its full sway for complete automation. The challenge lies in recognizing the natural language answers and extracting the precise meaning so as to appropriately evaluate the knowledge obtained by the student. The proposed method contains stages such as question classification, answer classification and answer evaluation for the answers given by the student and grade them with appropriate score. A syntactical relation-based feature extraction technique is proposed for automatic evaluation of descriptive-type answers. The system has also adopted a cognitive-based approach where the student answers are judged for its correctness based on the phrases used for answering the questions. The score and feedback are provided to make aware of the understanding level of the subject. The experimental analysis shows .85% higher precision and recall when compared to the earlier systems.