DOI: 10.5176/2251-2195_CSEIT19.158
Authors: Mr. Rutvik Dixit
Abstract: Assessing numerous and distinct essays is both time and resource consuming which is considered one of the most significant activities and plays a dominant role in the education field. In this paper, we aim to develop a model that provides a cost-efficient and compatible alternative to human scoring. Traditional essay grading systems relied on crafted extracted features for the assessment of essays. The performanceofsucha kind of system is profoundly dependent on the nature orquality of the designed features. Despite, it is arduoustodevisethemost instructive features for such a system manually. Theprominent barrier to accepting constructed-response assessments over traditional multiple-choiceassessmentsisthesubstantialcostand laborious effort wanted for grading orscoring.Thisprojectisan attempt to use different neural network structures to develop a comprehensive automated essay grading system to solve this intricacy. Therefore, in this paper, we seek to develop a model based on the neural network approach for the assignment of automated essay grading or scoring. The result that our system was able toachieveaQuadraticWeightedKappa(QWK)scoreof 0.9608 that has marginally outperformed the baseline models developed by a team at Stanford University and a team of Carnegie Mellon Universitywhichmanagedtoobtainscore0.833 and 0.9447875 respectively and achieved a state of art performance in field of automated essay scoring. The scores achieved by the models based on the neural network approach verifies the immense potential of the development of the models using deep neural network approaches in the field of automated scoring of essays.
Keywords: component; Automation; Essay scorer; Recurrent Neural Networks; Machine Learning; LSTM; Word2vec; Neural Networks; Kaggle competition; ASAP dataset.
Updating...