Grading Descriptive Answer Scripts Using Deep Learning
Neethu George1, Sijimol PJ2, Surekha Mariam Varghese3
1Neethu George, Department of Computer Science and Engineering, Mar Athanasius College of Engineering, Kothamangalam (Kerala), India.
2Sijimol PJ, Department of Computer Science and Engineering, Mar Athanasius College of Engineering, Kothamangalam (Kerala), India.
3Surekha Mariam Varghese, Department of Computer Science and Engineering, Mar Athanasius College of Engineering, Kothamangalam (Kerala), India.
Manuscript received on 18 March 2019 | Revised Manuscript received on 27 March 2019 | Manuscript published on 28 March 2019 | PP: 991-996 | Volume-8 Issue-5, March, 2019 | Retrieval Number: E3139038519/19©BEIESP
Open Access | Ethics and Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: The evaluation of answer papers considering semantics is a complex process that requires great intellectual effort from evaluators. The lack of availability of expert evaluators makes the evaluation more time consuming. Nowadays, everything is automated. Hence, in order to reduce the effort during the evaluation of answer scripts an automated system is required to grade the answer scripts correctly. This paper presents a system for descriptive answer checking and grading application based on natural language processing and deep learning. Features are extracted to create a model from the human evaluated sample dataset of answer scripts. The proposed sequential model consists of LSTM-RNN layer which sequentially takes the glove vector representation in a sentence of each word and converts to embedding vector representation. Embedding vector corresponding to the glove vector of the last word will be the representation of the entire sentence in its semantic form. The sequential model consists of embedding layer, Long Short Term Memory layer, dropout layer and dense layer. The regularization technique, dropout reduces over fitting by preventing complex co-adaptations on training data. The softmax activation function in the dense layer (fully connected neural network layer) gives the one hot encoded score for each answer. The model can assign scores of non-evaluated descriptive answers by comparing it with answer key. This approach is very useful in valuation of essays, descriptive answer scripts, document similarity checking, and plagiarism detection.
Keywords: Long Short Term Memory (LSTM), Machine Learning (ML), Recurrent Neural Network (RNN), Deep Descriptive Answer Scoring (D-DAS).
Scope of the Article: Machine Learning