Udit Kr. Chakraborty

A Novel Semantic Similarity Based Technique for Computer Assisted Automatic Evaluation of Textual Answers

We propose in this paper a unique approach for the automatic evaluation of free text answers. A question answering module has been developed for the evaluation of free text responses provided by the learner. The module is capable of automatically evaluating the free text response of the learner SA to a given question Q and its model text based answer MA on a scale [0, 1] with respect to the MA. This approach takes into consideration not only the important key-words but also stop words and the positional expressions present in the learners’ response. Here positional expression implies the pre-expression and post-expression appearing before and after a keyword in the learners’ response. The results obtained on using this approach are promising enough for investing into future efforts.

A Fuzzy Indiscernibility Based Measure of Distance between Semantic Spaces Towards Automatic Evaluation of Free Text Answers

Quantitative models built as tools for evaluating human language performance, can prove useful for both theoretical and applied areas of discourse comprehension, assessment, and education. The curr...

A quantum parallel bi-directional self-organizing neural network (QPBDSONN) architecture for extraction of pure color objects from noisy background

This paper is aimed to propose a suitable real-time pure color image denoising procedure using a self-supervised network referred to as Quantum Parallel Bi-directional Self-Organizing Neural Network (QPBDSONN) architecture. The proposed QPBDSONN replicates the Parallel Bi-directional Self-Organizing Neural Network (PBDSONN) architecture and exploits by the power of quantum computation. To process three distinct basic color components (Red, Green and Blue) of noisy color image, QPBDSONN uses trinity of Quantum Bi-directional Self-organizing Neural Network (QBDSONN) architecture at the source layer in parallel mode. Each constituent QBDSONN comprises input, intermediate or hidden and output layers interconnected by 8-connected neighborhood topology based layer of neurons represented by qubits. Each constituent QBDSONN updates weighted interconnections in the form of quantum states through counter-propagation between hidden and output layer to obviate quantum back propagation. Rotation gates are introduced to represent weighted inter-links and values of activation. Finally, a quantum measurement operation is performed at the output layer of each constituent QBDSONN followed by fusion operation at the sink layer of QPBDSONN to concatenate the processed color image components resulting in the true output. The superiority of the proposed network architecture over the classical PBDSONN can be established using a real-life spanner pure color image and synthetic pure color image corrupted with various intensity of uniform and Gaussian noise in terms of extraction time and shapes.

Rough Set based keyword selection and weighing for textual answer evaluation

Automatic assessment of learners responses has gained wider acceptance and popularity in recent times. Due to associated complexities of free text evaluation, the trend has gradually shifted towards close ended question which have their limitations. The current work proposes a rough set based strategy to augment automated free text evaluation system(s) using keyword and associated expression based technique. The proposed method uses human evaluated answers as training data and using Rough Set Theory, extracts information from them to be used in the shortlisting and weighing of keywords which are to be used in assessment. The results of the proposed technique outperforms the manual keyword selection and weight association and also higher correlation with human evaluators.