Prediction of Question Tags Based on LDA and Deep Neural Network

Abstract

Students are being evaluated based on the examinations conducted by various institutions or organizations, which test the knowledge of that person. Exams like Computerized Adaptive Testing (CAT), offers a computer based test that adapts the examinee's ability level. Some of the CAT exams include tags which help students to understand the questions. Tags are metadata used to identify or describe an item. There are three types of tags: Manual Tagging, Semi-Automatic tagging and Fully Automatic tagging. Earlier manual tagging was used to construct question banks. However it is time consuming and leads to many other consistency issues. A Semi-Automatic tagging facilitates human intervention to increase the accuracy of tagging. Fully automatic tagging gives a more promising result as compared with manual and semi-automatic tagging. This paper proposes a fully automated tagging system which uses Deep Neural Network and Natural Language Processing to generate tags from the derived knowledge unit. This paper also discusses LDA (Latent Dirichlet Allocation) which gives the relevance of each tag.