Subjective Evaluation: A Comparison of Several Statistical Techniques

Evaluation of subjective examinations using computerized tools has been a topic of research for more than four decades. Several statistical and mathematical techniques have been proposed by various researchers. In this research work, the several methods proposed earlier like Latent Semantic Analysis (LSA), Generalized Latent Semantic Analysis (GLSA), Bilingual Evaluation Understudy (BLEU), and Maximum Entropy (MaxEnt) are compared on common input data. The techniques are implemented using Java programming language, MatLab, and other open source tools. Experiments have been conducted and developed prototypes are tested using a database of 4500 answers with approximately 50 questions of computer science. Comparison of these techniques on a common database is not available in the literature as far as the authors' review is concerned. The database used for testing is collected by conducting tests of students of graduate level in the field of computer science. The pros and cons of each technique on the basis of experiments are discussed in the paper.