A strategy for detection of inconsistency in evaluation of essay type answers

摘要

The quality of evaluation of essay type answer books involving multiple evaluators for courses with large number of enrollments is likely to be affected due to heterogeneity in experience, expertise and maturity of evaluators. In this paper, we present a strategy to detect anomalies in evaluation of essay type answers by multiple evaluators based on the relationship between marks/grades awarded and symbolic markers, opinionated words recorded in answer books during evaluation. Our strategy is based on the results of our survey with evaluators, analysis of large number of essay type evaluated answer books and our own experiences regarding grievances of students regarding marks/grades. Results of both survey and analysis of evaluated answer books identified underline, tick and cross as frequently used markers compared to circle and question mark. Further, both opinionated words and symbolic markers identified through the survey are used by evaluators to express either positive or negative sentiments. They have differential usage pattern of these symbols as single evaluator and as one amongst multiple evaluators. Tick and cross have well define purposes and have strong correlation with marks awarded. However, the underline marker is being used for dual purpose of expressing both correctness and incorrectness of answers. Our strategy of inconsistency detection first identifies outliers based on the relationship between marks/grades awarded and number of symbols and/or opinionated words used in evaluation. Subsequently, marks and number of symbolic markers of outliers are compared with peer non-outlier answer books having same marks but different number of markers used. Such outlier answer books are termed as anomalous. We discovered 36 anomalies out of total 425 evaluated answer books. We have developed a prototype tool to facilitate online evaluation of answer book and to proactively alert evaluators of possible anomalies.