Puanlama Yöntemleri

Abstract

This chapter examines scoring methods used in educational assessment from an epistemological perspective that focuses on how scores are generated. Accordingly, the methods are organized into three main categories: process-based scoring approaches, automated scoring, and comparative judgement. The first section discusses manual and digital scoring, dichotomous and partial-credit scoring, and weighted scoring approaches in terms of their applications, advantages, limitations, and psychometric properties. It also examines how scoring procedures vary according to the purpose of assessment, the construct being measured, and the conditions under which the assessment is administered. The second section explains the historical development of automated scoring systems, tracing their evolution from traditional rule-based and feature-based models to deep learning and generative artificial intelligence-based approaches. It summarizes the operating principles of rule-based, machine learning, neural network, and ontology-based methods, and discusses the roles of CNN, LSTM, and Transformer architectures in the evaluation of written performances. The opportunities provided by large language models for automated scoring are examined within the context of zero-shot, few-shot, and fine-tuning approaches. In addition, the chapter discusses the metrics used to evaluate automated scoring performance —particularly the QWK — and the limitations associated with interpreting these metrics. The final section introduces the comparative judgement approach to scoring and explains its theoretical foundations as an alternative to absolute scoring, in which performances are evaluated through pairwise comparisons. Overall, the chapter provides a comprehensive perspective on contemporary scoring methods, ranging from traditional human scoring to AI-assisted systems and comparative judgement approaches. It offers both a theoretical and practical framework for selecting appropriate scoring methods for different assessment purposes and implementation contexts.