Zero-1-to-3: Domain-level Zero-shot Cognitive Diagnosis Via One Batch of Early-bird Students Towards Three Diagnostic Objectives

Cognitive diagnosis seeks to estimate the cognitive states of students byexploring their logged practice quiz data. It plays a pivotal role inpersonalized learning guidance within intelligent education systems. In thispaper, we focus on an important, practical, yet often underexplored task:domain-level zero-shot cognitive diagnosis (DZCD), which arises due to theabsence of student practice logs in newly launched domains. Recent cross-domaindiagnostic models have been demonstrated to be a promising strategy for DZCD.These methods primarily focus on how to transfer student states across domains.However, they might inadvertently incorporate non-transferable information intostudent representations, thereby limiting the efficacy of knowledge transfer.To tackle this, we propose Zero-1-to-3, a domain-level zero-shot cognitivediagnosis framework via one batch of early-bird students towards threediagnostic objectives. Our approach initiates with pre-training a diagnosismodel with dual regularizers, which decouples student states into domain-sharedand domain-specific parts. The shared cognitive signals can be transferred tothe target domain, enriching the cognitive priors for the new domain, whichensures the cognitive state propagation objective. Subsequently, we devise astrategy to generate simulated practice logs for cold-start students throughanalyzing the behavioral patterns from early-bird students, fulfilling thedomain-adaption goal. Consequently, we refine the cognitive states ofcold-start students as diagnostic outcomes via virtual data, aligning with thediagnosis-oriented goal. Finally, extensive experiments on six real-worlddatasets highlight the efficacy of our model for DZCD and its practicalapplication in question recommendation. The code is publicly available athttps://github.com/bigdata-ustc/Zero-1-to-3.