Research on a Software System Size Evaluation Model Combining BiLSTM-CRF with Domain Adaptation Learning

In the field of software engineering, accurate and efficient measurement of system size is crucial for project management and decision-making. Traditional software system size evaluation methods, such as expert decision-making methods, COCOMO models, function point methods, and analogical methods, have problems such as relying on manual labor, low efficiency, and susceptibility to subjective factors. To address these issues, this paper proposes a Dalse (Domain Adaptive Learning System Evaluation) automated method for evaluating software function size. Based on the IFPUG function point standard, this method constructs a BiLSTM-CRF model for identifying function types, and introduces domain adaptive learning technology to solve the problem of insufficient labeled text data for domain-specific annotation. This enables the method to flexibly adapt to the characteristics of software in specific domains, achieving automated extraction and quantitative calculation of software function points. Experimental validation shows that the method's precision, recall, and F1 score reached 83%, 80%, and 82%, respectively, which is a 5% improvement compared to the accuracy of pure BiLSTM-CRF function point recognition.