Источник
FSE
Дата публикации
23.06.2025
Авторы
Карина Романова Сергей Сеничев Лина Вельтман Иван Насонов Андрей Кузнецов Илья Макаров
Поделиться

SODAOpt: Socio-Demographic and Textual Adaptive Fusion for Optimizing Developer Task Assignment

Аннотация

Modern software engineering faces challenges in task assignmentdue to over-reliance on static identifiers, which ignore contextualdata like documentation, commit logs, and developer profiles. Thislimits adaptability in resource allocation and bug triage.We proposeSODAOpt, a transformer-based framework that integrates textualand socio-demographic embeddings to enhance task matching. Ourapproach employs adaptive fusion layers to unify heterogeneousdata modalities and a composite loss function balancing contrastivelearning with assignment optimization. Evaluations on real-worlddatasets demonstrate SODAOpt’s superiority in precision and crossprojectdiversity, validated by tailored metrics (SP@K, CPR, CNDCG@K).

Присоединяйтесь к AIRI в соцсетях