Replication Package for "The State of RL4SE: A Task-Oriented Review of Research on Reinforcement Learning for Software Engineering"

Abstract

This record contains the replication package for the survey “The State of RL4SE: A Task-Oriented Review of Research on Reinforcement Learning for Software Engineering.” The package supports the study-identification, screening, classification, and artifact-availability results reported in the survey. It contains: • candidate-level screening records for the venue-based query, four rounds of bidirectional snowballing, and the supplementary AI/ML-conference search;• a deduplicated bibliography of 146 included studies;• software-engineering task and reinforcement-learning algorithm labels;• a summary of datasets, benchmarks, interaction environments, simulators, and publicly available artifact links;• snowballing seeds and citation-provenance paths;• search, eligibility, screening, and classification protocols;• study-selection and GUI-study figures; and• scripts for validating the package and regenerating its figures. The final corpus comprises 92 studies identified through the venue-based query, 48 through bidirectional snowballing, and 6 through the supplementary search of the main conference tracks of NeurIPS, ICLR, AAAI, IJCAI, and ICML. The publication window is from 1 January 2006 to 30 June 2025. The supplementary AI/ML-conference search screened 44 new candidates after records already represented in the Query and Snowballing branches were removed. Six studies were included and 38 were excluded. The package README describes the directory structure and validation procedure. The LICENSE file specifies the applicable terms for the authors’ curated materials, scripts, and third-party bibliographic content.