Constructing Diverse Seed Sets for Snowballing with LLM-Assisted Candidate Assessment

This repository provides supplementary materials to support the replication of the study “Constructing Diverse Seed Sets for Snowballing with LLM-Assisted Candidate Assessment.” The study replicates two existing Systematic Literature Reviews (SLRs) in software engineering to investigate the performance of snowballing using diversity-based seed set construction. It compares conventional seed set construction approaches with a diversity-driven approach and investigates two strategies for candidate study assessment: Keyword-based strategy: Candidate studies are assessed using general criteria based on keywords, synonyms, and citation counts, followed by diversity-based seed set construction. LLM-assisted strategy: Candidate studies are assessed using Gemini 3.1 Pro against the inclusion and exclusion criteria of the original SLRs, followed by independent manual validation and diversity-based seed set construction. Both strategies consider three diversity characteristics: authors (DC1), publication years (DC2), and publication venues (DC3). The repository includes a PDF document describing the replication procedures and spreadsheets containing the data used for seed set construction, candidate study assessment, snowballing execution, and analysis of the results. Replication materials: Replication Process of Study (PDF): Step-by-step instructions for reproducing the keyword-based and LLM-assisted strategies. Seed Set SLRs.xlsx: Original and diversity-based seed sets for SLR1 and SLR2. SLR1 - Snowballing Replication Data.xlsx: Snowballing execution data and results for SLR1. SLR2 - Snowballing Replication Data.xlsx: Snowballing execution data and results for SLR2. The materials support the analysis of study coverage, screening effort, relative recall, precision, and F-measure, facilitating the reproducibility and comparison of the investigated strategies.