MMAssist

An IntelliJ IDEA assistant that recommends Move Method refactorings and delegates behavior-preserving edits to the IDE engine.

Co-author ICSME 2025 Refactoring

MMAssist studies Move Method recommendation as a combined retrieval, reasoning, and refactoring problem. It ranks candidate target classes, uses an LLM to reason over the strongest candidates, and applies the selected transformation through IntelliJ IDEA's refactoring engine.

Problem

Move Method refactoring requires identifying a target class whose responsibilities better match the method. This decision is difficult to make manually at scale and unreliable when an LLM is used without structured candidate evidence.

Approach

MMAssist ranks candidate target classes by semantic affinity, asks an LLM to reason over the top candidates, and applies the accepted move through the IDE refactoring engine rather than generated source edits.

Results

On the synthetic corpus, Recall@1 and Recall@3 reached 73% and 80%. On 210 verified real-world Move Method refactorings, Recall@1 and Recall@3 reached 71% and 82%; in a 30-participant study, 290 of 350 classes received a positive recommendation.

Why it matters

Separating recommendation from transformation lets the model assist the developer while the IDE supplies the behavior-preserving edit mechanism.

Project site: cuboulder-se-research.github.io/move-method-assist.

Details and evaluation are in the paper (Bellur et al., 2025).

References

2025

  1. Leveraging LLMs, IDEs, and Semantic Embeddings for Automated Move Method Refactoring
    Abhiram Bellur, Fraol Batole, Malinda Dilhara, and 11 more authors
    In ICSME’2025: The 41st IEEE International Conference on Software Maintenance and Evolution, Auchland, New Zealand, Sep 2025