SaraCoder: Orchestrating Semantic and Structural Cues for Profit-Oriented Repository-Level Code Completion

2025-08-13

AI Summary

In warehouse-scale code completion, shallow textual retrieval leads to semantic misdirection, result redundancy, homogeneity, and external symbol ambiguity. To address these challenges, this paper proposes a Hierarchical Feature-Optimized Retrieval-Augmented Generation (RAG) framework. Our approach introduces three key innovations: (1) integrating deep semantic similarity with code graph structural similarity, enhanced by graph-topological importance scoring to improve edit-awareness; (2) performing cross-file dependency analysis for external-context-aware identifier disambiguation; and (3) designing a multi-objective re-ranking mechanism that jointly optimizes relevance and diversity. Evaluated on the CrossCodeEval and RepoEval-Updated benchmarks, our method achieves significant improvements over existing state-of-the-art approaches. It demonstrates strong effectiveness, generalizability across programming languages, and robustness across diverse LLM backends—validating its practical utility for large-scale, real-world code completion tasks.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Search and Optimization: Sampling/Simulation-based Search
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Retrieval-augmented generation (RAG) for repository-level code completion commonly relies on superficial text similarity, leading to results plagued by semantic misguidance, redundancy, and homogeneity, while also failing to resolve external symbol ambiguity. To address these challenges, we introduce Saracoder, a Hierarchical Feature-Optimized retrieval framework. Its core Hierarchical Feature Optimization module systematically refines candidates by distilling deep semantic relationships, pruning exact duplicates, assessing structural similarity with a novel graph-based metric that weighs edits by their topological importance, and reranking results to maximize both relevance and diversity. Furthermore, an External-Aware Identifier Disambiguator module accurately resolves cross-file symbol ambiguity via dependency analysis. Extensive experiments on the challenging CrossCodeEval and RepoEval-Updated benchmarks demonstrate that Saracoder significantly outperforms existing baselines across multiple programming languages and models. Our work proves that systematically refining retrieval results across multiple dimensions provides a new paradigm for building more accurate and robust repository-level code completion systems.

Problem

• Addresses semantic misguidance and redundancy in code completion
• Resolves external symbol ambiguity via dependency analysis
• Improves retrieval accuracy and diversity in repository-level coding

Innovation

• Hierarchical Feature Optimization for code retrieval
• Graph-based structural similarity assessment
• Dependency-aware symbol disambiguation