MSLEF: Multi-Segment LLM Ensemble Finetuning in Recruitment

2025-09-07

AI Summary

To address low parsing accuracy caused by heterogeneous resume formats, this paper proposes a multi-paragraph large language model (LLM) ensemble fine-tuning framework. The method employs a segment-aware architecture that assigns field-specific weights to different resume sections and integrates outputs from fine-tuned models—including Gemma-9B, LLaMA-3.1-8B, and Phi-4-14B—via a higher-order aggregator (Gemini-2.5-Flash), combined with weighted voting and multi-task evaluation. Compared to the best-performing single-model baseline, our framework achieves consistent improvements across all evaluation metrics: Exact Match (EM), F1, BLEU, ROUGE, and Recruitment Similarity (RS), with RS increasing significantly by 7%. This demonstrates substantially enhanced cross-structural generalization capability, effectively supporting accurate candidate representation in real-world recruitment scenarios.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This paper presents MSLEF, a multi-segment ensemble framework that employs LLM fine-tuning to enhance resume parsing in recruitment automation. It integrates fine-tuned Large Language Models (LLMs) using weighted voting, with each model specializing in a specific resume segment to boost accuracy. Building on MLAR , MSLEF introduces a segment-aware architecture that leverages field-specific weighting tailored to each resume part, effectively overcoming the limitations of single-model systems by adapting to diverse formats and structures. The framework incorporates Gemini-2.5-Flash LLM as a high-level aggregator for complex sections and utilizes Gemma 9B, LLaMA 3.1 8B, and Phi-4 14B. MSLEF achieves significant improvements in Exact Match (EM), F1 score, BLEU, ROUGE, and Recruitment Similarity (RS) metrics, outperforming the best single model by up to +7% in RS. Its segment-aware design enhances generalization across varied resume layouts, making it highly adaptable to real-world hiring scenarios while ensuring precise and reliable candidate representation.

Problem

• Enhancing resume parsing accuracy in recruitment automation
• Overcoming single-model limitations with multi-segment ensemble framework
• Adapting to diverse resume formats and structures effectively

Innovation

• Segment-aware ensemble with weighted voting per resume part
• Specialized LLMs for different resume segments integration
• High-level aggregator LLM for complex sections handling