Divide, Cache, Conquer: Dichotomic Prompting for Efficient Multi-Label LLM-Based Classification
This addresses the efficiency problem in multi-label classification for LLM users, though it is incremental as it builds on existing prompting and distillation techniques.
The paper tackles multi-label text classification with large language models by decomposing it into sequences of yes/no decisions and using prefix caching, achieving substantial efficiency gains for short-text inference without accuracy loss. They demonstrate the approach on affective text analysis with 24 dimensions, showing significant improvements over zero-shot baselines through distillation and fine-tuning of smaller models.
We introduce a method for efficient multi-label text classification with large language models (LLMs), built on reformulating classification tasks as sequences of dichotomic (yes/no) decisions. Instead of generating all labels in a single structured response, each target dimension is queried independently, which, combined with a prefix caching mechanism, yields substantial efficiency gains for short-text inference without loss of accuracy. To demonstrate the approach, we focus on affective text analysis, covering 24 dimensions including emotions and sentiment. Using LLM-to-SLM distillation, a powerful annotator model (DeepSeek-V3) provides multiple annotations per text, which are aggregated to fine-tune smaller models (HerBERT-Large, CLARIN-1B, PLLuM-8B, Gemma3-1B). The fine-tuned models show significant improvements over zero-shot baselines, particularly on the dimensions seen during training. Our findings suggest that decomposing multi-label classification into dichotomic queries, combined with distillation and cache-aware inference, offers a scalable and effective framework for LLM-based classification. While we validate the method on affective states, the approach is general and applicable across domains.