ToM: Leveraging Tree-oriented MapReduce for Long-Context Reasoning in Large Language Models
ToM: Leveraging Tree-oriented MapReduce for Long-Context Reasoning in Large Language Models [107.9] ToMは、長期コンテキスト推論のための新しいツリー指向MapReduceフレームワークである。 ToM は既存の分割・クエリー・フレームワークや検索拡張生成手法よりも大幅に優れていることを示す。 論文参考訳（メタデータ） (Sat, 01 Nov 2025 10:43:58 GMT)
「Leveraging a tree- structured MapReduce approach, ToM performs recursive reasoning over documents to enhance long-context understanding. It consists of two key components: DocTree Construction: ToM first applies Hierarchical Semantic Parsing to convert each chunk into a structured subtree, then combines these subtrees into a hierarchical DocTree through Bottom-up Aggregation. 2). Recursive Reasoning via MapReduce: ToM performs recursive reasoning on the DocTree in a MapReduce fashion, enabling systematic aggregation of rationales across the hierarchy. 」とTree構造化＆MapReduceを用いる長文処理の提案。一般的なRAGよりも性能が良いとのこと。