ReTAG: Retrieval-Enhanced, Topic-Augmented Graph-Based Global Sensemaking

TL;DR

ReTAG addresses global sensemaking by constructing a contextualized graph $G_c=(\mathcal{V},\mathcal{E})$ over a corpus and organizing it into hierarchical communities. It extends prior graph-based sensemaking with topic augmentation to build topic-specific subgraphs and retrieval augmentation to select the most relevant summaries during answer generation, all under a $W$-token context window across $L$ levels. The approach yields significant gains in comprehensiveness and diversity while substantially reducing inference time, demonstrated on Podcast and News Articles datasets, with improvements up to $90.3\%$ in speed and higher relevance when comparing topic-augmented and retrieval-augmented configurations. By integrating topic mining, keyword-expanded retrieval, and end-to-end prompting, ReTAG enables scalable, high-quality global sensemaking across large document collections.

Abstract

Recent advances in question answering have led to substantial progress in tasks such as multi-hop reasoning. However, global sensemaking-answering questions by synthesizing information from an entire corpus remains a significant challenge. A prior graph-based approach to global sensemaking lacks retrieval mechanisms, topic specificity, and incurs high inference costs. To address these limitations, we propose ReTAG, a Retrieval-Enhanced, Topic-Augmented Graph framework that constructs topic-specific subgraphs and retrieves the relevant summaries for response generation. Experiments show that ReTAG improves response quality while significantly reducing inference time compared to the baseline. Our code is available at https://github.com/bykimby/retag.

Figures (5)

• Figure 1: Comparison between Local Traditional QA and Global Sensemaking QA. Local traditional QA focuses on local details in documents (the performer of Golden), while global sensemaking QA requires examining entity relationships (AI, humanity, etc.) across the entire corpus.
• Figure 2: Overall Process of Graph-based Global Sensemaking. All entities and relations are extracted by an LLM to build a contextualized entity-relation graph (a), which is then clustered forming hierarchical communities (b). A community summary is generated for each community based on the contexts within its corresponding subgraph (c).
• Figure 3: Topic Augmentation Pipeline. At each level $l$, given a set of community summaries (a), mining all relevant topics is essentially equivalent to performing global sensemaking on the given summaries (b). The final set of topics is constructed by taking the union of the sets of topics across all levels (c). For each topic, we apply the same community summary generation process (d), resulting in multi-level sets of community summaries generated for each topic (e).
• Figure 4: Cross-Level Winning Rates between Models with and without Topic Augmentation. We measured the winning rates (%) of the model with topic augmentation (rows) over the one without it (column) across community levels in both Podcast (a) and News Articles (b). The results indicate that Topic Augmentation generally performs better than the baseline across different levels.
• Figure 5: Recall@$p$ with Relevant Summaries with and without Keyword Expansion (KE). Recalls are measured across four levels. Relevant summaries are identified by LLM.