← News

ResearchSources: arXiv · 19 Nov 2026

TCC-RAG guides retrieval with causal chains for temporal knowledge graph question answering

Researchers propose TCC-RAG, a retrieval method that restructures temporal knowledge graphs around causal cues so large language models can answer time-dependent questions with less redundant context.

In plain words

Question answering over time-stamped knowledge graphs often needs reasoning across several linked events. The proposed TCC-RAG method rebuilds the graph to highlight cause-and-effect links over time, then retrieves only the knowledge relevant to a specific query. The authors report this helps language models avoid misleading reasoning paths.

What was announced

The authors argue that semantic retrieval augmented generation, the common approach for feeding supporting facts to large language models, has two weaknesses in temporal question answering: it struggles to capture multi-hop causal knowledge, and it pulls in redundant material that can trap models in spurious reasoning chains and produce wrong answers. TCC-RAG addresses this by reformulating the temporal knowledge graph into a multi-view causal cue graph whose connections encode temporal evolution and causal dynamics, then building a query-specific temporal causal chain to steer retrieval toward a concise set of causally relevant knowledge. A causal evidence filter then refines what is retrieved, checking its consistency with the temporal logic of the query.

In experiments reported by the authors, TCC-RAG delivered gains of 10.72% and 8.40% on two TKGQA benchmarks, along with a 62.97% improvement on multi-step temporal reasoning. These figures come from the paper's abstract alone, which does not name the benchmarks, baselines, or evaluation setup, so the results cannot be independently assessed from this source.

What it does not show

  • Evidence is limited to the paper's abstract; benchmark names, baselines, datasets, and evaluation details are not provided, and no independent verification of the reported gains is available.

Sources

Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 20 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.