RECIPE-TKG: From Sparse History to Structured Reasoning for LLM-based Temporal Knowledge Graph Completion.
| Authors | Ömer Faruk Akgül et al. |
| Year | 2026 |
| Venue | EACL 2026 |
| Paper | View on ACL Anthology |
| PDF | Download |
Abstract
Abstract not yet available in this stub. Read the full paper →
Engineering Breakdown
Plain English
This paper addresses temporal knowledge graph completion—predicting missing facts in knowledge bases where relationships change over time—by helping LLMs reason better from sparse historical data. The key contribution is RECIPE-TKG, a method that structures how LLMs access and reason over incomplete temporal histories to make more accurate predictions about future or missing facts.
Key Engineering Insight
The core insight is that sparse temporal data needs explicit structuring before feeding it to LLMs; rather than raw historical sequences, organizing facts into structured reasoning patterns dramatically improves completion accuracy and reduces hallucination on time-dependent relationships.
Why It Matters for Engineers
Production systems dealing with evolving data—recommendation systems, knowledge bases, entity resolution across time—struggle with incomplete histories. This work directly applies to engineers building temporal reasoning into LLM pipelines, offering a concrete approach to improve accuracy without retraining base models.
Research Context
Prior work treated temporal KG completion as a static problem or required expensive temporal embeddings. This paper advances the field by showing LLMs can perform competitively when given proper structured reasoning guidance, shifting the problem from "train better embeddings" to "engineer better prompts and reasoning paths" for existing models.
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::