cs.AIOct 19, 2026
SaveTimelineAgent: An LLM-based Framework for Query-Centric Event Timeline Extraction Across Multiple Documents
Organizations: Zhejiang University · School of Engineering, Westlake University · Institute of Advanced Technology, Westlake Institute for Advanced Study · Universität des Saarlandes ·
Abstract
Query-centric cross-document event timeline extraction aims to construct event timelines from multiple documents under diverse query focuses. The task is very challenging due to the heterogeneous event modeling and reasoning requirements arising from both the intrinsic complexity of cross-document timeline construction and the diversity of query focuses. In this paper, we propose TimelineAgent, an LLM-based multi-agent framework featuring a multi-stage reasoning workflow, where heterogeneous event features are organized into reusable skills and orchestrated through a goal-oriented protocol. To facilitate evaluation, we construct a benchmark covering diverse query focuses. Experimental results show that TimelineAgent consistently outperforms existing methods, offering a practical framework for complex real-world timeline construction.
Figures & tables
Figure 1: Examples of query-centric cross-document event timelines under different query focuses. Entity queries target event arguments (participants), Aspect queries characterize event types, and Seed Event queries capture event-to-event relations, together forming the basis for event understanding.
| Workflow Stage | Skill Categories (No. of Skills) |
|---|---|
| Event Detection | Syntactic(Parsing(3)+Resource(1)) |
| Relevance Matching | Semantic(2) 2,3 , Argument(4) 1,3 , Causal(1) 2,3 |
| Temporal Anchoring | Temporal(Extract(4)+Inference(3)) |
| Eligibility Filtering | Factuality(4) 1,2,3 , Semantic(1) 3 , Temporal(1) 3 |
| Event Coreference | Semantic(1), Argument(1) |
Table 1: Stage-wise coupling of skill categories and query types: Entity 1 , Aspect 2 , and Seed-event 3 . The specific skills are listed in Table 7 (Appendix A.2 ).
Figure 2: Our proposed TimelineAgent framework.
| Entity | Aspect † | Seed-event † | |
| # documents | 120 | 46 | 1475 |
| # events | 1102 | 647 | 406 |
| # timelines | 42 | 28 | 26 |
| events / timeline | 26.3 | 23.1 | 15.6 |
| event chains / timeline | 19.9 | 13.9 | 11.1 |
| docs / timeline | 6.9 | 8.6 | 6.0 |
Table 2: Dataset statistics of our benchmark. † denotes datasets newly annotated in this work.
| Method | Entity | Aspect | Seed-event | ||||||
|---|---|---|---|---|---|---|---|---|---|
| P | R | F1 | P | R | F1 | P | R | F1 | |
| DLT | 20.7 | 11.0 | 14.3 | - | - | - | - | - | - |
| Qwen3-32B | |||||||||
| ICL | 17.8 | 12.1 | 14.4 | 28.1 | 23.8 | 25.8 | 8.3 | 12.9 | 10.1 |
| TDAG | 15.3 | 20.8 | 17.6 | 33.5 | 27.5 | 30.2 | 11.3 | 19.6 | 14.3 |
| SFT | 24.0 | 27.9 | 25.8 | 40.2 | 31.8 | 35.5 | 17.6 | 25.8 | 20.9 |
Table 3: Performance comparison on QUEST.
Figure 3: Effects of skill library, orchestration protocol and staged execution of TimelineAgent.
Figure 4: Effects of different workflow stages of TimelineAgent.
| Oracle Setting | Entity | Aspect | Seed-event |
|---|---|---|---|
| TimelineAgent | 37.9 | 42.9 | 33.1 |
| Gold Detection, Matching, & Filtering | 67.6 | 69.8 | 71.4 |
| Gold Temporal Anchoring | 55.6 | 52.4 | 41.0 |
| Gold Event Coreference | 42.1 | 56.1 | 39.4 |
Table 4: Error propagation analysis using gold annotations for different workflow stages.
Figure 5: Comparison of performance across different LLM backbones under entity setting.
| Backbone | Method | Entity | Aspect | Seed-event |
|---|---|---|---|---|
| Qwen3-32B | ICL | 14.4 | 25.8 | 10.1 |
| SFT | 25.8 | 35.5 | 20.9 | |
| SFT (medium) | 28.6 | 38.0 | 24.4 | |
| SFT (large) | 29.9 | 39.5 | 26.2 | |
| TimelineAgent | 32.1 | 38.9 | 27.3 | |
| DeepSeek-V3 | TimelineAgent | 37.9 | 42.9 | 33.1 |
Table 5: F1 scores at different supervised training scales for end-to-end SFT. The standard, medium, and large SFT settings use 6/6/5, 18/12/10, and 30/18/15 training timelines for Entity, Aspect, and Seed-event queries, respectively.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
| Category | Atomic Skills | Configurations |
|---|---|---|
| Syntactic | Part-of-speech tagging, constituency parsing, dependency parsing, phrase span detection | Collins English Dictionary for phrase span detection |
| Semantic | Event type classification, semantic similarity modeling | predefined semantic categories for classification |
| Argument | Argument extraction, entity coreference resolution, bridging anaphora | PropBank schemas, FrameNet schemas, predefined schemas for artument extraction |
| Factuality | Polarity classification, certainty classification, factuality classification, conditionality classification | labels for certainty and factuality classification |
| Temporal | Temporal expression extraction, tense analysis, aspect analysis, ordering cue extraction, temporal normalization, date calculation, temporal relation classification | Temporal relation label sets for classification |
| Causal | Causal relation classification | ConceptNet causal knowledge |
Table 6: Atomic skills and configurations in the feature-oriented skill library.
| Pipeline Stage | Skills |
|---|---|
| Event Candidate Detection | Syntactic (POS tagging, constituency parsing, dependency parsing, lexical resource (Collins English Dictionary)) |
| Query-relevant Event Matching | Semantic (predefined event semantic typing 2 , semantic similarity 3 ) Argument (argument extraction (predefined 3 , PropBank 1 , FrameNet 3 ), entity coreference resolution 1 , bridging anaphora 1 ) Causal (causal relation (ConceptNet 2,3 )) |
| Event Temporal Anchoring | Temporal (tense, aspect, time expression, ordering cues, date calculation, temporal normalization, temporal relation) |
| Event Eligibility Filtering | Factuality (polarity, factuality, certainty, condition construction) Temporal (date calculation 3 ) Semantic (predefined semantic typing) |
| Cross-document Event Coreference Resolution | Semantic (semantic similarity) Argument extraction (PropBank) |
Table 7: Goal-oriented orchestration configurations across workflow stages and query settings: Entity 1 , Aspect 2 , and Seed-event 3 .
| Parameter | Value |
|---|---|
| Train Type | LoRA |
| LoRA Rank ( ) | 16 |
| LoRA Alpha | 32 |
| Target Modules | all-linear |
| Learning Rate | |
| Warmup Ratio | 0.05 |
Table 8: Training configuration for LoRA-based SFT experiments.