Chronological Knowledge Retrieval: A Retrieval-Augmented Generation Approach to Construction Project Documentation

Fran\c{c}ois Denis; Ioannis-Aris Kostis; Natalia Sanchiz; Pierre Schaus; Steeve De Schryver

arxiv: 2604.14169 · v1 · submitted 2026-03-25 · 💻 cs.CL

Chronological Knowledge Retrieval: A Retrieval-Augmented Generation Approach to Construction Project Documentation

Ioannis-Aris Kostis , Natalia Sanchiz , Steeve De Schryver , Fran\c{c}ois Denis , Pierre Schaus This is my paper

Pith reviewed 2026-05-15 01:01 UTC · model grok-4.3

classification 💻 cs.CL

keywords Retrieval-Augmented GenerationRAGMeeting MinutesConstruction DocumentationChronological RetrievalSemantic SearchNatural Language Queries

0 comments

The pith

A retrieval-augmented generation system lets professionals query construction meeting minutes in natural language while receiving explicitly time-annotated answers.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to show that a RAG pipeline can convert large archives of project meeting minutes into a conversational resource that surfaces the chronological sequence of decisions. Manual reconstruction of decision histories is slow and prone to error because later minutes routinely override earlier ones. The authors argue that semantic search followed by LLM generation can deliver both relevance and correct temporal ordering at once. They test the idea on a real, anonymized Belgian construction dataset that has been enriched with expert queries. Both the data and the code are released so others can reproduce or extend the approach.

Core claim

The central claim is that combining semantic retrieval with large language models in a RAG framework produces answers that are semantically relevant to user questions about construction decisions and that are explicitly annotated with the dates and order of the underlying minutes, thereby allowing users to trace how decisions evolved and overrode one another.

What carries the argument

A RAG framework that pairs semantic search over meeting-minute embeddings with an LLM generator to produce time-stamped, context-aware responses.

If this is right

Professionals gain conversational access to the full history of project choices without manual archive searches.
Answers carry explicit time stamps, making it easier to identify which decisions are still current.
The same pipeline can be applied to any collection of dated documents that record evolving decisions.
Releasing the dataset and code enables direct replication and comparison with other retrieval methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same chronological-retrieval pattern could be tested on legal case files or medical progress notes where later entries supersede earlier ones.
If the approach scales, project-management software could embed live query interfaces over meeting archives rather than static search.
Future work could measure how often the LLM still requires post-editing to correct subtle temporal misreads.

Load-bearing premise

Semantic similarity search plus an LLM will reliably recover the correct chronological sequence of decisions and surface overrides without hallucinating or dropping key context.

What would settle it

Run the system on the released expert-annotated queries and check whether the generated answers list decisions in correct date order and correctly flag overrides; any systematic mismatch in ordering or missed overrides would falsify the claim.

Figures

Figures reproduced from arXiv: 2604.14169 by Fran\c{c}ois Denis, Ioannis-Aris Kostis, Natalia Sanchiz, Pierre Schaus, Steeve De Schryver.

**Figure 1.** Figure 1: Example of system output. System response to the query “Could I have a list of the remarks made by SECO?". Query and output(s) are both translated from French. The contributions of this work are as follows: First, we describe the generalized, time-aware RAG architecture designed to extract chronologically structured information from sequential project records. Second, we demonstrate the system’s reliabilit… view at source ↗

**Figure 2.** Figure 2: Dataset composition and temporal statistics. (a)-(c) Summarize document structure: page, passage, and word counts. (d) Illustrates the temporal distribution of documents over the project’s time span. 4.3. Subfloor sprinkler pipe installation: 23/11/23 TS asks to check the regulations to see if the installation of the sprinkler pipes is not too far from the ceiling. The sprinkler heads must be very close to… view at source ↗

**Figure 3.** Figure 3: Evolution of a subject in the meeting minutes. The initial decision is amended over meetings and truncated for brevity toward the end of the project. Decision text translated from French. 5 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: System implementation overview. Schematic representation of the implemented system designed to meet the use case specifications. Dashed lines denote offline operations, while solid lines represent online processes. Modules enclosed within the gray dotted frame are optional and not essential for core system functionality. 6 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: System Graphical User Interface. Screenshot of the application interface showing the bottom input box for user queries, the left panel with the decision timeline and system replies, and the right panel displaying retrieved source pages grouped by time span and document. scope of the present study, we focus in this Section on an evaluation based on measurable performance metrics, which provide a more objec… view at source ↗

**Figure 6.** Figure 6: Query - document relevance heatmap. Heatmap showing the distribution of pages containing relevant passages for each query in the benchmarking set. Rows correspond to evaluated queries, and columns represent documents ordered chronologically (dates shown on the top axis). Color intensity indicates the number of relevant pages cited per document. corresponding pages, with their order (ascending, based on the… view at source ↗

**Figure 7.** Figure 7: Retrieval performance metrics at query and global levels. Evaluation of retrieval metrics at different cutoff values keval = 2, 3, 4, 5, comparing system performance across queries and in aggregate. Finally, beyond the scope of our case study and towards more generalizable system architectures, several modular improvements can also be envisioned. A first step would be to disassociate the explicit time-st… view at source ↗

**Figure 8.** Figure 8: System output comparison for different batch sizes. Comparison of systemgenerated outputs for batch sizes nbatch = 6, 10, and 60. Each sample reply illustrates how varying the batch size affects the completeness and consistency of the retrieved and synthesized responses. References [1] Reciprocal rank fusion outperforms condorcet and individual rank learning methods, 2009. [2] Dense Passage Retrieval for … view at source ↗

read the original abstract

In large-scale construction projects, the continuous evolution of decisions generates extensive records, most often captured in meeting minutes. Since decisions may override previous ones, professionals often need to reconstruct the history of specific choices. Retrieving such information manually from raw archives is both labor-intensive and error-prone. From a user perspective, we address this challenge by enabling conversational access to the whole set of project meeting minutes. Professionals can pose natural-language questions and receive answers that are both semantically relevant and explicitly time-annotated, allowing them to follow the chronology of decisions. From a technical perspective, our solution employs a Retrieval-Augmented Generation (RAG) framework that integrates semantic search with large language models to ensure accurate and context-aware responses. We demonstrate the approach using an anonymized, industry-sourced dataset of meeting minutes from a completed construction project by a large company in Belgium. The dataset is annotated and enriched with expert-defined queries to support systematic evaluation. Both the dataset and the open-source implementation are made available to the community to foster further research on conversational access to time-annotated project documentation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Applies standard RAG to construction meeting minutes with a chronology focus and releases an annotated dataset, but the abstract shows no metrics or mechanism for enforcing temporal order.

read the letter

The main takeaway is that this paper takes an off-the-shelf RAG pipeline and points it at a concrete industry problem: letting users query the evolving history of decisions in construction project meeting minutes. They work with an anonymized set from a real Belgian project, add expert queries, and release both the data and the code. That release is the part that actually adds value beyond the usual RAG description. Construction records are full of overrides and time-sensitive choices, so a conversational interface that surfaces chronology could save real time if it works. Making the dataset public gives other groups something concrete to test against, which is more useful than another generic framework paper. The approach itself is not new technically; it combines semantic search with an LLM for generation and time annotation. The soft spot is exactly what the stress-test note flags. Semantic retrieval ranks by embedding similarity, not by timestamp or causal order, so the system relies on the LLM to reconstruct the correct sequence and catch later overrides from whatever passages get pulled in. The abstract gives no numbers on accuracy, no error analysis on missed overrides or hallucinations, and no description of any explicit temporal indexing or filtering step. Without those details it is impossible to tell whether the central claim holds. This is aimed at applied researchers or practitioners who need domain-specific retrieval examples rather than core method advances. People working on similar document streams in law or engineering might find the released data worth looking at. The work is coherent and engages honestly with the workflow problem, even if the evaluation is missing at this stage. I would bring it to a reading group as a maybe, mainly for the dataset. I would not cite it yet because there are no results to reference. It deserves peer review so the authors can supply the quantitative checks on chronology handling.

Referee Report

2 major / 0 minor

Summary. The paper presents a RAG framework that combines semantic search with LLMs to enable natural-language queries over construction project meeting minutes, producing responses that are both semantically relevant and explicitly time-annotated so users can follow decision chronology and overrides. It demonstrates the approach on an anonymized industry dataset of meeting minutes enriched with expert queries and releases both the dataset and open-source implementation.

Significance. If the pipeline reliably surfaces correct chronological order and detects overrides, the work would offer a practical tool for domain-specific document retrieval in project management and support further research through the released dataset and code.

major comments (2)

Abstract: the claim that the RAG framework 'ensures accurate and context-aware responses' with explicit time annotation lacks any quantitative metrics, error analysis, or ablation results, leaving the central chronological-retrieval claim unsupported by evidence.
Abstract: the architecture is described as standard semantic search followed by LLM generation; no mechanism is specified for enforcing temporal ordering, detecting overrides, or guaranteeing that retrieved passages contain the full relevant timeline, which is load-bearing for the paper's main contribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and indicate the planned revisions.

read point-by-point responses

Referee: Abstract: the claim that the RAG framework 'ensures accurate and context-aware responses' with explicit time annotation lacks any quantitative metrics, error analysis, or ablation results, leaving the central chronological-retrieval claim unsupported by evidence.

Authors: We agree that the abstract phrasing is too strong given the evidence presented. The manuscript demonstrates the framework on an industry dataset enriched with expert queries to enable systematic evaluation, but does not report specific quantitative metrics, error analysis, or ablations in the abstract. We will revise the abstract to moderate the language (e.g., replace 'ensures' with 'supports') and explicitly reference the evaluation approach and key findings from the experiments section. revision: yes
Referee: Abstract: the architecture is described as standard semantic search followed by LLM generation; no mechanism is specified for enforcing temporal ordering, detecting overrides, or guaranteeing that retrieved passages contain the full relevant timeline, which is load-bearing for the paper's main contribution.

Authors: The current description relies on timestamp metadata attached to meeting-minute passages during indexing, so that semantic retrieval surfaces time-stamped content and the LLM prompt instructs the model to produce chronologically ordered, time-annotated answers. No dedicated post-retrieval step enforces ordering or detects overrides. We acknowledge the description is insufficiently explicit. We will revise the abstract and expand the methods section to detail the temporal handling, state the assumptions about timeline completeness, and note the absence of explicit override-detection logic. revision: yes

Circularity Check

0 steps flagged

No circularity; standard RAG application on external components

full rationale

The paper describes a conventional Retrieval-Augmented Generation pipeline that combines semantic vector search over meeting-minute embeddings with an LLM for generating time-annotated answers. No equations, parameter-fitting steps, uniqueness theorems, or ansatzes are introduced; the central claim simply invokes established external RAG primitives (semantic similarity ranking followed by LLM synthesis) without any reduction of outputs to inputs by construction. Dataset release and open-source code are offered as evaluation artifacts, but these do not create self-referential loops. The architecture therefore remains self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review reveals no explicit free parameters, new entities, or non-standard axioms beyond the general assumption that standard RAG pipelines can handle chronological project records accurately.

axioms (1)

domain assumption Semantic search plus LLM generation produces accurate time-annotated responses from meeting minutes
Invoked in the description of the RAG framework as the basis for context-aware chronological answers

pith-pipeline@v0.9.0 · 5504 in / 1098 out tokens · 48462 ms · 2026-05-15T01:01:20.660292+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

our solution employs a Retrieval-Augmented Generation (RAG) framework that integrates semantic search with large language models
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

performing an independent retrieval for each discrete time-span

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

[1]

Reciprocal rank fusion outperforms condorcet and individu al rank learning methods , 2009

work page 2009
[2]

Dense Passage Retrieval for Open-Domain Question Answerin g., 2020

work page 2020
[3]

A Retrieval-Augmented Framework For Meeting Insight Extra ction, 2025. 22

work page 2025
[4]

Tempretriever: Fusion-based temporal dense passage retri eval for time-sensitive questions

Abdelrahman Abdallah, Bhawna Piryani, Jonas Wallat, A vi shek Anand, and Adam Jatowt. Tempretriever: Fusion-based temporal dense passage retri eval for time-sensitive questions. arXiv preprint arXiv:2502.21024 , 2025

work page arXiv 2025
[5]

Autom eet: a proof-of-concept study of genai to automate meetings in automotive engineering

Simon Baeuerle, Max Radyschevski, and Ulrike Pado. Autom eet: a proof-of-concept study of genai to automate meetings in automotive engineering. arXiv preprint arXiv:2507.16054 , 2025

work page arXiv 2025
[6]

Meet2mitigate: An llm-powered framework for real-time iss ue identiﬁcation and mitiga- tion from construction meeting discourse

Gongfan Chen, Abdullah Alsharef, Anto Ovid, Alex Albert , and Edward Jaselskis. Meet2mitigate: An llm-powered framework for real-time iss ue identiﬁcation and mitiga- tion from construction meeting discourse. Advanced Engineering Informatics , 64:103068, 2025

work page 2025
[7]

The llama 3 herd of models

Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhis hek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angel a Fan, et al. The llama 3 herd of models. arXiv e-prints , pages arXiv–2407, 2024

work page 2024
[8]

It’s About Time: Incorporating Temporality in Retrieval Au gmented Language Mod- els, 2025

IEEE. It’s About Time: Incorporating Temporality in Retrieval Au gmented Language Mod- els, 2025

work page 2025
[9]

Hybrid large language model ap proach for prompt and sensitive defect management: A comparative analysis of hyb rid, non-hybrid, and graphrag approaches

Kahyun Jeon and Ghang Lee. Hybrid large language model ap proach for prompt and sensitive defect management: A comparative analysis of hyb rid, non-hybrid, and graphrag approaches. Advanced Engineering Informatics , 64:103076, 2025

work page 2025
[10]

Evaluat ing answer reranking strate- gies in time-sensitive question answering

Mehmet Kardan, Bhawna Piryani, and Adam Jatowt. Evaluat ing answer reranking strate- gies in time-sensitive question answering. arXiv preprint arXiv:2503.04972 , 2025

work page arXiv 2025
[11]

Read- ing between the timelines: Rag for answering diachronic que stions

Kwun Hang Lau, Ruiyuan Zhang, Weijie Shi, Xiaofang Zhou , and Xiaojun Cheng. Read- ing between the timelines: Rag for answering diachronic que stions. arXiv preprint arXiv:2507.22917, 2025

work page arXiv 2025
[12]

Performance comparison of retrieval-augmented generation and ﬁne-tuned large lan guage models for construction safety management knowledge retrieval

Jungwon Lee, Seungjun Ahn, Daeho Kim, and Dongkyun Kim. Performance comparison of retrieval-augmented generation and ﬁne-tuned large lan guage models for construction safety management knowledge retrieval. Automation in Construction , 168:105846, 2024

work page 2024
[13]

TimeR4 : Time-aware retrieval-augmented large language models fo r temporal knowledge graph question answering

Xinying Qian, Ying Zhang, Yu Zhao, Baohang Zhou, Xuhui Su i, Li Zhang, and Kehui Song. TimeR4 : Time-aware retrieval-augmented large language models fo r temporal knowledge graph question answering. In Yaser Al-Onaizan, Mohit Bansal , and Yun-Nung Chen, edi- tors, Proceedings of the 2024 Conference on Empirical Methods in Natural Language Process- ing, pa...

work page 2024
[14]

Ranking models for the temporal dimension of text

Stefano Giovanni Rizzo, Matteo Brucato, and Danilo Mont esi. Ranking models for the temporal dimension of text. ACM Transactions on Information Systems , 41(2):1–34, 2022

work page 2022
[15]

The probabili stic relevance framework: Bm25 and beyond

Stephen Robertson, Hugo Zaragoza, et al. The probabili stic relevance framework: Bm25 and beyond. Foundations and Trends® in Information Retrieval , 3(4):333–389, 2009

work page 2009
[16]

Colbertv2: Effective and efficient retrieval via lightweight late interaction

Keshav Santhanam, Omar Khattab, Jon Saad-Falcon, Chri stopher Potts, and Matei Za- haria. Colbertv2: Eﬀective and eﬃcient retrieval via light weight late interaction. arXiv preprint arXiv:2112.01488 , 2021

work page arXiv 2021
[17]

Mrag: A modular retrieval framework for time-sensitive que stion answering

Zhang Siyue, Xue Yuxiang, Zhang Yiming, Wu Xiaobao, Luu Anh Tuan, and Zhao Chen. Mrag: A modular retrieval framework for time-sensitive que stion answering. arXiv preprint arXiv:2412.15540, 2024. 23

work page arXiv 2024
[18]

Dyg-rag: Dynamic graph retrieval-augmente d generation with event- centric reasoning

Qingyun Sun, Jiaqi Yuan, Shan He, Xiao Guan, Haonan Yuan , Xingcheng Fu, Jianxin Li, and Philip S Yu. Dyg-rag: Dynamic graph retrieval-augmente d generation with event- centric reasoning. arXiv preprint arXiv:2507.13396 , 2025

work page arXiv 2025
[19]

Improv ing knowledge management in building engineering with hybrid retrieval-augmented g eneration framework

Zhiqi Wang, Zhongcun Liu, Weizhen Lu, and Lu Jia. Improv ing knowledge management in building engineering with hybrid retrieval-augmented g eneration framework. Journal of Building Engineering, 103:112189, 2025

work page 2025
[20]

Retrieval augmented generation-driven i nformation retrieval and ques- tion answering in construction management

Chengke Wu, Wenjun Ding, Qisen Jin, Junjie Jiang, Rui Sh iping Jiang, Qinge Xiao, Longhui Liao, and Xiao Li. Retrieval augmented generation-driven i nformation retrieval and ques- tion answering in construction management. Advanced Engineering Informatics, 65:103158, 2025

work page 2025
[21]

Judg ing llm-as-a-judge with mt-bench and chatbot arena

Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhua ng, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. Judg ing llm-as-a-judge with mt-bench and chatbot arena. Advances in neural information processing systems , 36:46595– 46623, 2023. 24

work page 2023

[1] [1]

Reciprocal rank fusion outperforms condorcet and individu al rank learning methods , 2009

work page 2009

[2] [2]

Dense Passage Retrieval for Open-Domain Question Answerin g., 2020

work page 2020

[3] [3]

A Retrieval-Augmented Framework For Meeting Insight Extra ction, 2025. 22

work page 2025

[4] [4]

Tempretriever: Fusion-based temporal dense passage retri eval for time-sensitive questions

Abdelrahman Abdallah, Bhawna Piryani, Jonas Wallat, A vi shek Anand, and Adam Jatowt. Tempretriever: Fusion-based temporal dense passage retri eval for time-sensitive questions. arXiv preprint arXiv:2502.21024 , 2025

work page arXiv 2025

[5] [5]

Autom eet: a proof-of-concept study of genai to automate meetings in automotive engineering

Simon Baeuerle, Max Radyschevski, and Ulrike Pado. Autom eet: a proof-of-concept study of genai to automate meetings in automotive engineering. arXiv preprint arXiv:2507.16054 , 2025

work page arXiv 2025

[6] [6]

Meet2mitigate: An llm-powered framework for real-time iss ue identiﬁcation and mitiga- tion from construction meeting discourse

Gongfan Chen, Abdullah Alsharef, Anto Ovid, Alex Albert , and Edward Jaselskis. Meet2mitigate: An llm-powered framework for real-time iss ue identiﬁcation and mitiga- tion from construction meeting discourse. Advanced Engineering Informatics , 64:103068, 2025

work page 2025

[7] [7]

The llama 3 herd of models

Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhis hek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angel a Fan, et al. The llama 3 herd of models. arXiv e-prints , pages arXiv–2407, 2024

work page 2024

[8] [8]

It’s About Time: Incorporating Temporality in Retrieval Au gmented Language Mod- els, 2025

IEEE. It’s About Time: Incorporating Temporality in Retrieval Au gmented Language Mod- els, 2025

work page 2025

[9] [9]

Hybrid large language model ap proach for prompt and sensitive defect management: A comparative analysis of hyb rid, non-hybrid, and graphrag approaches

Kahyun Jeon and Ghang Lee. Hybrid large language model ap proach for prompt and sensitive defect management: A comparative analysis of hyb rid, non-hybrid, and graphrag approaches. Advanced Engineering Informatics , 64:103076, 2025

work page 2025

[10] [10]

Evaluat ing answer reranking strate- gies in time-sensitive question answering

Mehmet Kardan, Bhawna Piryani, and Adam Jatowt. Evaluat ing answer reranking strate- gies in time-sensitive question answering. arXiv preprint arXiv:2503.04972 , 2025

work page arXiv 2025

[11] [11]

Read- ing between the timelines: Rag for answering diachronic que stions

Kwun Hang Lau, Ruiyuan Zhang, Weijie Shi, Xiaofang Zhou , and Xiaojun Cheng. Read- ing between the timelines: Rag for answering diachronic que stions. arXiv preprint arXiv:2507.22917, 2025

work page arXiv 2025

[12] [12]

Performance comparison of retrieval-augmented generation and ﬁne-tuned large lan guage models for construction safety management knowledge retrieval

Jungwon Lee, Seungjun Ahn, Daeho Kim, and Dongkyun Kim. Performance comparison of retrieval-augmented generation and ﬁne-tuned large lan guage models for construction safety management knowledge retrieval. Automation in Construction , 168:105846, 2024

work page 2024

[13] [13]

TimeR4 : Time-aware retrieval-augmented large language models fo r temporal knowledge graph question answering

Xinying Qian, Ying Zhang, Yu Zhao, Baohang Zhou, Xuhui Su i, Li Zhang, and Kehui Song. TimeR4 : Time-aware retrieval-augmented large language models fo r temporal knowledge graph question answering. In Yaser Al-Onaizan, Mohit Bansal , and Yun-Nung Chen, edi- tors, Proceedings of the 2024 Conference on Empirical Methods in Natural Language Process- ing, pa...

work page 2024

[14] [14]

Ranking models for the temporal dimension of text

Stefano Giovanni Rizzo, Matteo Brucato, and Danilo Mont esi. Ranking models for the temporal dimension of text. ACM Transactions on Information Systems , 41(2):1–34, 2022

work page 2022

[15] [15]

The probabili stic relevance framework: Bm25 and beyond

Stephen Robertson, Hugo Zaragoza, et al. The probabili stic relevance framework: Bm25 and beyond. Foundations and Trends® in Information Retrieval , 3(4):333–389, 2009

work page 2009

[16] [16]

Colbertv2: Effective and efficient retrieval via lightweight late interaction

Keshav Santhanam, Omar Khattab, Jon Saad-Falcon, Chri stopher Potts, and Matei Za- haria. Colbertv2: Eﬀective and eﬃcient retrieval via light weight late interaction. arXiv preprint arXiv:2112.01488 , 2021

work page arXiv 2021

[17] [17]

Mrag: A modular retrieval framework for time-sensitive que stion answering

Zhang Siyue, Xue Yuxiang, Zhang Yiming, Wu Xiaobao, Luu Anh Tuan, and Zhao Chen. Mrag: A modular retrieval framework for time-sensitive que stion answering. arXiv preprint arXiv:2412.15540, 2024. 23

work page arXiv 2024

[18] [18]

Dyg-rag: Dynamic graph retrieval-augmente d generation with event- centric reasoning

Qingyun Sun, Jiaqi Yuan, Shan He, Xiao Guan, Haonan Yuan , Xingcheng Fu, Jianxin Li, and Philip S Yu. Dyg-rag: Dynamic graph retrieval-augmente d generation with event- centric reasoning. arXiv preprint arXiv:2507.13396 , 2025

work page arXiv 2025

[19] [19]

Improv ing knowledge management in building engineering with hybrid retrieval-augmented g eneration framework

Zhiqi Wang, Zhongcun Liu, Weizhen Lu, and Lu Jia. Improv ing knowledge management in building engineering with hybrid retrieval-augmented g eneration framework. Journal of Building Engineering, 103:112189, 2025

work page 2025

[20] [20]

Retrieval augmented generation-driven i nformation retrieval and ques- tion answering in construction management

Chengke Wu, Wenjun Ding, Qisen Jin, Junjie Jiang, Rui Sh iping Jiang, Qinge Xiao, Longhui Liao, and Xiao Li. Retrieval augmented generation-driven i nformation retrieval and ques- tion answering in construction management. Advanced Engineering Informatics, 65:103158, 2025

work page 2025

[21] [21]

Judg ing llm-as-a-judge with mt-bench and chatbot arena

Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhua ng, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. Judg ing llm-as-a-judge with mt-bench and chatbot arena. Advances in neural information processing systems , 36:46595– 46623, 2023. 24

work page 2023