pith. sign in

arxiv: 2606.04555 · v1 · pith:FDB2P6OKnew · submitted 2026-06-03 · 💻 cs.CL · cs.AI

Temporal Order Matters for Agentic Memory: Segment Trees for Long-Horizon Agents

Yifan Simon Liu , Liam Gallagher , Faeze Moradi Kalarde , Jiazhou Liang , Armin Toroghi , Scott Sanner This is my paper

Pith reviewed 2026-06-28 06:11 UTC · model grok-4.3

classification 💻 cs.CL cs.AI
keywords segment tree memorytemporal orderlong-horizon agentsconversational memoryagentic memorymemory architectureonline insertion
0
0 comments X

The pith

A segment tree that inserts conversation turns in chronological order improves answer quality on long-horizon agent benchmarks.

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

The paper presents SegTreeMem, a memory architecture that builds a segment tree over utterances using an online rightmost-frontier insertion rule to keep events in their original time sequence. Retrieval works by propagating relevance scores down the tree so that local semantic matches are combined with hierarchical temporal context. Experiments on three long-horizon benchmarks with two different LLM backbones show higher answer quality than flat retrieval, graph memory, and other tree baselines. A follow-up test that randomly permutes temporal order during construction finds that the gains disappear, indicating that chronological structure is necessary for the observed improvement.

Core claim

SegTreeMem represents conversation history as a temporally ordered Segment Tree over utterances, incrementally inserts new utterances through an online rightmost-frontier update rule while forming hierarchical memory segments, and propagates relevance scores through the tree to combine local semantic matching with hierarchical temporal context; across three long-horizon memory benchmarks and two LLM backbones this yields higher answer quality than flat retrieval, graph-structured memory, and tree-structured memory baselines, and a temporal-order permutation analysis shows that the performance gain depends on preserving temporal order during memory construction.

What carries the argument

Segment tree with online rightmost-frontier update rule that preserves chronological order while forming hierarchical segments, plus relevance-score propagation for retrieval.

If this is right

  • Long-horizon agents can maintain coherent recall of evolving events without explicit time-stamping at query time.
  • Memory systems that ignore insertion order will underperform even when they use hierarchical or graph structures.
  • Online incremental construction is sufficient to produce usable temporal hierarchy without full offline reordering.
  • Retrieval that combines local match scores with ancestor context captures both topical and sequential relevance.

Where Pith is reading between the lines

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

  • The same insertion rule could be tested on non-conversational event streams such as sensor logs or code commit histories.
  • If temporal order is the dominant factor, simpler linear buffers with time-aware scoring might close much of the gap.
  • The approach suggests that agent memory benchmarks should include explicit order-permutation controls as a standard diagnostic.

Load-bearing premise

That the measured gains come from temporal-order preservation rather than from other differences in how the segment tree inserts items or scores them.

What would settle it

Running the same benchmarks after building the segment tree with the rightmost-frontier rule disabled or after randomly shuffling utterance order at insertion time and finding no drop in answer quality.

Figures

Figures reproduced from arXiv: 2606.04555 by Armin Toroghi, Faeze Moradi Kalarde, Jiazhou Liang, Liam Gallagher, Scott Sanner, Yifan Simon Liu.

Figure 1
Figure 1. Figure 1: Memory tree representations. Semantic trees may group non-consecutive utterances, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Online Segment Tree update. For a new utterance xt+1, the compatibility model selects a frontier node v3, to which a subtree with leaf xt+1 is attached. Online tree update via rightmost frontier. To select an attachment node, we use a compat￾ibility model, which either returns the non-leaf frontier node most compatible with the incom￾ing utterance xt+1, or indicates that no compat￾ible frontier node exists… view at source ↗
Figure 3
Figure 3. Figure 3: Two score propagation policies: top: top-down, where scores propagate from parents to [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Controlled comparison of tree construction strategies. We compare a non-temporal similarity [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Construction and retrieval efficiency as memory grows. We report per-utterance construc [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Effect of propagation policy and decay factor on retrieval quality. The dashed line [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Retrieved-node level distribution as a function of the propagation setting, using batch-LLM [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Whole-tree visualizations of SEGTREEMEM (top), MEMTREE [31] (middle), and RAP￾TOR [34] (bottom) on LoCoMo conv-47, which spans 329 utterances across 31 dialogues. Each leaf (filled circle) is one utterance; internal nodes are drawn as outlined circles. Both leaves and internal nodes are colored by source dialogue index. C.3 Adversarial Cases We analyze two adversarial input patterns for SEGTREEMEM’s online… view at source ↗
Figure 9
Figure 9. Figure 9: Degenerate cascading segment tree produced by maximally topic-switching input. Each [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Long-range topical recurrence. Two leaves on the same topic (purple) are separated by an [PITH_FULL_IMAGE:figures/full_fig_p021_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Q1 ground-truth answer-bearing utterance and ancestor nodes in the LoCoMo [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Effect of propagation direction and decay factor under the [PITH_FULL_IMAGE:figures/full_fig_p027_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Accuracy as a function of the decay factor [PITH_FULL_IMAGE:figures/full_fig_p029_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Retrieved-node level distribution as a function of the propagation setting, using batch-LLM [PITH_FULL_IMAGE:figures/full_fig_p030_14.png] view at source ↗
read the original abstract

Long-horizon conversational agents need to interact with users through evolving events, tasks, and goals. Such histories are naturally temporal, yet many existing memory systems organize information primarily by topical similarity and may ignore the order in which events occur. We introduce Segment Tree Memory, or SegTreeMem, a memory architecture that represents conversation history as a temporally ordered Segment Tree over utterances. SegTreeMem incrementally inserts new utterances through an online rightmost-frontier update rule, preserving chronological order while forming hierarchical memory segments. For retrieval, SegTreeMem propagates relevance scores through the tree to combine local semantic matching with hierarchical temporal context. Across three long-horizon memory benchmarks and two LLM backbones, SegTreeMem improves answer quality over flat retrieval, graph-structured memory, and tree-structured memory baselines. Additional temporal-order permutation analysis shows that the performance gain depends on preserving temporal order during memory construction, supporting the claim that temporal order is a key structure for agentic memory.

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.

Referee Report

1 major / 2 minor

Summary. The paper introduces SegTreeMem, a memory architecture that represents conversation history as a temporally ordered segment tree over utterances. New utterances are inserted incrementally via an online rightmost-frontier update rule that preserves chronological order while forming hierarchical segments; relevance scores are propagated through the tree at retrieval time to combine local semantic matching with hierarchical temporal context. Empirical results across three long-horizon benchmarks and two LLM backbones show improved answer quality relative to flat retrieval, graph-structured memory, and other tree-structured baselines. A temporal-order permutation analysis is presented to support the claim that performance gains depend on preserving temporal order during memory construction.

Significance. If the central empirical claims and the attribution to temporal order hold after addressing the noted methodological gap, the work would strengthen the case that chronological structure is a load-bearing inductive bias for agentic memory systems, distinct from topical similarity alone. The choice of a standard segment-tree data structure with an explicit online insertion rule is a clear methodological strength that could facilitate reproducibility.

major comments (1)
  1. [Permutation analysis] Permutation analysis (described in the abstract and presumably §4 or §5): randomly reordering utterances necessarily alters the rightmost-frontier insertion sequence and therefore the resulting hierarchical segment boundaries and tree topology. No control experiment is described that holds segment structure fixed while only permuting order (or vice versa), so the observed performance drop cannot be attributed solely to loss of chronological order rather than to disruption of the tree topology or scoring propagation rules. This directly affects the load-bearing claim that 'the performance gain depends on preserving temporal order.'
minor comments (2)
  1. The abstract states empirical improvements but supplies no numerical values, error bars, or statistical tests; the full results section should include these details with explicit baseline comparisons.
  2. Clarify the exact relevance-score propagation rule (e.g., how scores are combined across levels) with a worked example or pseudocode, as the current description leaves the interaction between local matching and hierarchical context underspecified.

Simulated Author's Rebuttal

1 responses · 0 unresolved

Thank you for the constructive feedback. We respond to the major comment on the permutation analysis and will revise the manuscript accordingly to address the methodological concern.

read point-by-point responses

  1. Referee: [Permutation analysis] Permutation analysis (described in the abstract and presumably §4 or §5): randomly reordering utterances necessarily alters the rightmost-frontier insertion sequence and therefore the resulting hierarchical segment boundaries and tree topology. No control experiment is described that holds segment structure fixed while only permuting order (or vice versa), so the observed performance drop cannot be attributed solely to loss of chronological order rather than to disruption of the tree topology or scoring propagation rules. This directly affects the load-bearing claim that 'the performance gain depends on preserving temporal order.'

    Authors: We thank the referee for highlighting this important point. The permutation analysis was intended to demonstrate the importance of chronological order in the construction process, which inherently determines the tree topology via the online insertion rule. However, we acknowledge that this does not fully disentangle the contribution of order from the resulting structure. To address this, we will include an additional control experiment in the revised manuscript where the segment tree topology is fixed based on the original temporal order, but the leaf utterances are permuted in content. This will allow us to assess the impact of order independently of topology changes. We believe this will strengthen the evidence for our central claim. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical claims rest on external benchmarks

full rationale

The paper introduces SegTreeMem via an online insertion rule and reports answer-quality gains on three external long-horizon benchmarks plus a temporal-order permutation test. No equations, fitted parameters, self-citations, or ansatzes are described that reduce any claimed result to an input by construction. The reported improvements and the dependence on temporal order are presented as measured outcomes on independent benchmarks rather than quantities defined in terms of the method itself; the derivation chain is therefore self-contained against external data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The segment tree itself is a standard structure repurposed here, so no new postulated entities are introduced in the provided text.

pith-pipeline@v0.9.1-grok · 5711 in / 1289 out tokens · 20897 ms · 2026-06-28T06:11:50.512956+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 27 canonical work pages · 13 internal anchors

  1. [1]

    Qwen3.5-Flash model documentation

    Alibaba Cloud. Qwen3.5-Flash model documentation. https://www.alibabacloud.com/ help/en/model-studio/getting-started/models, 2026. Alibaba Cloud Model Studio documentation forqwen3.5-flash; snapshotqwen3.5-flash-2026-02-23

    2026

  2. [2]

    Longbench: A bilingual, multitask benchmark for long context understanding

    Yushi Bai, Xin Lv, Jiajie Zhang, Hongchang Lyu, Jiankai Tang, Zhidian Huang, Zhengxiao Du, Xiao Liu, Aohan Zeng, Lei Hou, et al. Longbench: A bilingual, multitask benchmark for long context understanding. InProceedings of the 62nd annual meeting of the association for computational linguistics (volume 1: Long papers), pages 3119–3137, 2024

    2024

  3. [3]

    Realmem: Benchmarking llms in real-world memory-driven interaction.arXiv preprint arXiv:2601.06966, 2026

    Haonan Bian, Zhiyuan Yao, Sen Hu, Zishan Xu, Shaolei Zhang, Yifu Guo, Ziliang Yang, Xueran Han, Huacan Wang, and Ronghao Chen. Realmem: Benchmarking llms in real-world memory-driven interaction.arXiv preprint arXiv:2601.06966, 2026

    work page arXiv 2026

  4. [4]

    Walking down the memory maze: Beyond context limit through interactive reading.arXiv preprint arXiv:2310.05029, 2023

    Howard Chen, Ramakanth Pasunuru, Jason Weston, and Asli Celikyilmaz. Walking down the memory maze: Beyond context limit through interactive reading.arXiv preprint arXiv:2310.05029, 2023

    work page arXiv 2023

  5. [5]

    Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory

    Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Yadav. Mem0: Building production-ready ai agents with scalable long-term memory.arXiv preprint arXiv:2504.19413, 2025. Includes Mem0 and graph-enhanced Mem0g variants

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  6. [6]

    Springer, 3 edition, 2008

    Mark de Berg, Otfried Cheong, Marc van Kreveld, and Mark Overmars.Computational Geometry: Algorithms and Applications. Springer, 3 edition, 2008

    2008

  7. [7]

    McKeown, Eric Fosler-Lussier, and Hongyan Jing

    Michel Galley, Kathleen R. McKeown, Eric Fosler-Lussier, and Hongyan Jing. Discourse segmentation of multi-party conversation. InProceedings of the 41st Annual Meeting of the Association for Computational Linguistics, pages 562–569, Sapporo, Japan, July 2003. Association for Computational Linguistics. doi: 10.3115/1075096.1075167. URL https: //aclantholog...

    work page doi:10.3115/1075096.1075167 2003

  8. [8]

    Nilesh Gupta, Wei-Cheng Chang, Ngot Bui, Cho-Jui Hsieh, and Inderjit S. Dhillon. LLM-guided hierarchical retrieval.arXiv preprint arXiv:2510.13217, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  9. [9]

    J.; Shu, Y.; Gu, Y.; Yasunaga, M.; and Su, Y

    Bernal Jiménez Gutiérrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su. Hipporag: Neurobiologically inspired long-term memory for large language models.arXiv preprint arXiv:2405.14831, 2024

    work page arXiv 2024

  10. [10]

    Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions

    Yuanzhe Hu, Yu Wang, and Julian McAuley. Evaluating memory in llm agents via incremental multi-turn interactions.arXiv preprint arXiv:2507.05257, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  11. [11]

    Scaling personalized web search

    Glen Jeh and Jennifer Widom. Scaling personalized web search. InProceedings of the 12th International Conference on World Wide Web, pages 271–279. ACM, 2003

    2003

  12. [12]

    Shafiq Joty, Giuseppe Carenini, and Raymond T. Ng. Topic segmentation and labeling in asynchronous conversations.Journal of Artificial Intelligence Research, 47:521–573, 2013. doi: 10.1613/jair.3940. URLhttps://doi.org/10.1613/jair.3940

    work page doi:10.1613/jair.3940 2013

  13. [13]

    InFindings of the Conference on Empirical Methods in Natural Language Processing (EMNLP)

    Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering. InProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6769–6781, Online, November 2020. Association for Computational Lin- guistics. ...

    work page doi:10.18653/v1/2020.emnlp-main.550 2020

  14. [14]

    Bayesian Active Learning with Gaussian Processes Guided by LLM Relevance Scoring for Dense Passage Retrieval

    Junyoung Kim, Anton Korikov, Jiazhou Liang, Justin Cui, Yifan Simon Liu, Qianfeng Wen, Mark Zhao, and Scott Sanner. Bayesian active learning with gaussian processes guided by llm relevance scoring for dense passage retrieval.arXiv preprint arXiv:2604.17906, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  15. [15]

    Evaluating Scene-based In-Situ Item Labeling for Immersive Conversational Recommendation

    Jiazhou Liang, Yifan Simon Liu, David Guo, Minqi Sun, Yilun Jiang, and Scott Sanner. Evaluating scene-based in-situ item labeling for immersive conversational recommendation. arXiv preprint arXiv:2604.09698, 2026. 11

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  16. [16]

    Goal-Oriented Reasoning for RAG-based Memory in Conversational Agentic LLM Systems

    Jiazhou Liang, Armin Toroghi, Yifan Simon Liu, Faeze Moradi Kalarde, Liam Gallagher, and Scott Sanner. Goal-oriented reasoning for rag-based memory in conversational agentic llm systems.arXiv preprint arXiv:2605.12213, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  17. [18]

    G-eval: Nlg evaluation using gpt-4 with better human alignment

    Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. G-eval: Nlg evaluation using gpt-4 with better human alignment. InProceedings of the 2023 conference on empirical methods in natural language processing, pages 2511–2522, 2023

    2023

  18. [19]

    A compar- ative study of static and contextual embeddings for analyzing semantic changes in medieval latin charters

    Yifan Liu, Gelila Tilahun, Xinxiang Gao, Qianfeng Wen, and Michael Gervers. A compar- ative study of static and contextual embeddings for analyzing semantic changes in medieval latin charters. InProceedings of the First Workshop on Language Models for Low-Resource Languages, pages 182–192, 2025

    2025

  19. [20]

    Multimodal item scoring for natural language recommendation via gaussian process regression with llm relevance judgments.arXiv preprint arXiv:2510.22023, 2025

    Yifan Liu, Qianfeng Wen, Jiazhou Liang, Mark Zhao, Justin Cui, Anton Korikov, Armin Toroghi, Junyoung Kim, and Scott Sanner. Multimodal item scoring for natural language recommendation via gaussian process regression with llm relevance judgments.arXiv preprint arXiv:2510.22023, 2025

    work page arXiv 2025

  20. [21]

    Ma-dpr: Manifold- aware distance metrics for dense passage retrieval

    Yifan Liu, Qianfeng Wen, Mark Zhao, Jiazhou Liang, and Scott Sanner. Ma-dpr: Manifold- aware distance metrics for dense passage retrieval. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 31073–31091, 2025

    2025

  21. [22]

    Semantic xpath: Structured agentic memory access for conversational ai.arXiv preprint arXiv:2603.01160, 2026

    Yifan Simon Liu, Ruifan Wu, Liam Gallagher, Jiazhou Liang, Armin Toroghi, and Scott Sanner. Semantic xpath: Structured agentic memory access for conversational ai.arXiv preprint arXiv:2603.01160, 2026

    work page arXiv 2026

  22. [23]

    2504.08266,arXiv:2504.08266,doi:10.48550/ARXIV.2504.08266

    Xueguang Ma, Xinyu Zhang, Ronak Pradeep, and Jimmy Lin. Zero-shot listwise document reranking with a large language model.CoRR, abs/2305.02156, 2023. doi: 10.48550/arXiv. 2305.02156

    work page internal anchor Pith review doi:10.48550/arxiv 2023

  23. [24]

    Evaluating Very Long-Term Conversational Memory of LLM Agents

    Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. Evaluating very long-term conversational memory of llm agents.arXiv preprint arXiv:2402.17753, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  24. [25]

    GPT-4o mini model documentation

    OpenAI. GPT-4o mini model documentation. https://platform.openai.com/docs/ models/gpt-4o-mini, 2024. Official OpenAI API documentation for gpt-4o-mini; snap- shotgpt-4o-mini-2024-07-18

    2024

  25. [26]

    New embedding models and api updates

    OpenAI. New embedding models and api updates. https://openai. com/index/new-embedding-models-and-api-updates/ , 2024. Introduces text-embedding-3-small

    2024

  26. [27]

    Gpt-5.4 mini model documentation

    OpenAI. Gpt-5.4 mini model documentation. https://developers.openai.com/ api/docs/models/gpt-5.4-mini, 2026. Official OpenAI API documentation for gpt-5.4-mini

    2026

  27. [28]

    MemGPT: Towards LLMs as Operating Systems

    Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, and Joseph E. Gonzalez. Memgpt: Towards llms as operating systems.arXiv preprint arXiv:2310.08560, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  28. [29]

    The pagerank citation ranking: Bringing order to the web

    Lawrence Page, Sergey Brin, Rajeev Motwani, and Terry Winograd. The pagerank citation ranking: Bringing order to the web. Technical report, Stanford InfoLab, 1999

    1999

  29. [30]

    Bernstein

    Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Interactive simulacra of human behavior. InProceed- ings of the 36th Annual ACM Symposium on User Interface Software and Technology, UIST ’23, 2023. 12

    2023

  30. [31]

    From isolated conversations to hierarchi- cal schemas: Dynamic tree memory representation for llms.arXiv preprint arXiv:2410.14052, 2024

    Alireza Rezazadeh, Zichao Li, Wei Wei, and Yujia Bao. From isolated conversations to hierarchi- cal schemas: Dynamic tree memory representation for llms.arXiv preprint arXiv:2410.14052, 2024

    work page arXiv 2024

  31. [32]

    The probabilistic relevance framework: Bm25 and beyond.Foundations and Trends in Information Retrieval, 3(4):333–389, 2009

    Stephen Robertson and Hugo Zaragoza. The probabilistic relevance framework: Bm25 and beyond.Foundations and Trends in Information Retrieval, 3(4):333–389, 2009. doi: 10.1561/ 1500000019

    2009

  32. [33]

    Robertson, Steve Walker, Susan Jones, Micheline Hancock-Beaulieu, and Mike Gatford

    Stephen E. Robertson, Steve Walker, Susan Jones, Micheline Hancock-Beaulieu, and Mike Gatford. Okapi at trec-3. InText Retrieval Conference, 1994

    1994

  33. [34]

    Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh Khanna, Anna Goldie, and Christopher D. Manning. RAPTOR: Recursive abstractive processing for tree-organized retrieval.arXiv preprint arXiv:2401.18059, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  34. [35]

    Adith Swaminathan, Akshay Krishnamurthy, Alekh Agarwal, Miroslav Dudík, John Langford, Damien Jose, and Imed Zitouni

    Weiwei Sun, Lingyong Yan, Xinyu Ma, Shuaiqiang Wang, Pengjie Ren, Zhumin Chen, Dawei Yin, and Zhaochun Ren. Is ChatGPT good at search? investigating large language models as re-ranking agents. InProceedings of the 2023 Conference on Empirical Meth- ods in Natural Language Processing, pages 14918–14937, Singapore, December 2023. As- sociation for Computati...

    work page doi:10.18653/v1/2023.emnlp-main.923 2023

  35. [36]

    MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries

    Yixuan Tang and Yi Yang. Multihop-rag: Benchmarking retrieval-augmented generation for multi-hop queries.arXiv preprint arXiv:2401.15391, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  36. [37]

    Fast random walk with restart and its applications

    Hanghang Tong, Christos Faloutsos, and Jia-Yu Pan. Fast random walk with restart and its applications. InProceedings of the Sixth IEEE International Conference on Data Mining, pages 613–622, 2006

    2006

  37. [38]

    Elaborative subtopic query reformulation for broad and indirect queries in travel destination recommendation.arXiv preprint arXiv:2410.01598, 2024

    Qianfeng Wen, Yifan Liu, Joshua Zhang, George Saad, Anton Korikov, Yury Sambale, and Scott Sanner. Elaborative subtopic query reformulation for broad and indirect queries in travel destination recommendation.arXiv preprint arXiv:2410.01598, 2024

    work page arXiv 2024

  38. [39]

    A simple but effective elaborative query reformulation approach for natural language recommendation.arXiv preprint arXiv:2510.02656, 2025

    Qianfeng Wen, Yifan Liu, Justin Cui, Joshua Zhang, Anton Korikov, George-Kirollos Saad, and Scott Sanner. A simple but effective elaborative query reformulation approach for natural language recommendation.arXiv preprint arXiv:2510.02656, 2025

    work page arXiv 2025

  39. [40]

    LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory

    Di Wu, Hongwei Wang, Wenhao Yu, Yuwei Zhang, Kai-Wei Chang, and Dong Yu. Long- memeval: Benchmarking chat assistants on long-term interactive memory.arXiv preprint arXiv:2410.10813, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  40. [41]

    A-MEM: Agentic Memory for LLM Agents

    Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. A-mem: Agentic memory for llm agents.arXiv preprint arXiv:2502.12110, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  41. [42]

    Rank-without- GPT: Building GPT-independent listwise rerankers on open-source large language models

    Xinyu Zhang, Sebastian Hofstätter, Patrick Lewis, Raphael Tang, and Jimmy Lin. Rank-without- GPT: Building GPT-independent listwise rerankers on open-source large language models. CoRR, abs/2312.02969, 2023. doi: 10.48550/arXiv.2312.02969

    work page doi:10.48550/arxiv.2312.02969 2023

  42. [43]

    John advised him to practice first . . . using a gamepad and good tim- ing

    Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. Memorybank: Enhancing large language models with long-term memory.Proceedings of the AAAI Conference on Artificial Intelligence, 38(17), 2024. 13 A Algorithmic Details This subsection provides pseudocode for the two SEGTREEMEMoperations. Node annotations are denoted byA(v), node intervals byI(...

    work page arXiv 2024

  43. [44]

    The question category (1=single-hop, 2=temporal, 3=open-ended, 4=multi-hop, 5=adversarial)

  44. [45]

    The gold answer (or adversarial answer for category 5)

  45. [46]

    For category 5, the candidate provides the adversarial (wrong) answer as if it were true

    The candidate answer to evaluate Scoring criteria: Score 0: The candidate answer is incorrect or contradicts the gold answer. For category 5, the candidate provides the adversarial (wrong) answer as if it were true. Score 1: The candidate answer is vague or generic, not using specific information from the conversation. For category 5, the candidate is unc...

  46. [47]

    The user’s current query

  47. [48]

    The user-related memory, representing the latest valid user state

  48. [49]

    A reference answer based on the relevant memory

  49. [50]

    facts, constraints, preferences, and confirmed states

    The candidate answer to be evaluated Please follow these rules during evaluation: - Focus only on whether "facts, constraints, preferences, and confirmed states" are correctly used - Do NOT evaluate language style, tone, politeness, empathy, or fluency - Do NOT give a high score just because the answer "sounds reasonable" - The reference answer is only to...

  50. [51]

    Answer: [Yes] Justification: We use LLMs as core components of both our method and evaluation

    Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals (or an equivalent approval/review based on the requirements of your country or ...