Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 28 of the 28 references it cites were found in a registry or an index; 6 of the 11 sources its claims cite were checked for support.

'My agent understands me better': Integrating Dynamic Human-like Memory Recall and Consolidation in LLM-Based Agents

Open access
Hou, Yuki, Tamoto, Haruki, Miyashita, Homei · 2024 · Extended Abstracts of the CHI Conference on Human Factors in Computing Systems

Abstract

In this study, we propose a novel human-like memory architecture designed for enhancing the cognitive abilities of large language model (LLM)-based dialogue agents. Our proposed architecture enables agents to autonomously recall memories necessary for response generation, effectively addressing a limitation in the temporal cognition of LLMs. We adopt the human memory cue recall as a trigger for accurate and efficient memory recall. Moreover, we developed a mathematical model that dynamically quantifies memory consolidation, considering factors such as contextual relevance, elapsed time, and recall frequency. The agent stores memories retrieved from the user’s interaction history in a database that encapsulates each memory’s content and temporal context. Thus, this strategic storage allows agents to recall specific memories and understand their significance to the user in a temporal context, similar to how humans recognize and recall past experiences.

Overview

CoChat Summary

## TL;DR

This CHI EA 2024 paper proposes an LLM-agent memory architecture that combines human-like cued recall with a consolidation model based on relevance, elapsed time, and recall frequency. In the reported evaluation, the proposed model outperformed a Generative Agent baseline on recall accuracy, while also showing promising personalized response generation.

## Key findings

  • The model’s recall accuracy had a statistically significant lower loss than the Generative Agent baseline, with t = -5.687 and p = 0.000299; the 95% confidence interval for the mean difference was [-0.27, -0.12].
  • The authors say they built a dataset of 10 tasks from actual conversational histories, and also ran a qualitative dialogue study with 6 participants over one week to three months.
  • In the response-generation examples, the agent used recalled memories to produce personalized replies, such as reminding a user that “Ice cream awaits” and responding to a Hawaii mention by referencing a paper deadline in Hawaii on 1/25.
  • The paper reports a trigger threshold of 0.86 as appropriate in trials, though it says further research is needed to determine the best threshold.

## Methods

  • The paper proposes an LLM-based dialogue agent architecture that stores episodic memories from user dialogues in a database and computes recall via a mathematical model using event relevance, elapsed time, and recall frequency.
  • For evaluation, the authors built a system in Python using GPT-4-0613 as the baseline model, Qdrant for vector search/memory triggering, and Firestore for chat history management.
  • Quantitative analysis compared the proposed model against Generative Agents on 10 tasks using a loss based on normalized scores, Softmax probabilities, and mean squared error; qualitative examples were taken from participant chat histories.

## Limitations

  • The paper explicitly states that the method relies on users’ long-term behavioral patterns, so it may be less adaptable when behavior changes significantly, such as after starting a new job or school.
  • The authors note that the evaluation focused on specific domains and tasks, and that generalizability across other domains and dialogue tasks remains to be tested.
  • The text makes evident that the paper reports only a small-scale evaluation (10 tasks and 6 participants); this limits how far the reported results can be generalized.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3613905.3650839

Cites

28 of 28 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Image 1
From the PDF
Figure 2 (A) Decline in recall probability over time. The black curve ( 𝑟 =1, 𝑔 =1) shows a rapid loss of recall, while the red curve ( 𝑟 =0.6, 𝑔 =2) represents a slower forgetting rate. This difference indicates the challenge in designing dialogue agents that must distinguish between recent and distant events. (B) At time 𝑡 0 , Event D is recalled by the user, and the model updates its temporal significance. This exemplifies how memory is reinforced through repetition, becoming less susceptible to forgetting at 𝑡 0 + 𝑡 .
Figure 2 (A) Decline in recall probability over time. The black curve ( 𝑟 =1, 𝑔 =1) shows a rapid loss of recall, while the red curve ( 𝑟 =0.6, 𝑔 =2) represents a slower forgetting rate. This difference indicates the challenge in designing dialogue agents that must distinguish between recent and distant events. (B) At time 𝑡 0 , Event D is recalled by the user, and the model updates its temporal significance. This exemplifies how memory is reinforced through repetition, becoming less susceptible to forgetting at 𝑡 0 + 𝑡 .From the PDF
Figure 3 (A) Comparison of Loss Values Between two Models Across Different Tasks (B) Verification of Significance of Results
Figure 3 (A) Comparison of Loss Values Between two Models Across Different Tasks (B) Verification of Significance of ResultsFrom the PDF

Verification

Claims the paper makes

6 Valid · 3 Not valid · 5 Not checked · 38 Own results and interpretations

17 more claims

Create a free account to see all 52 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 14
  • Probably not a substantial claim. 21

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 16 references

Research works printed but not linked: 2

  1. S.D.L.R.S.P.P.U. California. 1987. Memory and Brain. Oxford University Press USA. https://books.google.co.jp/books?id=WH-HF5E9XSsC
  2. Atsushi Yamadori. 2002. Frontiers of Human Memory : a collection of contributions based on lectures presented at Internationl Symposium, Sendai, Japan, October 25-27, 2001. Tohoku University Press. https://ci.nii.ac.jp/ncid/BA57511014

Cited by

See all 38

Metadata

Authors
Hou, Y., Tamoto, H., & Miyashita, H.
Affiliations
  • Meiji University (JP)
  • Kyoto University (JP)
Venue
Extended Abstracts of the CHI Conference on Human Factors in Computing Systems
Publisher
ACM
Published
2024-05-11
Pages
1-7
OpenAlex
W4393905041
Cited by
31
References
16
Citations by year
2026: 5 · 2025: 22 · 2024: 4
FWCI
9.11
Type
conference paper
Language
en
Open access
Locations
  • CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) (publishedVersion)
  • arXiv (Cornell University) (submittedVersion, cc-by-nc-sa)
  • arXiv (Cornell University)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
  • Physical Sciences › Computer Science › Artificial Intelligence › AI-based Problem Solving and Planning
  • Physical Sciences › Computer Science › Artificial Intelligence › Psychiatry, Mental Health, Neuroscience
Keywords
memory retrieval, memory consolidation, temporal cognition, in-memory databases, interaction history, response generation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: 'My agent understands me better': Integrating Dynamic Human-like Memory Recall and Consolidation in LLM-Based Agents. https://app.cochat.ai/papers/10.1145/3613905.3650839
The paper
Hou, Y., Tamoto, H., & Miyashita, H. (2024). 'My agent understands me better': Integrating Dynamic Human-like Memory Recall and Consolidation in LLM-Based Agents. Extended Abstracts of the CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3613905.3650839

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.