Cao Zihang - Blog

【阅】本周阅读摘选2026-02-16 → 2026-02-22

Mon, 23 Feb 2026 00:00:00 +0000

本周阅读摘选 2026-02-16 → 2026-02-22 目录

技术技巧

技术技巧

Python & Google丨利用LLM从非结构化数文本中提取结构化数据

前两年自己工具里写的正则提取规则实在是写得累，这下子方便了

特性

自动分块
自动并发
自动对齐
自动去重

项目信息

DOI: 10.5281/zenodo.17015089
License: Apache-2.0

字节跳动技术团队丨当 OpenClaw 内置 LanceDB：为个人 AI 智能体打造超强长期记忆

Agent Runner为最关键的组件

Durable Knowledge 关于知识或用户偏好的持久化记忆
- 工作区：~/.openclaw/workspace
- 精选记忆：MEMORY.md精选长期记忆，仅在私聊时加载
- 每日记忆：memory/YYYY-MM-DD.md
- 索引存储：~/.openclaw/memory/<agentId>.sqlite
- QMD备选：~/.openclaw/agents/<agentId>/qmd/... 存放QMD后端的索引与配置
Task Memory 长周期、多步骤任务的中间记忆
- memory/YYY-MM-DD[-slug].md hook生成的特定任务或会话的快照 -Conversational History 完整对话历史和模型理解的隐式意图
- 状态目录：~/.openclaw/agents/<agentId>/sessions/
- sessions.json：会话元数据 (sessionId, token技术, memoryFlush状态)
- .jsonL：仅追加的JSONL格式记录完整对话树，包括消息、工具调用和压缩摘要
External Resources 代码库、技术文档、API规范等
- agent.defaults.memorySearch.extraPaths 允许递归索引外部MD文件夹，将其添加到记忆搜索范围
- memory.qmd.paths：声明QMD集合源，使其可以索引任意位置的文档

Memory模块解读

File/Backend Based Memory
- 日常记忆层
- 核心记忆层Memory.md：提炼过的、稳定持久的核心事实与偏好
- bank 记忆层（实验性）
  - world.md ：世界事实
  - experience.md ：Agent经历
  - opinion.md ：Agent观点
  - entities/*.md ：实体信息库
- Backend 处理Memory文件索引、查询和管理的底层引擎
  - builtin: SQLite + fts5 (BM25) + sqlite-vec
  - qmd (实验性): 高级混合搜索，继承BM25，向量搜索和重排
LanceDB Based Memory
- Plugin
  - memory-core 将能力暴露为标准的Agent工具和CLI命令
  - memory-lancedb 可选、功能完备的第三方长时记忆解决方案
    - 独立存储：LanceDB
    - 独立embedding：OpenAI API
    - 高级记忆操作工具
      - memory_recall, memory_stoore, memory_forget
    - hook
      - 记忆auto-capture & auto-recall
- LanceDB plugin
  - Tool
    - memory_store, memory_recall, memory_forget
  - CLI commands
    - list, search, stats
  - Lifecycle Hooks
    - before_agent_start, agent_end
  - Service
    - start/stop

LanceDBmemoreis表结构

id 唯一标识
text 记忆内容的原始文本
vector 嵌入向量
importance 记忆的权重0-1，默认0.7
category 记忆类别
- preference; fact; decision; entity; other
createdAt 记录创建时间戳

A complete guide to building skills for Claude

33页指南

EvoMap

一个协同进化的skill共享平台，核心理念是一个Agent习得的能力，全球Agents可继承

基础术语

Gene 最小可复用skill单元
Capsule 成功任务执行路径的标准化封装
Event 不可变的进化日志

【阅】本周阅读摘选2026-02-09 → 2026-02-15

Mon, 16 Feb 2026 00:00:00 +0000

本周阅读摘选 2026-02-09 → 2026-02-15 目录

学术相关

学术相关

Analyzing consumer footprints on E-commerce platforms: a multichannel sequential search model with reference price ¹

AlphaOPT: formulating optimization programs with self-improving LLM experience library ²

Microsoft丨Agent Lightning 训练方案

开源跨框架的LLM Agent训练方案，使用强化学习、自动提示词优化、监督微调等算法提升Agent的效果

Zhang, H., Guo, Z., Liu, J., & Wang, M. (2026). Analyzing consumer footprints on E-commerce platforms: A multichannel sequential search model with reference price. Information Systems Research, isre.2024.991. https://doi.org/10.1287/isre.2024.0991 ↩
Kong, M., Qu, A., Guo, X., Ouyang, W., Jiang, C., Zheng, H., Ma, Y., Zhuang, D., Tang, Y., Li, J., Wang, S., Koutsopoulos, H., Wang, H., Wu, C., & Zhao, J. (2026). AlphaOPT: Formulating optimization programs with self-improving LLM experience library (arXiv:2510.18428). arXiv. https://doi.org/10.48550/arXiv.2510.18428 ↩

【阅】本周阅读摘选2026-02-02 → 2026-02-08

Mon, 09 Feb 2026 00:00:00 +0000

本周阅读摘选 2026-02-02 → 2026-02-08 目录

学术相关
- SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue

学术相关

SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue ¹

Dai, Y., Gao, N., Zhang, W., Wang, J., Luo, Z., Wang, J., Wang, Y., Wu, R., & Wang, C. (2026). SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue (arXiv:2602.03548). arXiv. https://doi.org/10.48550/arXiv.2602.03548 ↩

【阅】本周阅读摘选2026-01-26 → 2026-02-01

Mon, 02 Feb 2026 00:00:00 +0000

本周阅读摘选 2026-01-26 → 2026-02-01 目录

学术相关
- OnePiece: bringing context engineering and reasoning to industrial cascade ranking system
技术技巧
- ElevenLabs UI Vue
娱乐
- 复杂系统三元交互：Mining higher-order triadic interactions

学术相关

OnePiece: bringing context engineering and reasoning to industrial cascade ranking system ¹

技术技巧

ElevenLabs UI Vue

多模态Agent应用UI库

娱乐

复杂系统三元交互：Mining higher-order triadic interactions ²

Dai, S., Tang, J., Wu, J., Wang, K., Zhu, Y., Chen, B., Hong, B., Zhao, Y., Fu, C., Wu, K., Ni, Y., Zeng, A., Wang, W., Chen, X., Xu, J., & Ng, S.-K. (2025). OnePiece: Bringing context engineering and reasoning to industrial cascade ranking system (arXiv:2509.18091). arXiv. https://doi.org/10.48550/arXiv.2509.18091 ↩
Niedostatek, M., Baptista, A., Yamamoto, J., Kurths, J., Sanchez Garcia, R., MacArthur, B. D., & Bianconi, G. (2025). Mining higher-order triadic interactions. Nature Communications, 16(1), 11613. https://doi.org/10.1038/s41467-025-66577-z ↩

【阅】本周阅读摘选2026-01-19 → 2026-01-25

Mon, 26 Jan 2026 00:00:00 +0000

本周阅读摘选 2026-01-19 → 2026-01-25 目录

学术相关
- Conscious awareness, sensory integration, and evidence accumulation in bodily self-perception
- Testing the drift-diffusion model
技术技巧

学术相关

Conscious awareness, sensory integration, and evidence accumulation in bodily self-perception ¹

Testing the drift-diffusion model ²

技术技巧

Skills资料

Agent Skills 官方独立文档
Claude Blog

对于复杂的skill文档，可以在skill目录中绑定其他文档并在SKILL.md中通过名称引用这些文档

可以通过skill执行脚本

开发建议

先识别Agent能力缺陷，评估skills对能力的提升
当SKILL.md臃肿时切分内容；对上下文互斥或少用内容单独分割
Code可以作为执行脚本和文档，需明确运行还是读取脚本
监控Agent实际使用，分析意外轨迹或错误，迭代skill，特别是name和description
完成任务后，让工具反思成功和创建的错误，提取到可复用的技能

Is semantic chunking worth the computational cost? ³

在拼接数据集上：语义分块效果更佳
在真实数据集上：语义分块与固定分块差异不大

Automatic Prompt Engineering (APE) 自动提示词优化

opik

Optn-source logging, debugging, and otpimization platform for AI agent & LLM Applications

logs, annotation, and scoring functions
pre-configured LLM-as-a-judge eval metrics
automated agent optimization algorithms

Lanfranco, R. C., Katyal, S., Hägerdal, A., Luan, X., Nos, V., & Ehrsson, H. H. (2025). Conscious awareness, sensory integration, and evidence accumulation in bodily self-perception. Proceedings of the National Academy of Sciences, 122(49), e2503629122. https://doi.org/10.1073/pnas.2503629122 ↩
Fudenberg, D., Newey, W., Strack, P., & Strzalecki, T. (2020). Testing the drift-diffusion model. Proceedings of the National Academy of Sciences, 117(52), 33141–33148. https://doi.org/10.1073/pnas.2011446117 ↩
Qu, R., Tu, R., & Bao, F. (2024). Is semantic chunking worth the computational cost? (arXiv:2410.13070). arXiv. https://doi.org/10.48550/arXiv.2410.13070 ↩

【阅】本周阅读摘选2026-01-12 → 2026-01-18

Mon, 19 Jan 2026 00:00:00 +0000

本周阅读摘选 2026-01-12 → 2026-01-18 目录

学术相关
技术技巧
- Learning Prompting

学术相关

Revenue Maximization and Learning in Product Ranking ¹

Learning user preferences in livestreaming market: A graphical model considering temporal effect ²

Matrix Factorization (补充)

在推荐系统中，用户与商品之间的交互数据（如评分、点击、购买等）通常表示为$m\times n$的系数矩阵$R$

$m$ 为用户数量
$n$ 为商品数量
$R_{i,j}$ 表示用户$i$对商品$j$的交互数据

矩阵分解的核心目的是将高维系数矩阵$R$分解为两个低维稠密矩阵的乘积，从而挖掘交互的潜在特征

$R\approx P\times Q^T$

$P \in \mathbb{R}^{m \times k}$ 为用户隐因子矩阵，每行$p_u$表示用户$u$的隐因子向量
$Q \in \mathbb{R}^{n \times k}$ 为商品隐因子矩阵，每列$q_i$表示商品$i$的隐因子向量
$k$ 为隐因子的维数，通常远小于$m$和$n$
$r_{ui}\approx p_u^Tq_i$ 为用户$u$对商品$i$的评分$

该方法擅长处理（适度）稀疏矩阵，常用于显示评分预测，隐式反馈推荐，具有较强的泛化能力和可解释性；缺点是存在冷启动问题，计算复杂度高，超参数敏感

Replicator dynamics and behavior-augmented multiscale epidemic modeling ³

技术技巧

Learning Prompting

一个比较全面的提示词学习网站

Chen, N., Li, A., & Yang, S. (2026). Revenue Maximization and Learning in Product Ranking. Operations Research, opre.2020.781. https://doi.org/10.1287/opre.2020.0781 ↩
Lin, Q., Li, Y., Kadziński, M., & Guo, M. (2026). Learning user preferences in livestreaming market: A graphical model considering temporal effect. Decision Support Systems, 202, 114600. https://doi.org/10.1016/j.dss.2025.114600 ↩
Wan, J., Liu, Q., Zacks, K., Zhao, J., Ichinose, G., Small, M., & Cheng, C. (2025). Replicator dynamics and behavior-augmented multiscale epidemic modeling. Chaos: An Interdisciplinary Journal of Nonlinear Science, 35(12), 123128. https://doi.org/10.1063/5.0297383 ↩

【阅】本周阅读摘选2026-01-05 → 2026-01-11

Mon, 12 Jan 2026 00:00:00 +0000

本周阅读摘选 2026-01-05 → 2026-01-11 目录

学术相关
- Sequential learning under informational ambiguity
- Is Competition Only One Click Away? The Digital Markets Act’s Impact on Google Maps
技术技巧
- DeepHub IMBA丨多智能体模式
  - 常见的多智能体模式

学术相关

Sequential learning under informational ambiguity ¹

Is Competition Only One Click Away? The Digital Markets Act’s Impact on Google Maps ²

技术技巧

DeepHub IMBA丨多智能体模式

单智能体在功能复杂情况下会变得不可用，该临界点被称为“单智能体墙” （Single-agent capacity threshold）

Once intelligence reaches a threshold, multi-agent systems become a vital way to scale performance. … Our internal evaluations show that multi-agent research systems excel especially for breadth-first queries that involve pursuing multiple independent directions simultaneously. We found that a multi-agent system with Claude Opus 4 as the lead agent and Claude Sonnet 4 subagents outperformed single-agent Claude Opus 4 by 90.2% on our internal research eval.³

核心原因是指令迷雾和工具过载

常见的多智能体模式

Sequential Muti-Agent
Parallel Multi-Agent
Layered or Hierarchical
Router
- 与layer的差异是不分解任务，而是直接分发
Reflect and Critique
Custom and Consensus

Chen, J. Y. (2026). Sequential learning under informational ambiguity. American Economic Review, 116(1), 209–245. https://doi.org/10.1257/aer.20231394 ↩
Pape, L.-D., & Rossi, M. (2026). Is Competition Only One Click Away? The Digital Markets Act’s Impact on Google Maps. Marketing Science, mksc.2025.159. https://doi.org/10.1287/mksc.2025.0159 ↩
Engineering at Anthropic. (2025, June 13). How we built our multi-agent research system. https://www.anthropic.com/engineering/multi-agent-research-system ↩

【阅】本周阅读摘选2025-12-29 → 2026-01-04

Mon, 05 Jan 2026 00:00:00 +0000

本周阅读摘选 2025-12-29 → 2026-01-04 目录

学术相关
- Designing information to engage customers
- 集智俱乐部丨综述：信息论如何成为复杂系统科学的核心工具
  - 部分信息分解 partial information decomposition (PID)
  - 传递熵 transfer entropy

学术相关

Designing information to engage customers ¹

集智俱乐部丨综述：信息论如何成为复杂系统科学的核心工具 ²

复杂系统：具有高度互联性、非线性、多尺度和动态性等特征，常表现出临界相变、级联故障和自组织等现象

香农熵

$H(X) = - \sum_{x \in \mathcal{X}} p(x) log (p(x))$

对数底通常为2，单位为比特

联合熵H(X,Y)与条件熵H(Y|X)

不妨令$H(Y) \leq H(X)$，则$H(Y)\leq H(X) \leq H(X,Y) \leq H(X) + H(Y)$

$H(X,Y) = H(X) + H(Y|X)$

$\therefore$ 当且仅当$Y$完全由$X$决定$H(Y|X)=0$时，联合熵等于最大的单熵；当且仅当两个变量完全独立$H(Y|X) \geq H(Y)$时，联合熵等于两个单熵之和

相对熵 Kullback-Leibler散度

$D_{KL}(P||Q)=\sum_{x \in \mathcal{X}} P(x) \log \frac{P(x)}{Q(x)}$

假设$Q(X)$为先验概率分布，$P(X)$为后验概率分布，则KL散度为将先验信念修正为P时获取的信息

KL散度可以被视作描述两个分布的“距离”，但不能假设$D_{KL}(P||Q) = D_{KL}(Q||P)$或假设满足三角不等式

修正的KL散度Jensen-Shannon散度满足对称性

$\begin{align}D_{JS}(P||Q) = \frac{D_{KL}(P||M)+ D_{KL}(Q||M)}{2}\ M := \frac{P+Q}{2}\end{align}$

$D_{JS}$不是距离指标，$\sqrt{D_{JS}(P||Q)}$为$P$到$Q$之间的Jensen-Shannon距离

局部相对熵

即将相对熵的期望“展开”

$d_{kL}(P(x)||Q(x)) = log \frac{P(x)}{Q(x)} = h_Q(x) - h_P(x)$

互信息I(X;Y)

互信息量化了两个变量的（任何形式的）统计依赖关系

$I(X;Y)=H(X) +H(Y) - H(X,Y)=D_{KL}(P(X,Y)||P(X)P(Y))=\sum_{\begin{align}X\in\mathcal{X}_1\\ Y\in\mathcal{Y} \end{align}}P(X,Y)log\frac{P(X|Y)}{P(X)}$

局部互信息

$i(X,Y)=h(X) - h(X|Y)=log \frac{P(X,Y)}{P(X)P(Y)}$

条件互信息

$I(X_1; X_2| X_3) = H(X_1,X_3) + H(X_2,X_3) - H(X_1,X_2,X_3) - H(X_3)$

部分信息分解 partial information decomposition (PID)

对于$I(S;X,Y)$由X和Y决定的关于目标S的总信息

冗余信息 (Redundancy): 由X和Y各自单独提供的，关于S的相同信息
特有信息 (Unique): 仅由X或Y提供的，关于S的独特信息
协同信息 (Synergy): 仅当同时考虑X和Y时，产生关于S的信息

$I(S;X,Y) = Red(X,Y\to S) + X\ Unq(X\to S|Y) + Y\ Unq(Y\to S|X) + Syn(X,Y\to S)$

当源数$N>2$时PID迅速复杂化，需要使用Redundancy Lattice等方法分解

传递熵 transfer entropy

传递熵为互信息在时间序列熵的推广，衡量在已知$Y$自身过去历史的情况下，$X$过去历史能为预测$Y$当前状态提供多少额外信息，从而推断出因果关系的方向。

局限性：能够检验统计依赖性，但无法完全确定因果关系（不能排除存在未观测的共同驱动变量）；对弱平稳过程

$T_{Y\to X} = H(X_{t+1}|X_t^{(k)}) - H(X_{t+1}|X_t^{(k)},Y_t^{(l)}) = I(Y_{t}^{(l)};X_{t+1}|X_t^{(k)})$

Guo, L. (2025). Designing information to engage customers. Management Science, 71(10), 8169–8187. https://doi.org/10.1287/mnsc.2022.03503 ↩
Varley, T. F. (2025). Information theory for complex systems scientists: What, why, and how. Physics Reports, 1148, 1–55. https://doi.org/10.1016/j.physrep.2025.09.007 ↩

【阅】本周阅读摘选2025-12-22 → 2025-12-28

Mon, 29 Dec 2025 00:00:00 +0000

本周阅读摘选 2025-12-22 → 2025-12-28 目录

学术相关
- Applying large language models to sponsored search advertising
业界动态
- McKinsey Global Institute’s favorite data visualizations from 2025

学术相关

Applying large language models to sponsored search advertising ¹

业界动态

McKinsey Global Institute’s favorite data visualizations from 2025

下面几张图在统计口径、对象、指标定义上都有值得玩味的地方。

Reisenbichler, M., Reutterer, T., & Schweidel, D. A. (2025). Applying large language models to sponsored search advertising. Marketing Science, mksc.2023.611. https://doi.org/10.1287/mksc.2023.0611 ↩

【阅】本周阅读摘选2025-12-15 → 2025-12-21

Mon, 22 Dec 2025 00:00:00 +0000

本周阅读摘选 2025-12-15 → 2025-12-21 目录

学术相关
- How Perceptual Disfluency Affects Consumer Choices
技术技巧
- Python丨云朵君 - Python数据库选型
- UV
  - 项目工作流

学术相关

How Perceptual Disfluency Affects Consumer Choices ¹

主题	Title	Author	Journal	ROF
决策	How perceptual disfluency affects consumer choices	Shahryar Mohsenin; Kurt P Munz【2025】	JCR	进行消费选择时，感知不畅可能导致消费者更依赖直观信息，如品牌知名度、推荐等，而非需要仔细考虑的产品规格数字；当消费者追求决策速度时，这种效应会被放大；该效应至少部分源自消费者将信息感知不畅错误地归因为决策复杂，当消费者被告知决策不畅的实际来源（如字体）或不进行消费决策（如仅排序）时，该效应会消失

SPL	CPL	GAP	RFW
对某些非选择任务，如逻辑推理，感知不畅可以作为异常信号，提高警觉并激发深层处理；部分研究认为感知不畅性会导致消费者更多关注分析型线索，提升多样性寻求，另一部分认为不存在该效应；感知不畅会导致决策延迟和妥协	缺少涉及消费者选择任务的研究，多集中在逻辑题上；感知不畅与增强被试分析处理之间的关系存在质疑	目前感知不畅度对消费者选择领域的影响及其潜在机制理解缺失	基于直觉型线索决策即可被理解为减少分析深度，也可被理解为细致分析后的方法选择，本研究没有区分；本研究不流畅集中在视觉线索上

POC-RPP	IV	DV	Method
作者测试的产品都是具有单一功能的平价实用型产品，对于贵重或复杂产品缺少测试；实验2中显示涉及感知决策难度的串行中介效应大小约占总效应的8.5%，实验4引导被试不将感知不畅归因为决策难度后总效应变得不显著，作者没有予以解释和进一步分析，简单地将实验干预归因为排除影响感知决策难度可能不严谨	是否感知不畅处理（字体、背景、图像、尺寸）	直觉型信息优势（熟悉/推荐产品）选项比例	对照实验；路径分析；attributional discounting paradigm（类似消融实验）

这项研究整体逻辑比较缜密，忽视一些对实验结果无意义的歧视性设计和几处关键笔误外，是一个比较好的实验设计

技术技巧

Python丨云朵君 - Python数据库选型

名称	类型	特征	适用场景
MySQL	关系型数据库	成熟完善	结构化数据
PostgreSQL	关系型数据库	强大的复杂数据规则；开源	复杂业务、GIS、JSON数据混合
SQLite	嵌入式关系型数据库	轻量级；My Favorite	应用、原型开发
MongoDB	文档型数据库	灵活数据类型	内容推送、日志存储、用户画像
Redis	内存键值数据库	极速读写小数据	缓存、会话、消息队列
Neo4j	图数据库	记录节点与关系	社交网络、推荐系统、知识图谱
Milvus	向量数据库	云原生架构	LLM

UV

帮助 uv [] --help
更新 pip install --upgrade uv
Python管理
- 安装版本 uv python install
- 列出版本 uv python list
- 删除版本 uv python uninstall
- 寻找版本 uv python find
- 指定版本 uv python pin
脚本管理
- 运行 uv run
- 添加脚本级依赖 uv add --script
- 删除脚本级依赖 uv remove --script
项目管理
- 初始化 uv init
- 添加项目依赖 uv add
- 删除项目依赖 uv remove
- 项目依赖与环境同步 uv sync (类似pip install -r)
- 创建lockfile uv lock
- 运行命令 uv run
- 依赖树（显示声明的） uv tree
- 打包 uv build
- 发布 uv publish 需要配置仓库
Tools
- 临时应用包 uvx / uv tool run
  - 创建一个隔离环境，下载运行工具后清理，避免污染全局环境
  - 常用于pytest等工具
- 安装到用户全局 uv tool install
- 卸载用户全局包 uv tool uninstall
- 列出用户全局包 uv tool list
环境管理
- 创建虚拟环境 (通常uv init会自动创建，无需手动) uv venv
- 安装到当前环境 uv pip install
- 卸载环境包 uv pip uninstall
- 显示安装包详情 uv pip show
- 列出依赖 uv pip freeze
- 易读列出 up pip list
- 兼容性校验 uv pip check
- 环境依赖树 uv pip tree
- lockfile uv pip compile
- 同步 uv pip sync
换源
- 临时换源uv add --default-index https://pypi.tuna.tsinghua.edu.cn/simple requests
- project换源：pyproject.toml
  - [[tool.uv.index]]
  - name = "tencent"
  - url = "https://mirrors.cloud.tencent.com/pypi/simple/"

项目工作流

# 初始化项目 生成配置文件
uv init
uv run main.py

# 添加依赖
uv add numpy
uv add 'pandas>=1.0.0"
uv add git+https://github.com/psf/request'
uv add -r requirements.txt -c constraints.txt
uv remove numpy
uv lock --upgrade-packages requests

# 发布
uv build
uv version 1.0.0
## 版本增量
uv version --bump [major|minor|patch|stable|aplpha|beta|rc]
uv publish

Mohsenin, S., & Munz, K. P. (2025). How Perceptual Disfluency Affects Consumer Choices. Journal of Consumer Research, ucaf062. https://doi.org/10.1093/jcr/ucaf062 ↩

Cao Zihang - Blog

【阅】本周阅读摘选2026-02-16 → 2026-02-22

技术技巧

Memory模块解读

【阅】本周阅读摘选2026-02-09 → 2026-02-15

学术相关

Analyzing consumer footprints on E-commerce platforms: a multichannel sequential search model with reference price 1

AlphaOPT: formulating optimization programs with self-improving LLM experience library 2

【阅】本周阅读摘选2026-02-02 → 2026-02-08

学术相关

SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue 1

【阅】本周阅读摘选2026-01-26 → 2026-02-01

学术相关

OnePiece: bringing context engineering and reasoning to industrial cascade ranking system 1

技术技巧

娱乐

复杂系统三元交互：Mining higher-order triadic interactions 2

【阅】本周阅读摘选2026-01-19 → 2026-01-25

学术相关

Conscious awareness, sensory integration, and evidence accumulation in bodily self-perception 1

Testing the drift-diffusion model 2

技术技巧

Skills资料

开发建议

Is semantic chunking worth the computational cost? 3

Automatic Prompt Engineering (APE) 自动提示词优化

【阅】本周阅读摘选2026-01-12 → 2026-01-18

学术相关

Revenue Maximization and Learning in Product Ranking 1

Learning user preferences in livestreaming market: A graphical model considering temporal effect 2

Matrix Factorization (补充)

Replicator dynamics and behavior-augmented multiscale epidemic modeling 3

技术技巧

【阅】本周阅读摘选2026-01-05 → 2026-01-11

学术相关

Sequential learning under informational ambiguity 1

Is Competition Only One Click Away? The Digital Markets Act’s Impact on Google Maps 2

技术技巧

常见的多智能体模式

【阅】本周阅读摘选2025-12-29 → 2026-01-04

学术相关

Designing information to engage customers 1

部分信息分解 partial information decomposition (PID)

传递熵 transfer entropy

【阅】本周阅读摘选2025-12-22 → 2025-12-28

学术相关

Applying large language models to sponsored search advertising 1

业界动态

【阅】本周阅读摘选2025-12-15 → 2025-12-21

学术相关

How Perceptual Disfluency Affects Consumer Choices 1

技术技巧

UV

项目工作流

Analyzing consumer footprints on E-commerce platforms: a multichannel sequential search model with reference price ¹

AlphaOPT: formulating optimization programs with self-improving LLM experience library ²

SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue ¹

OnePiece: bringing context engineering and reasoning to industrial cascade ranking system ¹

复杂系统三元交互：Mining higher-order triadic interactions ²

Conscious awareness, sensory integration, and evidence accumulation in bodily self-perception ¹

Testing the drift-diffusion model ²

Is semantic chunking worth the computational cost? ³

Revenue Maximization and Learning in Product Ranking ¹

Learning user preferences in livestreaming market: A graphical model considering temporal effect ²

Replicator dynamics and behavior-augmented multiscale epidemic modeling ³

Sequential learning under informational ambiguity ¹

Is Competition Only One Click Away? The Digital Markets Act’s Impact on Google Maps ²

Designing information to engage customers ¹

Applying large language models to sponsored search advertising ¹

How Perceptual Disfluency Affects Consumer Choices ¹