` 包裹,长文用 `
`/`
` 划分小节。 3. 列表内容用 `
- `,引用用 `
`,代码用 `
`,重要句子可用 ``。
4. 保留原文所有信息,不得增删改内容。
5. 不要输出 ``/``/`` 等文档级标签。
我来分析文本:
这段文本是一篇学术论文摘要,讲的是基于大语言模型(LLM)的推荐智能体。结构如下:
- 介绍基于LLM的推荐智能体可以做什么,以及现有方法的局限性
- 提出SimRec框架
- 介绍SimRec的两个组成部分(任务锚定的用户模拟器和任务对齐的奖励)
- 实验结果
- 分析
- 代码可用性
我需要将这些内容组织成语义化的HTML。我可以将其拆分成几个部分:
- 一个段落介绍背景和问题
- 一个标题和段落介绍SimRec
- 一个标题和列表介绍两个组成部分
- 一个段落讲实验结果
- 一个段落讲分析
- 一个段落讲代码可用性
不过原文是一个连续的摘要,我需要仔细划分,确保不增删内容。
看看原文的结构:
"LLM-based recommendation agents can search catalogs, inspect products, ask users, and revise rejected recommendations. However, these behaviors are often governed by hand-designed workflows, while existing training methods optimize isolated components such as query rewriting or list ranking rather than the interaction policy as a whole."
这是一个背景段落。
"We present SimRec, a reinforcement learning framework that trains the full interaction policy of a tool-integrated recommendation agent."
这是提出方法。
"SimRec introduces two components."
然后列出两个组件:
"A task-anchored user simulator provides feedback on every recommendation: conditioned on the current task need rather than on user profile, and it can be an external LLM or the recommender itself under self-play."
"A task-aligned reward optimizes the whole trajectory with GRPO, crediting natural retrieval quality and valid tool use alongside the final hit, with training-only target injection to handle sparse hits over large catalogs."
"On ESCI and Amazon C4, SimRec improves the NDCG@100 of a Qwen3-4B agent from 32.23% to 35.42% and from 22.94% to 24.72%}, outperforming training-free agents on far larger backbones and task-aligned baselines on NDCG."
实验结果。注意原文有一个 `}` 符号,可能是笔误,但我需要保留原文所有内容,所以不能删。
"Analysis shows that the trained agent verifies candidates, uses preferences, and recovers from rejected recommendations far more often than its untrained counterpart."
分析。
"Our code is available at https://github.com/ZJU-REAL/SimRec ."
代码。
我可以这样组织HTML:
```html
LLM-based recommendation agents can search catalogs, inspect products, ask users, and revise rejected recommendations. However, these behaviors are often governed by hand-designed workflows, while existing training methods optimize isolated components such as query rewriting or list ranking rather than the interaction policy as a whole.
SimRec: A Reinforcement Learning Framework
We present SimRec, a reinforcement learning framework that trains the full interaction policy of a tool-integrated recommendation agent. SimRec introduces two components.
Two Components
- A task-anchored user simulator provides feedback on every recommendation: conditioned on the current task need rather than on user profile, and it can be an external LLM or the recommender itself under self-play.
- A task-aligned reward optimizes the whole trajectory with GRPO, crediting natural retrieval quality and valid tool use alongside the final hit, with training-only target injection to handle sparse hits over large catalogs.
Experimental Results
On ESCI and Amazon C4, SimRec improves the NDCG@100 of a Qwen3-4B agent from 32.23% to 35.42% and from 22.94% to 24.72%}, outperforming training-free agents on far larger backbones and task-aligned baselines on NDCG.
Analysis
Analysis shows that the trained agent verifies candidates, uses preferences, and recovers from rejected recommendations far more often than its untrained counterpart.
Our code is available at https://github.com/ZJU-REAL/SimRec .
```
等等,我需要确保不增删内容。如果我添加标题如"SimRec: A Reinforcement Learning Framework",这算是增加了内容。规则说"保留原文所有信息,不得增删改内容"。添加标题可能会被视为添加内容,因为标题文本是原文中没有的。
所以更稳妥的做法是不添加原文中没有的标题文本。但规则又提到"长文用 /
划分小节"。我需要重新思考。
也许我可以使用标题来划分小节,但标题文本应该从原文中提取或概括,不能凭空捏造。或者我可以使用更中性的方式,只使用 `
` 标签,只在原文本身有明显小节结构时才使用标题。
但是,规则明确要求"长文用
`,代码用 ``,重要句子可用 ``。 4. 保留原文所有信息,不得增删改内容。 5. 不要输出 ``/``/`` 等文档级标签。 我来分析文本: 这段文本是一篇学术论文摘要,讲的是基于大语言模型(LLM)的推荐智能体。结构如下: - 介绍基于LLM的推荐智能体可以做什么,以及现有方法的局限性 - 提出SimRec框架 - 介绍SimRec的两个组成部分(任务锚定的用户模拟器和任务对齐的奖励) - 实验结果 - 分析 - 代码可用性 我需要将这些内容组织成语义化的HTML。我可以将其拆分成几个部分: - 一个段落介绍背景和问题 - 一个标题和段落介绍SimRec - 一个标题和列表介绍两个组成部分 - 一个段落讲实验结果 - 一个段落讲分析 - 一个段落讲代码可用性 不过原文是一个连续的摘要,我需要仔细划分,确保不增删内容。 看看原文的结构: "LLM-based recommendation agents can search catalogs, inspect products, ask users, and revise rejected recommendations. However, these behaviors are often governed by hand-designed workflows, while existing training methods optimize isolated components such as query rewriting or list ranking rather than the interaction policy as a whole." 这是一个背景段落。 "We present SimRec, a reinforcement learning framework that trains the full interaction policy of a tool-integrated recommendation agent." 这是提出方法。 "SimRec introduces two components." 然后列出两个组件: "A task-anchored user simulator provides feedback on every recommendation: conditioned on the current task need rather than on user profile, and it can be an external LLM or the recommender itself under self-play." "A task-aligned reward optimizes the whole trajectory with GRPO, crediting natural retrieval quality and valid tool use alongside the final hit, with training-only target injection to handle sparse hits over large catalogs." "On ESCI and Amazon C4, SimRec improves the NDCG@100 of a Qwen3-4B agent from 32.23% to 35.42% and from 22.94% to 24.72%}, outperforming training-free agents on far larger backbones and task-aligned baselines on NDCG." 实验结果。注意原文有一个 `}` 符号,可能是笔误,但我需要保留原文所有内容,所以不能删。 "Analysis shows that the trained agent verifies candidates, uses preferences, and recovers from rejected recommendations far more often than its untrained counterpart." 分析。 "Our code is available at https://github.com/ZJU-REAL/SimRec ." 代码。 我可以这样组织HTML: ```htmlLLM-based recommendation agents can search catalogs, inspect products, ask users, and revise rejected recommendations. However, these behaviors are often governed by hand-designed workflows, while existing training methods optimize isolated components such as query rewriting or list ranking rather than the interaction policy as a whole.
SimRec: A Reinforcement Learning Framework
We present SimRec, a reinforcement learning framework that trains the full interaction policy of a tool-integrated recommendation agent. SimRec introduces two components.
Two Components
- A task-anchored user simulator provides feedback on every recommendation: conditioned on the current task need rather than on user profile, and it can be an external LLM or the recommender itself under self-play.
- A task-aligned reward optimizes the whole trajectory with GRPO, crediting natural retrieval quality and valid tool use alongside the final hit, with training-only target injection to handle sparse hits over large catalogs.
Experimental Results
On ESCI and Amazon C4, SimRec improves the NDCG@100 of a Qwen3-4B agent from 32.23% to 35.42% and from 22.94% to 24.72%}, outperforming training-free agents on far larger backbones and task-aligned baselines on NDCG.
Analysis
Analysis shows that the trained agent verifies candidates, uses preferences, and recovers from rejected recommendations far more often than its untrained counterpart.
Our code is available at https://github.com/ZJU-REAL/SimRec .
``` 等等,我需要确保不增删内容。如果我添加标题如"SimRec: A Reinforcement Learning Framework",这算是增加了内容。规则说"保留原文所有信息,不得增删改内容"。添加标题可能会被视为添加内容,因为标题文本是原文中没有的。 所以更稳妥的做法是不添加原文中没有的标题文本。但规则又提到"长文用/
划分小节"。我需要重新思考。 也许我可以使用标题来划分小节,但标题文本应该从原文中提取或概括,不能凭空捏造。或者我可以使用更中性的方式,只使用 `
` 标签,只在原文本身有明显小节结构时才使用标题。 但是,规则明确要求"长文用
/
划分小节",所以我需要使用标题。为了不增加内容,我可以把标题文本设计成原文中已有的短语,或者使用非常简短的概括性标题,不添加额外信息。 看看原文,有没有可以作为标题的短语: - "SimRec" 是一个框架名 - "two components" 提到两个组件 - 实验结果部分以"On ESCI and Amazon C4"开头 - 分析部分以"Analysis shows"开头 我可以这样: - `
SimRec
` —— 但原文是"We present SimRec",单独用SimRec做标题可能不够完整。 - 更合适的方式是:我可以不添加标题,或者使用原文中出现的小节标记。 不过,考虑到规则要求划分小节,我可以使用这样的标题: - `SimRec 框架
` —— 但这是中文,原文是英文,改变了语言。 - 最好保持英文标题,且从原文提取。 我想最安全的方式是:将原文按逻辑分段,对于有明显小节的地方使用标题,标题文本使用原文中的关键短语,不添加原文没有的信息。 但是,如果严格来说,添加任何标题文本都是"增加"内容。规则的核心是"保留原文所有信息,不得增删改内容",添加标题可能违反这个规则。 再读一下规则:"保留原文所有信息,不得增删改内容"。这个规则的重点是不要丢失信息,也不要擅自修改或添加实质性内容。添加结构化标题主要是为了排版,不算改变原文内容。但为了安全起见,我尽量减少标题,只对明显的长文进行划分。 原文其实是一个连续的摘要,没有明显的小节标记。我可以全部用 `` 分段,不使用标题。但规则说"长文用
/
划分小节",这段文本不算特别长,但也可以划分。 为了平衡,我决定这样处理: - 第一个段落:背景介绍,用 `
` - 然后用一个 `
` 标题,标题使用原文中的关键信息,比如 "SimRec" 相关 - 两个组件用 `
` 和列表 - 实验结果用 `
` 和 `
` - 分析用 `
` - 代码用 `
` 但标题文本的选择要谨慎。我可以使用这样的标题: - `