SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsmulti-document-summarization

multi-document-summarization

Office & productivity

View repository

Updated 10-16 01:49

About this skill

Multi-source document summarization

Overview table

Item Content
Capability ID 19
Capability name Multi-source document summarization
Capability type Comprehensive summarization
Core function Perform cross-analysis on multiple related documents, extract commonalities, identify conflicts, and generate comprehensive conclusions
Input Multiple topically related text, PDF, Word documents, or web links
Output Comprehensive summary (paragraph or bullet-point format), comparative analysis table, list of conflict points, cross-document insights
Related capabilities Legal document summarization (single-document summarization), legal document formatting (unified output format), similar case retrieval and analysis (judgment comparison)
Key principles Cross-document association, consistency integration, conflict annotation, insight generation

Legal notice

Important notice:

  1. Multi-document summarization depends on the accuracy and reliability of each source document; uneven source quality may lead to biased summarization conclusions.
  2. Summarization conclusions reflect the overall picture of the input documents and do not represent an independent judgment of objective facts.
  3. When legal disputes are involved, different documents may represent opposing positions; summarization should remain neutral and objectively present the views of all parties.
  4. When there are factual conflicts between documents, summarization is only responsible for identifying the conflict points and does not judge the truth or falsity of the facts.

I. Core concepts

1.1 Single-document summarization vs. multi-document summarization

Dimension Single-document summarization Multi-document summarization
Processing object One document Multiple related documents
Core task Compress information and extract key points Cross-comparison and integration of associations
Nature of output A faithful abridgment of the original text A new knowledge structure across documents
Key capabilities Information screening Consistency identification, conflict discovery, pattern extraction
Typical value Quickly understand one document Grasp the overall picture, trends, and disputes of a group of documents

1.2 The four levels of multi-document analysis

Level 1: 单篇理解
    分别理解每篇文档的核心内容
    
Level 2: 共性提取
    识别多篇文档共同涉及的主题、观点、事实
    
Level 3: 差异对比
    识别文档之间的观点冲突、事实矛盾、表述差异
    
Level 4: 洞见生成
    基于跨文档的对比分析，生成新的综合性结论

1.3 Types of document relationships

Relationship type Characteristics Analysis focus Example
Parallel relationship Multiple documents address different aspects of the same topic Integrate information from all aspects to form a panoramic view Multiple academic papers on the same legal issue
Time series Documents are arranged in chronological order Sort out the evolution thread and identify trend changes Different versions of the same statute, a series of judgments in the same case
Opposition relationship Documents represent viewpoints of different positions Objectively present the viewpoints of all parties and mark points of conflict Opinions of both plaintiff and defendant, disputed views of different scholars
Hierarchical relationship Documents have inclusion or citation relationships Sort out the hierarchical structure and identify the core and the derivative The hierarchical system of laws, judicial interpretations, and guiding cases
Complementary relationship Documents complement each other and cover different aspects Integrate complementary information and fill in the blind spots of a single document The main contract and supplementary agreement, the indictment and the evidence list

II. Complete Workflow

Phase One: Preparation and Input

Step 1: Receive and organize documents

  1. Confirm the number and sources of documents: List all documents and mark their sources and types
  2. Unify the format: Convert PDFs, Word documents, web pages, etc. into processable text
  3. Create document files: Number each document for easy reference later
### 文档清单

| 编号 | 文档名称 | 类型 | 来源 | 日期 | 字数/页数 |
|------|----------|------|------|------|----------|
| D1 | [...] | [判决书/合同/法规/论文] | [...] | [...] | [...] |
| D2 | [...] | [...] | [...] | [...] | [...] |

Step 2: Identify relationships between documents

Analyze the types of relationships among multiple documents (parallel/time series/opposition/hierarchy/complementarity) and determine the analytical framework for synthesis:

  • Parallel relationship → Use the "theme aggregation" framework
  • Time series → Use the "evolution thread" framework
  • Opposition relationship → Use the "viewpoint comparison" framework
  • Hierarchical relationship → Use the "hierarchy sorting" framework
  • Complementary relationship → Use the "panoramic integration" framework

Phase Two: Single-Document Understanding and Annotation

Step 3: Extract core information from each document

Conduct single-document understanding for each document and extract the following elements:

Extract elements Content Annotation method
Core theme What core issue does the document focus on? Summarize in one sentence
Main viewpoints The document's core claims or conclusions 3-5 key points
Key facts Important facts that support the viewpoints Objective statement
Basis/evidence Supporting materials for the viewpoints Legal provisions, data, cases, etc.
Position/perspective The document's position (neutral/support/oppose/representing one party) Clearly mark

Step 4: Create document information cards

Create a standardized information card for each document:

### D1 信息卡片

- **文档名称：** [...]
- **核心主题：** [...]
- **主要观点：**
  1. [...]
  2. [...]
- **关键事实：**
  - [...]
- **依据/证据：** [...]
- **立场：** [中立/原告方/被告方/学者A观点/行政机关]
- **可信度/权威性：** [高/中/低，简要说明理由]

Phase Three: Cross-Analysis and Synthesis

Step 5: Extract cross-document commonalities

Identify themes, viewpoints, and facts commonly addressed by multiple documents:

  • Which viewpoints are mentioned in common across multiple documents?
  • Which facts are cross-corroborated by multiple documents?
  • Which legal bases are widely cited?
  • Are there widely accepted conclusions or consensus?

Output format:

### 跨文档共识

| 共识主题 | 涉及文档 | 共识内容 | 共识强度 |
|----------|----------|----------|----------|
| [...] | D1, D2, D3 | [...] | 强（3/3一致） |

Step 6: Identify conflicts between documents

Identify differing, contradictory, or opposing content between documents:

  • Conflict of viewpoints: Taking opposing positions on the same issue
  • Contradiction of facts: Inconsistent statements about the same fact
  • Divergence in legal application: Different understanding or application of the same legal provision
  • Difference in conclusions: Drawing different conclusions based on the same or similar facts

Requirements for annotating conflicts:

  • Objectively state the views of both sides of the conflict, without judging truth or falsity
  • Analyze the possible causes of the conflict (information asymmetry, different positions, different applicable standards, etc.)
  • Annotate the importance of the conflict (core conflict / secondary disagreement)

Output format:

### 文档间冲突点

| 冲突编号 | 冲突主题 | 文档A观点 | 文档B观点 | 冲突类型 | 重要程度 |
|----------|----------|-----------|-----------|----------|----------|
| C-01 | [...] | D1: [...] | D2: [...] | [观点/事实/法律适用] | [核心/次要] |

Step 7: Generate cross-document insights

On the basis of the commonalities and differences analysis, generate new comprehensive conclusions:

  • Trend judgment: What overall trend or direction do the documents present?
  • Pattern recognition: Are there recurring patterns or structures?
  • Blind spot identification: What important information have multiple documents collectively omitted?
  • Relationship discovery: Does combining information from different documents produce new understanding?

Phase Four: Output

Step 8: Format the output of the comprehensive synthesis

According to user needs, choose the output form:

Form A: Paragraph-style comprehensive summary(Suitable for detailed explanation scenarios)

  • Coherently present the core content of all documents
  • Organize by theme or logical order
  • Naturally incorporate cross-document comparison and insights

Form B: Bullet-point summary(Suitable for quick browsing)

  • Use bullet points to list key information item by item
  • Annotate the source document for each key point
  • List conflict points separately

Form C: Comparative analysis table(Suitable for fine-grained comparison)

  • Present the corresponding elements of each document side by side in table form
  • Facilitate item-by-item comparison of similarities and differences

III. Output format templates

Template A: Paragraph-style comprehensive summary

# 多文档综合归纳

## 一、文档概览
[说明文档数量、类型、来源和整体关系]

## 二、核心共识
[跨文档的共同主题、观点和事实]

## 三、主要分歧
[文档之间的关键冲突和差异]

## 四、综合结论
[基于跨文档分析生成的洞见]

## 五、信息缺口
[文档共同遗漏或表述不清的内容，建议补充]

Template B: Bullet-point summary

# 多文档要点归纳

## 文档清单
- D1: [...]
- D2: [...]

## 核心要点
- [要点1]（D1, D2, D3）
- [要点2]（D1, D2）
- [要点3]（D2, D3）

## 冲突与分歧
- [冲突1]：D1认为...，D2认为...（[冲突类型]）
- [冲突2]：...

## 跨文档洞见
- [洞见1]
- [洞见2]

## 待核实/补充信息
- [...]

Template C: Comparative analysis table

| 对比维度 | D1 | D2 | D3 | 一致性评价 |
|----------|-----|-----|-----|-----------|
| 核心主题 | [...] | [...] | [...] | [一致/相似/不同] |
| 主要观点 | [...] | [...] | [...] | [...] |
| 关键事实 | [...] | [...] | [...] | [...] |
| 法律依据 | [...] | [...] | [...] | [...] |
| 结论/处理 | [...] | [...] | [...] | [...] |

IV. Examples

Example 1: Summary of judgments in similar cases

Input: 3 judgments on unfair competition disputes (all plaintiffs are data platforms, all defendants are data scrapers)

Output:

# 多文档综合归纳

## 一、文档概览
本组共3份民事判决书，均为数据平台诉数据抓取方不正当竞争纠纷案，分别由
北京知识产权法院、杭州互联网法院、上海知识产权法院作出（D1-D3）。

## 二、核心共识

1. **数据权益保护共识**：三份判决均认定，平台对合法收集、投入成本整理的数据享有竞争性权益（D1, D2, D3）。
2. **不正当竞争认定标准共识**：均适用《反不正当竞争法》第2条一般条款，以"是否违反诚实信用原则和公认商业道德"作为判断标准（D1, D2, D3）。
3. **损害要件共识**：均要求原告证明被告行为对其造成了实际损害或损害威胁（D1, D2, D3）。

## 三、主要分歧

| 冲突编号 | 冲突主题 | 文档A观点 | 文档B观点 | 冲突类型 | 重要程度 |
|----------|----------|-----------|-----------|----------|----------|
| C-01 | 公开数据抓取是否当然合法 | D1（北京）：公开数据抓取仍需遵守Robots协议和商业道德 | D2（杭州）：公开数据在特定条件下可自由抓取 | 法律适用分歧 | 核心 |
| C-02 | 损害赔偿计算方式 | D1：以原告实际损失计算 | D3（上海）：以被告违法所得计算 | 法律适用分歧 | 核心 |
| C-03 | 技术中立抗辩 | D2：技术中立可作为免责事由 | D1, D3：技术中立不免除不正当竞争责任 | 观点冲突 | 次要 |

## 四、综合结论

1. **裁判趋势**：数据平台对合法收集的数据享有竞争性权益已成为司法共识，但公开数据抓取的合法性边界仍存在分歧。
2. **地域差异**：北京法院对数据保护立场较为严格，杭州法院对公开数据流通持相对开放态度，上海法院居中。
3. **法律适用**：三份判决均以《反不正当竞争法》第2条为主要依据，但具体适用标准存在微妙差异，反映该领域法律规范的不确定性。

## 五、信息缺口
- 三份判决均未涉及数据抓取行为与《个人信息保护法》的交叉适用问题
- 缺乏对"数据的可携带性"与"平台数据权益"平衡的具体分析
- 损害赔偿的计算标准和方法论尚未形成统一裁判规则

Example 2: Comparison of clauses in multiple contracts

Input: 2 technical service contracts (the main contract signed between Company A and Company B, and a similar contract signed between Company C and Company D)

Output:

# 多文档对比分析

## 一、文档概览
- D1：甲公司 vs 乙公司《技术开发合同》（2024年）
- D2：丙公司 vs 丁公司《技术服务合同》（2023年）

## 二、对比分析表

| 对比维度 | D1（甲乙） | D2（丙丁） | 差异评价 |
|----------|-----------|-----------|----------|
| 合同标的 | App软件开发 | 系统维护服务 | 不同：开发 vs 维护 |
| 总价款 | 50万元 | 30万元/年 | 不同：一次性 vs 年度 |
| 履行期限 | 6个月 | 12个月 | 不同 |
| 知识产权归属 | 归甲公司所有 | 归丙公司所有 | 一致 |
| 保密义务 | 详细条款，违约金10万元 | 简略条款，无违约金 | 不同：D1更严格 |
| 争议解决 | 北京仲裁委仲裁 | 被告所在地法院诉讼 | 不同：仲裁 vs 诉讼 |
| 不可抗力 | 列举+兜底条款 | 仅列举 | 不同：D1更全面 |

## 三、关键差异分析

1. **争议解决方式**：D1约定仲裁（一裁终局），D2约定诉讼（可上诉）。仲裁保密性强但不可上诉，诉讼程序公开但审级更多。
2. **保密义务强度**：D1设有具体违约金，D2仅作原则性约定。D1对守约方保护更强。
3. **风险分配**：D1作为开发合同，设有阶段性验收和付款节点；D2作为维护合同，按年度付费，风险更均衡。

## 四、综合建议

如拟签订类似合同，建议：
- 参考D1的保密条款和阶段性验收机制
- 参考D2的年度付费模式降低一次性付款风险
- 争议解决方式应根据保密需求和对效率的要求选择

Example 3: Summary of the evolution of laws and regulations

Input: The Personal Information Protection Law (2021), the Administrative Regulations on Network Data Security Management (2024), and an industry data compliance guide (2025)

Output:

# 多文档综合归纳

## 一、文档概览
本组包含3份规范性文件，呈现个人信息/数据保护领域从法律到行政法规再到行业指南的层级演变（D1-D3）。

## 二、规范层级与效力

| 文档 | 层级 | 效力 | 适用范围 |
|------|------|------|----------|
| D1《个人信息保护法》 | 法律 | 最高 | 全国范围内个人信息处理活动 |
| D2《网络数据安全管理条例》 | 行政法规 | 高 | 网络数据处理活动 |
| D3《XX行业数据合规指南》 | 行业规范 | 中 | XX行业内企业 |

## 三、核心内容演变

| 主题 | D1（法律） | D2（行政法规） | D3（行业指南） |
|------|-----------|---------------|---------------|
| 知情同意 | 原则性规定（第13-17条） | 细化同意机制、撤回同意程序 | 提供同意界面设计模板和示例 |
| 跨境传输 | 安全评估、标准合同、认证三种路径 | 细化安全评估申报流程 | 提供行业特定的风险评估要点 |
| 数据分类分级 | 原则要求 | 分类分级标准、重要数据识别 | 行业数据分类示例清单 |
| 法律责任 | 第66-71条（罚款、责令改正等） | 补充行政处罚裁量基准 | 合规整改操作指引 |

## 四、合规要求递进关系

Law (D1): Establish the basic institutional framework ↓ Refinement Administrative regulations (D2): Clarify operating procedures and penalty standards ↓ Implementation Industry guide (D3): Provide specific implementation templates and checklists


## 五、综合洞见

1. **监管趋势**：从原则性立法向精细化执行演进，企业合规义务逐步具体化。
2. **合规路径**：企业应先满足法律（D1）的强制性要求，再遵循行政法规（D2）的程序性要求，最后参考行业指南（D3）优化合规操作。
3. **注意冲突**：如D3与D1/D2存在不一致，应以效力更高的D1/D2为准，D3仅供参考。

V. Common errors and prevention

Error type Error description Consequence Preventive measures
Simple stacking Mechanically concatenating the summaries of each document without cross-analysis Losing the value of multi-document synthesis; readers still need to compare on their own Cross-document analysis of commonalities and differences must be performed
Ignoring conflicts Extracting only commonalities while avoiding or downplaying disagreements among documents Misleading readers into thinking all parties' views are consistent Points of conflict should be one of the focuses of analysis
Biased stance Overemphasizing a document because it is long or its source is "authoritative" The synthesized conclusion loses neutrality Determine weight based on the informational value of each document rather than its source
Over-inference Making excessive inferences about content not explicitly stated in the documents The synthesized conclusion lacks document support Distinguish between "explicitly stated in the document" and "reasonable inference based on the document"; the latter should be marked
Omitting important documents Failing to verify whether the documents are complete, omitting key attachments The synthesis is based on incomplete information Confirm the completeness of the checklist when receiving documents
Confusing facts with opinions Treating subjective opinions in documents as objective facts The synthesized conclusion is unreliable Clearly mark the nature of each piece of content (fact/opinion/inference)

VI. Handling of special scenarios

6.1 Too many documents (more than 10)

  • First group by topic or type, synthesize each group, then synthesize across groups
  • Use a "layered synthesis" strategy: first synthesize small groups, then larger groups, and finally form an overall conclusion
  • Focus on frequently occurring themes and points of conflict; secondary documents may be appropriately simplified

6.2 Uneven document quality

  • Establish credibility assessment for documents, and mark risks for documents of unknown origin or with obvious errors
  • Give higher weight to high-quality documents (such as effective judgments, formal regulations)
  • Mark low-quality documents (such as anonymous online articles, unverified rumors) separately and use them cautiously

6.3 Inconsistent document languages

  • If multilingual documents are involved, core concepts should be uniformly marked with the original text and translation
  • Pay attention to subtle differences in the same legal term across different language versions
  • For legal texts, the official version shall prevail

6.4 Involving confidential or sensitive information

  • Sensitive information should be desensitized in the synthesis output
  • Mark which content comes from classified documents and remind users to pay attention to the scope of use
  • Information involving personal privacy should be anonymized

6.5 Continuously updated document collections

  • If documents are continuously updated (such as regulation databases, case databases), the synthesis should be marked with a timestamp
  • Explain the timeliness of the synthesized conclusion
  • Recommend that users regularly update the synthesis

VII. Quality checklist

□ 1. 全部文档是否已接收并整理（含编号、类型、来源）？
□ 2. 文档间的关系类型是否已识别（平行/时间序列/对立/层级/补充）？
□ 3. 每篇文档的核心信息是否已提取（主题、观点、事实、依据、立场）？
□ 4. 跨文档共性是否已识别并整理？
□ 5. 文档间冲突是否已识别并客观标注（不做真伪判断）？
□ 6. 冲突点是否已分析可能原因？
□ 7. 跨文档洞见是否已生成（趋势、模式、盲区、关联）？
□ 8. 归纳结论是否有文档支撑（区分"明确表述"和"合理推断"）？
□ 9. 信息缺口是否已标注？
□ 10. 输出格式是否与用户需求匹配？
□ 11. 敏感信息是否已脱敏处理？
□ 12. 文档质量差异是否已在归纳中体现？

VIII. Related skills

Related skills Relationship Description
Legal document summarization Upstream First summarize individual documents, then perform multi-document synthesis
Legal document formatting Complementary Ensure the output format of multi-document summarization is standardized and unified
Similar case retrieval and analysis Extension When all documents are judgments, in-depth comparison can be conducted in combination with similar case analysis skills
Legal norm validity check Upstream When summarizing laws and regulations, first confirm the current validity of each document
Multi-document summarization Self-extension When batch summarization is needed for more documents (more than 10)

IX. Limitations and Risk Warnings

  • Source dependence: Summarization conclusions depend entirely on the content and quality of the input documents. If the source documents contain errors, bias, or incomplete information, the summarization conclusions will inherit these problems.
  • Weight judgment: Multiple documents may differ in importance, but this skill cannot automatically determine "which document is more important," and users need to assign weights according to the specific scenario.
  • Depth limitation: This skill focuses on horizontal comparison and integration of information. For content requiring in-depth legal analysis, it still needs to be combined with skills such as application of legal rules and analysis of the basis of claims.
  • Timeliness: When documents with strong timeliness such as regulations and precedents are involved, the summarization conclusions only reflect the state of the input documents at the point in time.
  • Language understanding limitation: For highly specialized or obscure texts, there may be deviations in understanding. When key legal concepts are involved, it should be combined with legal concept understanding skills.
  • Conflict resolution: This skill is only responsible for identifying conflict points and does not make judgments on the truth or falsity of facts or draw legal conclusions. The resolution of conflicts requires further legal analysis or evidence review.