paper-idea-generator

name	paper-idea-generator
version	2.1.0
description	Generate novel, actionable paper ideas from arXiv papers and LLM analysis. Includes selection reasons, score breakdowns, and gap evidence.

Paper Idea Generator v2.1

🆕 v2.1 新增功能

选择原因 (Selection Reason)

每个点子都会显示：

• 为什么选择这个研究空白
• 哪些论文暗示这个空白存在

评分区分度 (Score Breakdown)

不再是统一的 85 分，而是：

• 新颖性 (30%): 方法创新程度
• 可行性 (25%): 技术实现难度
• 相关性 (25%): 与主题的匹配度
• 影响力 (20%): 潜在研究价值

空白证据 (Gap Evidence)

• 统计多少论文提及该技术
• 找出间接暗示该空白的论文

Parameters

Parameter	Type	Default	Description
topic	string	required	Research topic (e.g., "LLM Agents")
num_ideas	number	5	Number of ideas (1-10)
max_papers	number	20	Max papers from arXiv (5-50)
language	string	"both"	Output language: "en", "zh", "both"
detail_level	string	"detailed"	Detail level: "brief", "standard", "detailed"
focus_areas	array	[]	Focus on specific categories

Output Format

{
  "success": true,
  "ideas": [
    {
      "id": "idea_1234567890",
      "title": "Hierarchical Memory-Constrained Task Decomposition...",
      "title_zh": "面向长周期大语言模型智能体的分层记忆约束任务分解方法",
      "research_gap": "hierarchical",
      "research_gap_category": "agent",
      "research_questions": ["Question 1", "Question 2", "Question 3"],
      "core_method": "Detailed methodology...",
      "experiments": [
        {
          "name": "Performance Evaluation",
          "dataset": "ALFWorld",
          "baselines": ["AutoGen", "MetaGPT"],
          "metrics": ["Success Rate", "Completion Time"]
        }
      ],
      "expected_contributions": ["Contribution 1", "Contribution 2"],
      "related_papers": ["Paper 1", "Paper 2", "Paper 3"],
      "score": 85,
      "difficulty": "medium",
      "novelty": "moderate",
      "feasibility": "high",
      "timeline": "3-6 months",
      "resources_needed": ["GPU cluster", "Dataset access"]
    }
  ],
  "papers_found": 20,
  "gaps_identified": 15,
  "markdown": "...",
  "summary": {
    "total_ideas": 5,
    "avg_score": 82,
    "top_score": 90
  }
}

Usage Examples

Basic Usage

const result = await skill.execute(ctx, {
  topic: "LLM Agents"
});

With Options

const result = await skill.execute(ctx, {
  topic: "RAG Enhancement",
  num_ideas: 5,
  max_papers: 30,
  language: "both",
  detail_level: "detailed",
  focus_areas: ["retrieval", "optimization"]
});

Quick Generation

const result = await skill.execute(ctx, {
  topic: "Multi-Agent Systems",
  num_ideas: 3,
  detail_level: "brief"
});

Research Gap Categories

Category	Keywords
agent	hierarchical, peer-to-peer, competitive, collaborative, hybrid, blackboard, event-driven, ...
memory	shared, hierarchical, episodic, semantic, working, long-term, short-term, ...
reasoning	chain-of-thought, tree-of-thought, multi-hop, speculative, self-consistency, ...
learning	reinforcement, supervised, unsupervised, meta-learning, continual, transfer, ...
optimization	quantization, pruning, distillation, speculative-decoding, cache, ...
retrieval	sparse, dense, hybrid, multi-modal, adaptive, re-ranking, query-expansion, ...

Dependencies

• Internet access for arXiv API
• Configured LLM client for idea generation

Performance

• Batch Processing: 3 ideas generated in parallel
• arXiv Fetch: ~2-3 seconds for 20 papers
• LLM Generation: ~2-5 seconds per idea
• Total Time: ~15-30 seconds for 5 ideas