Cyprex API

引用检测 API

检测 AI 生成的虚假/幻觉学术引文，并检查引用相关性。通过级联查询 CrossRef、Semantic Scholar、OpenAlex 验证引文真实性，通过 BM25、NLI 和 LLM 评估判断引用相关性。

POST/v1/citation/verify

引用验证

POST/v1/citation/verify

通过 CrossRef、Semantic Scholar 和 OpenAlex 验证一条或多条学术引用。检测 AI 生成的虚假/幻觉引文。支持单条和批量验证（最多 50 条引用）。

参数

请求体为 JSON 对象，包含 citations 数组（1-50 项）。每项包含：

参数类型必填描述
citation_textstring可选原始引用字符串（最少 10 字符）。citation_text 和 structured 至少需要提供一个。
structuredobject可选结构化元数据：title、authors、year、journal、doi 等
format_hintstring可选 (默认: auto)'apa'、'mla'、'chicago'、'ieee'、'gbt7714' 或 'auto'

批量验证

在 citations 数组中传入多个条目，单次请求最多验证 50 条引用。响应包含汇总计数（verified_count、uncertain_count、fabricated_count）。

引用相关性

POST/v1/citation/relevance

检查被引论文是否与引用上下文真正相关。使用 BM25、NLI 和 LLM 评估的级联方法判断相关性。支持通过 DOI 自动获取摘要。

参数

参数类型必填描述
citing_contextstring必填引用出现的句子或段落（最少 10 字符）
cited_paper_titlestring可选被引论文标题
cited_paper_abstractstring可选被引论文摘要（如提供 DOI 可自动获取）
cited_doistring可选被引论文的 DOI（用于自动获取摘要）
check_depthstring可选 (默认: auto)'fast'、'standard'、'deep' 或 'auto'

cited_paper_title、cited_paper_abstract、cited_doi 至少需要提供其中一个。

响应

Response
{
  "result": {
    "relevance_verdict": "RELEVANT",
    "relevance_score": 85.2,
    "citation_intent": "supporting",
    "nli_label": "entailment",
    "explanation": "The cited paper directly introduces the transformer architecture...",
    "tier_used": 2,
    "processing_time_ms": 450.3
  }
}

判定类型

验证判定

VERIFIED — 引用与真实出版物高置信度匹配

UNCERTAIN — 部分匹配，建议人工复核

FABRICATED — 未找到匹配的出版物，可能为 AI 幻觉

相关性判定

RELEVANT — 被引论文与引用上下文明确相关

WEAKLY_RELEVANT — 主题有部分重叠但匹配度不高

IRRELEVANT — 与引用上下文无实质相关性

CONTRADICTORY — 被引论文与引用主张相矛盾

参数

参数类型必填描述
citation_textstring可选原始引用字符串（最少 10 字符）。citation_text 和 structured 至少需要提供一个。
structuredobject可选结构化引用元数据，包含字段：title、authors (string[])、year (int)、journal、volume、issue、pages、doi。
format_hintstring可选 (默认: auto)引用格式提示：'apa'、'mla'、'chicago'、'ieee'、'gbt7714' 或 'auto'（自动检测）

响应

Response
{
  "result": {
    "parsed_metadata": {
      "title": "Attention Is All You Need",
      "authors": [
        "Vaswani, A.",
        "Shazeer, N.",
        "Parmar, N."
      ],
      "year": 2017,
      "journal": "Advances in Neural Information Processing Systems",
      "doi": null
    },
    "verdict": "VERIFIED",
    "confidence": 92.5,
    "best_match": {
      "source": "crossref",
      "title": "Attention is all you need",
      "authors": [
        "Vaswani, Ashish",
        "Shazeer, Noam",
        "Parmar, Niki"
      ],
      "year": 2017,
      "doi": "10.5555/3295222.3295349",
      "overall_similarity": 100
    },
    "sources_queried": [
      "crossref"
    ],
    "hallucination_signals": [],
    "processing_time_ms": 320.5
  }
}

示例

import requests

API_KEY = "your_api_key_here"
BASE_URL = "https://api.qinyanai.com"

# --- Verify a single citation ---
response = requests.post(
    f"{BASE_URL}/v1/citation/verify",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "citations": [{
            "citation_text": "Vaswani, A., Shazeer, N. et al. (2017). Attention Is All You Need. NeurIPS, 30."
        }]
    }
)
results = response.json()["results"]
for r in results:
    print(f"Verdict: {r['verdict']}  Confidence: {r['confidence']}%")

# --- Batch verification (up to 50) ---
response = requests.post(
    f"{BASE_URL}/v1/citation/verify",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "citations": [
            {"citation_text": "Vaswani, A. et al. (2017). Attention Is All You Need. NeurIPS."},
            {"citation_text": "Devlin, J. et al. (2019). BERT: Pre-training of Deep Bidirectional Transformers. NAACL."},
        ]
    }
)
batch = response.json()
print(f"Verified: {batch['verified_count']}, Fabricated: {batch['fabricated_count']}")

# --- Check citation relevance ---
response = requests.post(
    f"{BASE_URL}/v1/citation/relevance",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "citing_context": "Transformer models have achieved state-of-the-art results in NLP tasks [1].",
        "cited_paper_title": "Attention Is All You Need",
        "check_depth": "auto"
    }
)
result = response.json()["result"]
print(f"Relevance: {result['relevance_verdict']}  Score: {result['relevance_score']}")