uspto-database

uspto-database

概述

USPTO 提供两种主要的专利数据编程访问方式：PatentsView API（REST，免费，基本使用无需密钥），可按发明人、受让人、CPC 分类和关键词执行结构化查询；以及 Google Patents Public Data（BigQuery 公共数据集），可对完整专利语料进行大规模分析。两者提供的数据均采用 CC0 公共领域贡献协议。本技能介绍这两种方式的 Python 访问模式，以及基础的专利组合分析。

适用场景

• 现有技术检索：在提交申请前查找与某项技术相关的现有专利，或评估实施自由。
• 竞争对手知识产权格局分析：查询特定受让人（公司或机构）的所有专利，以绘制其技术组合。
• CPC 分类检索：使用合作专利分类代码查找特定技术领域的专利（例如，C12N 表示核苷酸／基因工程）。
• 发明人网络分析：识别某个领域中的高产发明人及其机构所属关系。
• 技术趋势追踪：按年份和技术类别统计专利申请数量，以识别新兴领域。
• 生命科学知识产权分析：检索生物技术专属分类（A61K 表示药品，C12N 表示遗传学，G16B 表示生物信息学）。
• 如需下载专利全文 PDF，请使用 USPTO Bulk Data Storage System（BDSS）或 Google Patents 直链。
• 速率限制：PatentsView API 在未使用 API 密钥时每分钟允许 45 次请求；申请免费密钥后，每分钟仍为 45 次请求，但每日限额更高。

前置条件

• Python 包：requests、pandas、matplotlib
• 可选：使用 Google Patents Public Data 查询时需要 google-cloud-bigquery
• 数据要求：PatentsView 基本查询无需账号；BigQuery 需要 Google Cloud 账号
• 速率限制：PatentsView——每分钟 45 次请求（未经身份验证），使用免费 API 密钥后限额更高

pip install requests pandas matplotlib
pip install google-cloud-bigquery  # optional: for BigQuery access

快速开始

import requests
import pandas as pd

# Search PatentsView API: patents assigned to "Genentech" in CPC class C12N
url = "https://api.patentsview.org/patents/query"
payload = {
    "q": {"_and": [
        {"_contains": {"assignee_organization": "Genentech"}},
        {"_contains": {"cpc_subgroup_id": "C12N"}},
    ]},
    "f": ["patent_number", "patent_title", "patent_date", "assignee_organization"],
    "o": {"per_page": 25},
}
resp = requests.post(url, json=payload)
data = resp.json()
df   = pd.DataFrame(data["patents"])
print(f"Found: {data['total_patent_count']} patents")
print(df[["patent_number", "patent_title", "patent_date"]].head())

核心 API

查询类型 1：按受让人（公司／机构）检索

查找授予特定组织的所有专利。

import requests
import pandas as pd

def search_by_assignee(assignee_name: str, per_page: int = 100) -> pd.DataFrame:
    url     = "https://api.patentsview.org/patents/query"
    payload = {
        "q": {"_contains": {"assignee_organization": assignee_name}},
        "f": [
            "patent_number", "patent_title", "patent_date",
            "patent_abstract", "assignee_organization", "assignee_country",
        ],
        "o": {"per_page": per_page, "sort": [{"patent_date": "desc"}]},
    }
    resp = requests.post(url, json=payload)
    resp.raise_for_status()
    data = resp.json()
    df   = pd.DataFrame(data.get("patents", []))
    print(f"Assignee '{assignee_name}': {data.get('total_patent_count', 0)} total patents")
    return df

# Example: patents from Broad Institute
df_broad = search_by_assignee("Broad Institute")
print(df_broad[["patent_number", "patent_title", "patent_date"]].head(10))

# Paginate through all results for large portfolios
def search_assignee_all_pages(assignee_name: str, page_size: int = 100) -> pd.DataFrame:
    url  = "https://api.patentsview.org/patents/query"
    all_patents = []
    page = 1
    while True:
        payload = {
            "q": {"_contains": {"assignee_organization": assignee_name}},
            "f": ["patent_number", "patent_title", "patent_date", "cpc_subgroup_id"],
            "o": {"per_page": page_size, "page": page},
        }
        resp = requests.post(url, json=payload)
        data = resp.json()
        patents = data.get("patents", [])
        if not patents:
            break
        all_patents.extend(patents)
        total = data.get("total_patent_count", 0)
        if len(all_patents) >= total:
            break
        page += 1

    df = pd.DataFrame(all_patents)
    print(f"Retrieved {len(df)} patents for '{assignee_name}'")
    return df

查询类型 2：按 CPC 分类检索

CPC（合作专利分类）代码用于按技术领域组织专利。生命科学代码包括 C12N（核苷酸／遗传学）、A61K（药品）和 G16B（生物信息学）。

import requests
import pandas as pd

# Search by CPC subgroup: C12N15 (mutation/genetic engineering)
url = "https://api.patentsview.org/patents/query"
payload = {
    "q": {"_begins": {"cpc_subgroup_id": "C12N15"}},
    "f": [
        "patent_number", "patent_title", "patent_date",
        "assignee_organization", "cpc_subgroup_id",
    ],
    "o": {"per_page": 50, "sort": [{"patent_date": "desc"}]},
}
resp = requests.post(url, json=payload)
data = resp.json()
df   = pd.DataFrame(data["patents"])
print(f"C12N15 patents: {data['total_patent_count']}")
print(df[["patent_number", "patent_title", "assignee_organization"]].head(10))

# Common life sciences CPC codes
CPC_LIFE_SCIENCES = {
    "C12N":    "Microorganisms / enzymes / compositions",
    "C12N15":  "Mutation / genetic engineering",
    "C12Q":    "Measuring / testing involving enzymes or microorganisms",
    "A61K":    "Preparations for medical use",
    "A61P":    "Therapeutic activity of chemical compounds",
    "G16B":    "Bioinformatics",
    "G16H":    "Healthcare informatics",
    "C07K":    "Peptides / proteins",
}
for code, desc in CPC_LIFE_SCIENCES.items():
    print(f"  {code:10s}: {desc}")

查询类型 3：全文关键词检索

在专利标题和摘要中搜索特定词语。

import requests
import pandas as pd

def keyword_search(keyword: str, per_page: int = 50) -> pd.DataFrame:
    url = "https://api.patentsview.org/patents/query"
    payload = {
        "q": {"_or": [
            {"_text_any": {"patent_title":    keyword}},
            {"_text_any": {"patent_abstract": keyword}},
        ]},
        "f": [
            "patent_number", "patent_title", "patent_date",
            "patent_abstract", "assignee_organization",
        ],
        "o": {"per_page": per_page, "sort": [{"patent_date": "desc"}]},
    }
    resp = requests.post(url, json=payload)
    resp.raise_for_status()
    data = resp.json()
    df   = pd.DataFrame(data.get("patents", []))
    print(f"Keyword '{keyword}': {data.get('total_patent_count', 0)} patents found")
    return df

# Search for CRISPR-related patents
df_crispr = keyword_search("CRISPR")
print(df_crispr[["patent_number", "patent_title", "patent_date"]].head(10))

查询类型 4：按发明人检索

按发明人姓名查找专利，或检索发明人的完整出版记录。

import requests
import pandas as pd

# Search by inventor name
url = "https://api.patentsview.org/inventors/query"
payload = {
    "q": {"_and": [
        {"inventor_last_name":  "Doudna"},
        {"inventor_first_name": "Jennifer"},
    ]},
    "f": ["inventor_id", "inventor_first_name", "inventor_last_name",
          "inventor_city", "inventor_state", "inventor_country"],
    "o": {"per_page": 10},
}
resp = requests.post(url, json=payload)
data = resp.json()
print(f"Found {data.get('total_inventor_count', 0)} inventors matching 'Jennifer Doudna'")
for inv in data.get("inventors", []):
    print(f"  ID: {inv['inventor_id']}, Location: {inv.get('inventor_city')}, {inv.get('inventor_country')}")

# Get all patents for a specific inventor by inventor_id
inventor_id = "fl:j_ln:doudna-1"   # PatentsView inventor ID format

url = "https://api.patentsview.org/patents/query"
payload = {
    "q": {"inventor_id": inventor_id},
    "f": ["patent_number", "patent_title", "patent_date", "assignee_organization"],
    "o": {"per_page": 100, "sort": [{"patent_date": "desc"}]},
}
resp  = requests.post(url, json=payload)
data  = resp.json()
df    = pd.DataFrame(data.get("patents", []))
print(f"Patents for inventor {inventor_id}: {data.get('total_patent_count', 0)}")
print(df.head(5))

查询类型 5：日期范围和组合筛选条件

组合多个筛选条件，执行针对性检索。

import requests
import pandas as pd

# Patents in gene therapy (CPC A61K48) filed 2020-2024 by a US assignee
url = "https://api.patentsview.org/patents/query"
payload = {
    "q": {"_and": [
        {"_begins":    {"cpc_subgroup_id": "A61K48"}},
        {"_gte":       {"patent_date": "2020-01-01"}},
        {"_lte":       {"patent_date": "2024-12-31"}},
        {"_eq":        {"assignee_country": "US"}},
    ]},
    "f": [
        "patent_number", "patent_title", "patent_date",
        "assignee_organization", "patent_num_claims",
    ],
    "o": {"per_page": 100, "sort": [{"patent_date": "desc"}]},
}
resp = requests.post(url, json=payload)
data = resp.json()
df   = pd.DataFrame(data.get("patents", []))
print(f"Gene therapy patents 2020-2024 (US assignee): {data.get('total_patent_count', 0)}")
print(df[["patent_number", "patent_title", "patent_date", "assignee_organization"]].head(10))

查询类型 6：Google Patents BigQuery

如需对专利语料进行大规模分析，请使用 BigQuery 中的 Google Patents 公共数据集。

from google.cloud import bigquery

client = bigquery.Client(project="YOUR_GCP_PROJECT")

# Count CRISPR patents by year (Google Patents public data)
query = """
SELECT
    EXTRACT(YEAR FROM filing_date) AS filing_year,
    COUNT(*)                        AS patent_count,
    COUNT(DISTINCT assignee)        AS unique_assignees
FROM `patents-public-data.patents.publications`
WHERE
    (LOWER(title_localized[SAFE_OFFSET(0)].text) LIKE '%crispr%'
     OR LOWER(abstract_localized[SAFE_OFFSET(0)].text) LIKE '%crispr%')
    AND filing_date >= '2010-01-01'
    AND country_code = 'US'
GROUP BY filing_year
ORDER BY filing_year
"""
df_bq = client.query(query).to_dataframe()
print(df_bq)
print(f"Peak year: {df_bq.loc[df_bq.patent_count.idxmax(), 'filing_year']} "
      f"({df_bq.patent_count.max()} patents)")

关键参数

参数	模块	默认值	范围／选项	作用
per_page	PatentsView "o"	25	1–10000	每次 API 调用返回的结果数
page	PatentsView "o"	1	1–最大页数	分页时的页码
sort	PatentsView "o"	API 默认值	任意字段 + "asc"／"desc"	结果排序方式
"f" 字段	PatentsView	最少字段	任意有效字段列表	响应中返回的字段（控制载荷大小）
"_begins"	查询运算符	—	字段 + 前缀字符串	前缀匹配（例如，CPC 代码前缀）
"_contains"	查询运算符	—	字段 + 子字符串	子字符串搜索（不区分大小写）
"_text_any"	查询运算符	—	字段 + 关键词	对标题／摘要字段进行全文搜索

最佳实践

1. 仅请求所需字段：「"f"（字段）参数控制返回的内容。为数千项专利请求 patent_abstract 会显著增加载荷大小和延迟。

2. 处理大型结果集时，始终进行分页：PatentsView 将每页响应上限设为 10,000 条。对于返回结果超过 10,000 条的查询，请按日期范围拆分，或缩小 CPC 代码范围以拆分查询。

3. 将 API 响应缓存到磁盘：PatentsView 有速率限制；如果要迭代构建数据集，请在每次 API 调用后将响应保存为 JSON/CSV。

   import json, pathlib
   cache = pathlib.Path("cache")
   cache.mkdir(exist_ok=True)
   cache_file = cache / "genentech_patents.json"
   if not cache_file.exists():
       resp = requests.post(url, json=payload)
       cache_file.write_text(resp.text)
   data = json.loads(cache_file.read_text())
   

4. 针对特定技术的搜索使用 CPC 代码，而不仅是关键词：关键词会漏掉同义词和外语专利；专利审查员会分配 CPC 代码，因此这种方式更系统。

5. 验证受让人名称：专利记录中的公司名称可能有所不同（例如「Genentech Inc.」「Genentech, Inc.」「GENENTECH INC」）。使用 _contains 进行模糊匹配，然后在 pandas 中去重。

常见工作流

工作流 1：技术格局分析——按年份统计申请趋势

目标：统计某个 CPC 类别中每年的专利申请数量，并绘制趋势图。

import requests
import pandas as pd
import matplotlib.pyplot as plt
from collections import defaultdict

def count_patents_by_year(cpc_prefix: str, start_year: int = 2010) -> pd.DataFrame:
    url  = "https://api.patentsview.org/patents/query"
    counts = defaultdict(int)
    page   = 1
    while True:
        payload = {
            "q": {"_and": [
                {"_begins": {"cpc_subgroup_id": cpc_prefix}},
                {"_gte":    {"patent_date": f"{start_year}-01-01"}},
            ]},
            "f": ["patent_number", "patent_date"],
            "o": {"per_page": 10000, "page": page},
        }
        resp    = requests.post(url, json=payload)
        patents = resp.json().get("patents", [])
        if not patents:
            break
        for p in patents:
            year = p["patent_date"][:4]
            counts[year] += 1
        total = resp.json().get("total_patent_count", 0)
        if sum(counts.values()) >= total:
            break
        page += 1

    df = pd.DataFrame(sorted(counts.items()), columns=["year", "count"])
    return df

df_trend = count_patents_by_year("C12N15", start_year=2010)

fig, ax = plt.subplots(figsize=(8, 4))
ax.bar(df_trend["year"], df_trend["count"], color="steelblue", edgecolor="white")
ax.set_xlabel("Year")
ax.set_ylabel("Patents granted")
ax.set_title("US Patents: C12N15 (Genetic Engineering) by Year")
plt.xticks(rotation=45)
plt.tight_layout()
plt.savefig("cpc_trend.png", dpi=150)
print(f"Trend plotted: {df_trend['count'].sum()} total patents -> cpc_trend.png")

工作流 2：受让人专利组合对比

目标：比较目标 CPC 类别中多家生物技术公司的专利数量。

import requests
import pandas as pd
import matplotlib.pyplot as plt

def count_patents_by_assignee(assignees: list, cpc_prefix: str) -> pd.DataFrame:
    url     = "https://api.patentsview.org/patents/query"
    records = []
    for assignee in assignees:
        payload = {
            "q": {"_and": [
                {"_contains": {"assignee_organization": assignee}},
                {"_begins":   {"cpc_subgroup_id": cpc_prefix}},
            ]},
            "f": ["patent_number"],
            "o": {"per_page": 1},  # only need total count
        }
        resp  = requests.post(url, json=payload)
        total = resp.json().get("total_patent_count", 0)
        records.append({"assignee": assignee, "patent_count": total})
        print(f"  {assignee}: {total} patents")

    df = pd.DataFrame(records).sort_values("patent_count", ascending=True)
    return df

companies = ["Genentech", "Amgen", "Regeneron", "AstraZeneca", "Novartis"]
df_comp   = count_patents_by_assignee(companies, cpc_prefix="A61K")

fig, ax = plt.subplots(figsize=(7, 4))
ax.barh(df_comp["assignee"], df_comp["patent_count"], color="salmon")
ax.set_xlabel("Patent count (A61K)")
ax.set_title("Pharmaceutical Patents by Assignee (CPC A61K)")
plt.tight_layout()
plt.savefig("assignee_comparison.png", dpi=150)
print("Comparison chart saved -> assignee_comparison.png")

预期输出

• 包含专利记录的 pd.DataFrame（列取决于请求的 "f" 字段）
• cpc_trend.png — 按年份统计专利数量的柱状图
• assignee_comparison.png — 对比各公司的横向柱状图
• API 响应中的 total_patent_count 给出查询对应的完整语料规模

故障排除

问题	原因	解决方案
HTTPError 429 Too Many Requests	超出每分钟 45 次请求的速率限制	在请求之间添加 time.sleep(1.5)；申请免费的 API 密钥
响应中的 patents 列表为空	查询范围太窄或字段名称不正确	检查 PatentsView API 文档中的字段名称；先在网页界面中测试查询
结果中缺少已知专利	对受让人名称进行精确字符串匹配	使用 _contains 而不是 _eq；检查名称变体
KeyError: patent_date	"f" 列表中未请求该字段	将 "patent_date" 添加到 "f" 数组
BigQuery 身份验证错误	未配置 GCP 凭据	运行 gcloud auth application-default login 或设置 GOOGLE_APPLICATION_CREDENTIALS
CPC 前缀未返回结果	CPC 代码无效或有拼写错误	在 CPC 分类浏览器中验证代码

参考资料

• PatentsView API 文档 — 完整字段列表和查询运算符参考
• PatentsView API 浏览器 — 用于测试的浏览器端查询构建器
• Google Patents 公共数据（BigQuery） — 架构文档
• CPC 分类浏览器 — 浏览生命科学领域的 CPC 代码
• USPTO 批量数据存储系统 — 专利全文 XML 下载