Difference between revisions of "Reproducible Patent Data"

From edegan.com
Jump to navigation Jump to search
Line 271: Line 271:
* Run the schema creation
* Run the schema creation
* Create an enum with the same names for attributes as in the DDL (case-insensitive! prefer all-caps screaming snake case)
* Create an enum with the same names for attributes as in the DDL (case-insensitive! prefer all-caps screaming snake case)
* Create a class which subclasses <code>AbstractIntertableData</code>
* Create a class which subclasses <code>AbstractInsertableData</code>
* Inside that class, create a static class which subclasses <code>AbstractTableMetadata</code> and has the proper values for getTableName(), getStringColumns(), getIntColumns()
* Inside that class, create a static class which subclasses <code>AbstractTableMetadata</code> and has the proper values for getTableName(), getStringColumns(), getIntColumns()
* (Optional) Implement builder pattern
* (Optional) Implement builder pattern

Revision as of 11:47, 14 November 2017

McNair Project
Reproducible Patent Data
Uspto web logo.jpg
Project Information
Project Title Reproducible Patent Data
Owner Oliver Chang
Start Date May 17
Primary Billing
Has project status Active
Subsumes: Redesigning Patent Database, Patent Assignment Data Restructure
Copyright © 2016 edegan.com. All Rights Reserved.

A continuation of Redesigning Patent Database that aims to write faster, more centralized code to deal with data from the United States Patent and Trademark Office (USPTO). By having an end-to-end pipeline we can easily reproduce or update data without worrying about unintentional side effects or missing data.


To get up and running with the code, do the following:

  1. Clone the git project (link at end of page) to your user directory
  2. Launch IntelliJ with >= Java 8 and Maven configured (default version installed on the RDP is setup to do this)
  3. Open project in IntelliJ
  4. Create an empty database (see #Database)
  5. Run the table creation scripts in src/db/schemas/ in your new database
  6. Modify the constant DATABASE_NAME in E:\McNair\Projects\SimplerPatentData\src\main\java\org\bakerinstitute\mcnair\postgres\DatabaseHelper.java
  7. Run the Driver scripts in IntelliJ with the correct value for DATA_DIRECTORY (or run RunInitialImport.java which will do all of the data directories for that patent item type)
  8. [Take a really, really long lunch...in total should take no more than five hours to load data on RDP]
  9. Run scripts in src/db/constraints to check data assumptions
  10. That's it!

Directory Layout

Where is the Data?


All of the information for this project is located at E:\McNair\Projects\SimplerPatentData

There are several interesting directories:

  • data/downloads/ is USPTO bulkdata, unmodified straight from the scraper
  • data/extracts/ is a directory of a strict subset of the information stored in data/downloads/. It is the result of running a bulk 7-zip job on that directory to get everything unzipped in a flat data structure. Note that these files have the USPTO modified-by time since that metadata is stored in the zipfiles. To extract files in this nice format, select all of the zipfiles and setup an extraction job like in this screenshot
  • data/backups/ is a 7zip'd backup of the corresponding directory in extracts
  • src/ is the main code repository for the java project

Input Files

All of the text-only Red Book files for granted patents from 1976 to 2016, inclusive. To find a specific year's XML file, find it in


To find application data from 2001 to 2016, inclusive, look in


To find assignment data, look in


To find maintenance fee data, look in


Where is the Code?

The code has the same parent directory as the data, so it is at E:\McNair\Projects\SimplerPatentData\src. You might notice a lot of single-entry directories; this is an idiomatic Java pattern that is used for package separation. If using IntelliJ or some other IDE, these directories are a bit less annoying.

The development environment is Java 8 JDK, IntelliJ Ultimate IDE, Maven build tools, and git VCS.

The git repository can be found at http://rdp.mcnaircenter.org/codebase/Repository/ReproduciblePatent

Prior Art

This tool is not so concerned with adding new functionality; rather, it aims to take a bunch of spread out Perl scripts and create a faster system that is easier to work with. As such, its functionality is largely stolen from those scripts:

  • Downloader: E:\McNair\Software\Scripts\Patent\USPTO_Parser.pl
  • XML Splitter: E:\McNair\PatentData\splitter.pl
  • XML Parsing: E:\McNair\PatentData\Processed\xmlparser_4.5_4.4_4.3.pl and E:\McNair\PatentData\Processed\*.pm

In addition, I used several non-standard Java libraries listed below:

If using maven, these dependencies are listed and should automatically be setup.

Using Code

Any file with a line that says public static void main(String[] args) { can be run as a standalone file. The easiest way to do this is to load the project and then the file in IntelliJ and click the little green play arrow next to this bit of code.

The code can also be run via the standard javac and java commands but since this project has a complicated structure you end up having to run commands like

"C:\Program Files\Java\jdk1.8.0_131\bin\java" "-javaagent:C:\Users\OliverC\IntelliJ IDEA 2017.1.3\lib\idea_rt.jar=62364:C:\Users\OliverC\IntelliJ IDEA 2017.1.3\bin" -Dfile.encoding=UTF-8 -classpath "[...contents truncated...];C:\Users\OliverC\.m2\repository\org\postgresql\postgresql\42.1.1\postgresql-42.1.1.jar" org.bakerinstitute.mcnair.uspto_assignments.XmlDriver

to include all of the runtime dependencies and it's just not worth it.

Altering Code

  • Use the IntelliJ command Reformat code (found in the menus at Code > Reformat Code
  • Use the optimize imports function found under the same menu
  • Use spaces for indentation
  • Loosely try to keep lines below 120 characters
  • Commit changes to the Git remote repository "bonobo"

Schema Reconciliation

For the work by Joe, see the Patent Schema Reconciliation page

Patents (Granted)

See E:\McNair\Projects\SimplerPatentData\data\examples\granted for extracted examples of what specific data is available for a sample of the data.

Granted Patent Data Formats
Dates Used Format Location Supported? Utility Reissue Design Plant
January 1976 to December 2001 APS data/extracts/granted/vintage Yes ~ ~ ~
January 2001 to December 2001 SGML Ignored; use concurrently recorded APS data No N/A N/A N/A N/A
January 2002 to December 2004 XML Version 2.5 data/extracts/granted/blunderyears Yes ~ ~ ~
January 2005 to December 2005 XML Version 4.0 ICE data/extracts/granted/modern Yes ~ ~ ~
January 2006 to December 2006 XML Version 4.1 ICE data/extracts/granted/modern Yes ~ ~ ~
January 2007 to December 2012 XML Version 4.2 ICE data/extracts/granted/modern Yes ~ ~ ~
January 2013 to September 24, 2013 XML Version 4.3 ICE data/extracts/granted/modern Yes ~ ~ ~
October 8, 2013 to December 2014 XML Version 4.4 ICE data/extracts/granted/modern Yes ~ ~ ~
January 2015 to December 2016 XML Version 4.5 ICE data/extracts/granted/modern Yes ~ ~ ~

APS Rosetta Stone

The Advanced Patent System (APS) is a fixed-width text format used to store historical patent grant data. The documentation for this sucks; there are pages missing at random. Luckily, we only care about the content contained here: File:PatentFullTextAPSDoc GreenBook pgs13-22.pdf.

It's worth mentioning that the APS contains an advanced text markup system for chemical formulae, basic text markup, tables, etc. that can lead to seemingly garbled text that is perfectly well-formed.

APS Gotchas

  • PATN.WKU is the granted patent number. It is 7 digits while the spec promises 6 digits. The rightmost digit is a check digit modulus 11. See File:Aps-wku-modulus11.pdf for the words from the horse's mouth.

Patents (Applications)

Patent Application Data Formats
Dates Used Format Location Supported by Parser?
March 15, 2001 to December 2001 XML Version 1.5 data/extracts/applications/vintage Yes, for basic information, inventors, and correspondents
January 2002 to December 2004 XML Version 1.6 data/extracts/applications/vintage Ditto
January 2005 to December 2005 XML Version 4.0 ICE data/extracts/applications/modern Ditto
January 2006 to December 2006 XML Version 4.1 ICE data/extracts/applications/modern Ditto
January 2007 to December 2012 XML Version 4.2 ICE data/extracts/applications/modern Ditto
January 2013 to December 2014 XML Version 4.3 ICE data/extracts/applications/modern Ditto
January 2015 to Present XML Version 4.4 ICE data/extracts/applications/modern Ditto


Because there isn't a compelling reason not to, I used the existing PostgreSQL infrastructure on the RDP. The "Java Way" of interacting with databases is the Java Database Connectivity API (JDBC), an implementation-agnostic API for interacting with databases. This project uses the stock Postgres JDBC, version 42.1.1

Create an empty database on RDP

To create an empty database, run this command: $ createdb --username=postgres database-name-goes-here # password is tabspaceenter

Abstraction Layer

Since writing raw SQL is a bit cumbersome and error-prone, I have added some abstraction layers that make it much easier to quickly add bulk data. By using Postgres's CopyManager class, we buffer SQL copy commands in memory (as many as possible) and then flush these rows. To understand how the abstraction layers work, see the code in E:\McNair\Projects\SimplerPatentData\src\main\java\org\bakerinstitute\mcnair\postgres. See E:\McNair\Projects\SimplerPatentData\src\main\java\org\bakerinstitute\mcnair\models\GrantedPatent.java for an example of how to extend the abstraction layer to deal with more complex scenarios.

New Table Checklist

  • Create schema DDL SQL code for the new table in E:\McNair\Projects\SimplerPatentData\src\db
  • Run the schema creation
  • Create an enum with the same names for attributes as in the DDL (case-insensitive! prefer all-caps screaming snake case)
  • Create a class which subclasses AbstractInsertableData
  • Inside that class, create a static class which subclasses AbstractTableMetadata and has the proper values for getTableName(), getStringColumns(), getIntColumns()
  • (Optional) Implement builder pattern
  • (Optional) Create a custom databasehelper for complex extras (see PatentApplication and GrantedPatent for examples)
  • Write the data to the table (see DatabaseHelper for the pattern I use)

Address Data

To get the most granular address data (street level, or at least postcode level) about who owns patents, the path is not so straightforward because off the complicated mapping of ownership to a granted patent. This is the final part of this project that I am working on and it is all at the level of SQL. See E:\McNair\Projects\SimplerPatentData\src\db\joins for my attempts to create a clean mapping. Optimistically speaking, the data generated here should be superset of the data present in the Patent Assignment Data Restructure project.

Note that as of the beginning of August 2017, this part has not been completed.


Use assignments_longform.last_update_date to find current/latest (or first/earliest) date of assignment. Then match with properties.docid on reelno, frameno to find patent application id. With this mapping to granted patents, we can discover the details of the original granted patent. And with the right date and reelno and frameno, we can match to the assignees table and get fine granularity addresses.

Related Pages

External Links