Patent Data Processing - SQL Steps
Contents
Objective
The McNair Center owns two sets of patent data - one set that is inherited from Harvard, the Harvard dataverse, which is stored in the database patentdata and another that is generated by crawlers pulling data from the USPTO website, which is stored in the database patent_2015.
We are now merging and cleaning the two data sets, and storing them in a schema that is amalgamation of the two underlying schema for the citations tables, assignees tables, and patents tables. The destination schema is allpatent.
Assignees Data
The schema for the assignees table in patentdata database is:
Column | Type | Modifiers -------------+-------------------+----------- patent | integer | asgtype | integer | assignee | character varying | city | character varying | state | character varying | country | character varying | nationality | character varying | residence | character varying | asgseq | integer |
The schema for the assignees table in patent_2015 is :
Column | Type | Modifiers --------------+---------+----------- lastname | text | firstname | text | orgname | text | city | text | country | text | patentcountry | text | patentnumber | integer | state | text | address | text | postcode | text |
To merge both schemas, we have some columns that overlap, and some columns that don't.
Overlapping Columns
patent_2015 | patentdata --------------+-------------- orgname | assignee city | city country | country patentnumber | patent state | state
These columns will have entries for most rows in the table, because they exist in both tables. The rest of the columns will be populated based on which table the row is coming from.
Final Schema
Table "public.assignees" Column | Type | Modifiers ---------------+-------------------+----------- lastname | character varying | firstname | character varying | address | character varying | postcode | character varying | orgname | character varying | city | character varying | country | character varying | patentnumber | integer | state | character varying | patentcountry | character varying | nationality2 | character varying | residence | character varying | asgseq | integer | asgtype | integer |
Non-overlapping Columns These are the columns that belong to either one of the assignees tables, and not to both. For these cases, to help users understand where the row is coming from, the following insert rules have been followed:
- For columns of type int, insert -1
- For columns of type string (character varying), the string 'null' has been inserted.
Therefore, if a row has appropriate values for orgname, state, city ,etc, but 'null' values for lastname, firstname, address and postcode, the row has come from the patentdata table.
Index
Since the table is relatively large, and is likely to be searched often, an index has been imposed on the table.
allpatent=# CREATE INDEX ON assignees (orgname); CREATE INDEX
Sample insert and copy commands
INSERT INTO assignees_merge ( SELECT 'null', 'null', 'null', 'null', a.assignee, a.city, a.country, a.patent, a.state, 'null', a.nationality, a.residence, a.asgseq, a.asgtype FROM assignees a );
INSERT INTO assignees_merge ( SELECT assignees.lastname, assignees.firstname, assignees.address, assignees.postcode, assignees.orgname, assignees.city, assignees.country, assignees.patentnumber, assignees.state, assignees.patentcountry, 'null', 'null', -1, -1 FROM assignees );
\COPY assignees_merge TO '/tmp/assignees_merge_export.txt' DELIMITER AS E'\t' HEADER NULL AS CSV;
\COPY assignees FROM '/tmp/assignees_merge_export.txt' DELIMITER AS E'\t' HEADER NULL AS CSV;
--1607724
\COPY assignees_merge TO '/tmp/assignees_merge_export1.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; \COPY assignees FROM '/tmp/assignees_merge_export1.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; --3818842
Note : The assignees table was updated on 6/23 to remove the 'null' string and the '-1' values.
Patents
Patentdata Schema:
Column | Type | Modifiers --------+-------------------+----------- patent | integer | kind | character varying | claims | integer | apptype | integer | appnum | integer | gdate | date | gyear | integer | appdate | date | appyear | integer |
Patent_2015 Schema:
Column | Type | Modifiers -------------------------------+---------+----------- patentnumber | int | not null kind | varchar | grantdate | date | type | varchar | applicationnumber | varchar | filingdate | date | prioritydate | date | prioritycountry | varchar | prioritypatentnumber | varchar | ussubclass | varchar | maingroup | varchar | subgroup | varchar | cpcsubclass | varchar | cpcmaingroup | varchar | cpcsubgroup | varchar | classificationnationalcountry | varchar | classificationnationalclass | varchar | title | varchar | numberofclaims | int | primaryexaminerfirstname | varchar | primaryexaminerlastname | varchar | primaryexaminerdepartment | varchar | pctpatentnumber | varchar | filename | varchar |
Overlapping Columns
patent_data patent_2015 --------------+------------- patent | patentnumber kind | kind claims | numberofclaims apptype | type appnum | applicationnumber gdate | grantdate appdate | filingdate
Combined Schema:
The final schema of the patents table is :
Column | Type | Modifiers ----------------------+-------------------+----------- patent | integer | not null grantdate | date | prioritydate | date | prioritycountry | character varying | prioritypatentnumber | character varying | cpcsubgroup | character varying | pctpatentnumber | character varying | claims | integer | appnum | integer | gyear | integer | appdate | date | appyear | integer | nber | integer | uspc | character varying | uspc_sub | character varying |
From the total list of columns belonging to both the tables (patentdata and patent_2015), a few columns, most of them related to classification of patents, have been dropped since the data in the tables was not clean.
Additionally, three columns - nber, uspc, uspc_sub have been added from the historicalpatentdata, a table built from data downloaded from the USPTO Bulk Data Storage. The join was executed on the patent number.
Note : The addition, deletion of columns as through separate scripts, therefore the scripts below will be slightly discrepant.
Index and Key Creation
Patent numbers are distinct in this table, and are central to the rest of the fields in the table. A primary key can therefore be imposed on the column. Also, since a number of searches are likely to be conducted on this table, an index has been imposed as well.
Code:
ALTER TABLE patents ADD PRIMARY KEY (patentnumber); -- RESULT : ALTER TABLE allpatent=# CREATE UNIQUE INDEX patent_idx ON patents (patentnumber);
Sample Insert and Copy Statements
patentdata: INSERT INTO patents_merged ( SELECT patent,
kind, gdate, 'NULL', 'NULL', NULL, NULL, 'NULL', 'NULL', 'NULL', 'NULL', 'NULL', 'NULL', 'NULL', 'NULL', 'NULL', 'NULL', 'NULL', -1, 'NULL', 'NULL', 'NULL', 'NULL', 'NULL', claims, apptype, appnum, gyear, appdate,
appyear FROM patents ); -- RESULT : INSERT 0 3984771 patent_2015: INSERT INTO patents_merged ( SELECT patentnumber,
kind, grantdate, type, applicationnumber, filingdate, prioritydate, prioritycountry, prioritypatentnumber, ussubclass, maingroup, subgroup, cpcsubclass, cpcmaingroup, cpcsubgroup, classificationnationalcountry, classificationnationalclass, title, numberofclaims, primaryexaminerfirstname, primaryexaminerlastname, primaryexaminerdepartment, pctpatentnumber, filename, -1, -1, -1, -1, NULL,
-1 FROM patents ); -- RESULT : INSERT 0 1646225
COPY SCRIPTS: patentdata:
\COPY patents_merged TO '/tmp/merged_patents_export1.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; --COPY 3984771
patent_2015:
\COPY patents_merged TO '/tmp/merged_patents_export.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; --COPY 1646225
PATENTS TABLE
\COPY patents FROM '/tmp/merged_patents_export1.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; -- RESULT : COPY 3984771 \COPY patents FROM '/tmp/merged_patents_export.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; -- RESULT : COPY 1646225
TESTING
select count(*) FROM (SELECT DISTINCT patentnumber FROM patents) AS t; --RESULT: 5411151 EXPECTED: 5426566
We found some copies of a few rows, where both the patent_2015 and patentdata
SELECT COUNT(*), * FROM patents GROUP BY patentnumber,
kind, grantdate, type, applicationnumber, filingdate, prioritydate, prioritycountry, prioritypatentnumber, ussubclass, maingroup, subgroup, cpcsubclass, cpcmaingroup, cpcsubgroup, classificationnationalcountry, classificationnationalclass, title, numberofclaims, primaryexaminerfirstname, primaryexaminerlastname, primaryexaminerdepartment, pctpatentnumber, filename, claims, apptype, appnum, gyear, appdate, appyear
HAVING COUNT(*) > 1;
SELECT patentnumber, count(*) FROM patents GROUP BY patentnumber HAVING count(*)>1; --7640598
SELECT * FROM patents op WHERE op.patentnumber IN
( SELECT ip.patentnumber FROM patents ip GROUP BY ip.patentnumber HAVING COUNT(*)>1 )
ORDER BY op.patentnumber;
( SELECT * INTO patentsCleaned FROM patents op WHERE op.patentnumber IN
( SELECT ip.patentnumber FROM patents ip GROUP BY ip.patentnumber HAVING COUNT(*)=1 )
ORDER BY op.patentnumber ) --SELECT 5191306
INSERT INTO patentsCleaned( SELECT * FROM patents op WHERE op.patentnumber IN
( SELECT ip.patentnumber FROM patents ip GROUP BY ip.patentnumber HAVING COUNT(*)>1 )
AND op.applicationnumber NOT LIKE 'NULL' ORDER BY op.patentnumber );
--219845
TESTING:
allpatent=# select count(*) from patentsCleaned; count --------- 5411151 (1 row)
allpatent=# select count(*), patentnumber FROM patentsCleaned group by patentnumber having count(*) > 1; count | patentnumber -------+-------------- (0 rows)
Citations
In the citations table, we needed to define another function that would convert a textual patent number into a number (big int, since the patents number were exceeding the range of regular integers.)
To Extract Patents with Numbers Only and to Ignore Other RegExes
CREATE OR REPLACE FUNCTION cleanpatno (text) RETURNS bigint AS $$
if ($_[0]) { my $var=$_[0]; if ($var=~/^\d*$/) {return $var;} return undef; } return undef; $$ LANGUAGE plperl;
patentdata schema:
Column | Type | Modifiers ------------+-------------------+----------- patent | integer | cit_date | date | cit_name | character varying | cit_kind | character varying | cit_country | character varying | citation | integer | category | character varying | citseq | integer |
SELECT patent as citingpatentnumber, citation AS citedpatentnumber INTO citations_merged FROM citations; --SELECT 38452957
patent_2015 schema:
Column | Type | Modifiers ---------------------+---------+----------- citingpatentnumber | integer | citingpatentcountry | text | citedpatentnumber | text | citedpatentcountry | text |
SELECT CAST(citingpatentnumber AS bigint), CAST(cleanpatno( citedpatentnumber) AS bigint) as citedpatentnumber INTO citations_merged FROM citations; -- RESULT : SELECT 59227881
Overlapping Columns
patent_2015 | patentdata | ---------------------+---------------+ citingpatentnumber | patent | citedpatentnumber | citation |
Combined Schema:
Column | Type | Modifiers
+--------+-----------
citingpatentnumber | bigint | citedpatentnumber | bigint |
Copy Statements:
patentdata:
\COPY citations_merged TO '/tmp/merged_citations_export1.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; --COPY 38452957
patent_2015:
\COPY citations_merged TO '/tmp/merged_citations_export.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; -- RESULT : COPY 59227881
allpatent:
\COPY citations FROM '/tmp/merged_citations_export.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; --RESULT : COPY 59227881
\COPY citations FROM '/tmp/merged_citations_export1.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; --RESULT: COPY 38452957
CLONING:
CREATE DATABASE allpatentsProcessed WITH TEMPLATE allpatent OWNER researcher;
USPTO Consolidated Patent Data
Scripts:
/* creating patent data tables from : https://bulkdata.uspto.gov/data2/patent/maintenancefee/*/
CREATE TABLE PatentMaintenanceFee( patentnumber varchar, applicationnumber int, smallentity varchar, filingdate date, grantissuedate date, maintenancefeedate date, maintenancefeecode varchar );
\COPY PatentMaintenanceFee FROM '/bulk/USPTO_Consolidated/MaintFeeEvents_20160613.txt' DELIMITER AS E'\t' HEADER NULL AS CSV; -- RESULT : COPY 14042059
/* creating tables for historical patent data - USPTO */
CREATE TABLE HistoricalPatentData( applicationid int, pubno varchar, patentnumber varchar, NBER int, USPC varchar, USPC_sub varchar, applicationdate date, prioritydate date, pubdate date, displaydate date, disptype varchar, exp_dt date, exp_dt_max date, pta int );
\COPY historicalpatentdata FROM '/bulk/USPTO_Consolidated/HistoricalFiles/historical_masterfile.csv' DELIMITER AS ',' HEADER NULL AS CSV;
--COPY 11191813