XML Functionality Inside CLAIMS Direct Data Warehouse

Blog post by admin


The CLAIMS Direct Web Services (CDWS) offer a variety of entry points into both the data warehouse and Solr index. These are mid-to-high-level entry points and can satisfy most requirements pertaining to searching and extracting data for a typical search/view application. There are, however, corner cases which may require more intricate extraction of particular information. On the other hand, there may also be situations where massive amounts of data need to be extracted for further, down-stream processing. The following discussion will touch on solutions for both of these cases.

Basic Data Warehouse Table Structure

The basic structure of the PostgreSQL database housing the XML content is best described as a collection of tables containing the raw XML of every patent document available in CLAIMS Direct. Each table is a section of the XML document, referred to as a container, and each table is named according to the XML container data it contains, e.g.,

xml.t_publication_referencepublication-referenceDocument publication information
xml.t_invention_titleinvention-titleDocument title(s)
xml.t_claimsclaimsDocument claims

The structure of each table is identical with the exception of xml.t_patent_document_values which functions as a meta table.

<container>_idserial primary key
publication_idinteger not null
modified_load_idinteger not null
statuscharacter(1) (v=valid XML, i=invalid XML)

The XML is always in the content (PostgreSQL type: XML) column.

select content
from xml.t_publication_reference
where publication_id=xml.f_ucid2id('US-5551212-A');
<publication-reference fvid="71329085" ucid="US-5551212-A">

The invaluable utility function xml.f_ucid2id takes a ucid and returns a publication_id.

Extracting Pieces of the XML Content

As mentioned above, all content columns are of type XML and therefore the internal PostgreSQL xpath functionality can be used. There are a variety of utility functions provided by CLAIMS Direct that mimic DOM functionality, e.g.




Extract text of node:

select xml.f_textContent('./document-id/lang', content)
from xml.t_publication_reference
where publication_id=xml.f_ucid2id('US-5551212-A');



Extract single attribute value:

select xml.f_getAttribute('ucid', content)
from xml.t_publication_reference
where publication_id=xml.f_ucid2id('US-5551212-A');



Test presence of a particular node:

select xml.f_nodeExists('//date', content)
from xml.t_publication_reference
where publication_id=xml.f_ucid2id('US-5551212-A');



Select a node (type XML) satisfying the given xpath expression:

select xml.f_findNode('./main-classification', content)
from xml.t_classification_national
where publication_id=xml.f_ucid2id('US-5551212-A');
 <main-classification mxw-id="PCL893513162" load-source="mcf">053397</main-classification>



Select an array of nodes satisfying the given xpath expression:

select xml.f_findNodes('./classification-ipcr', content)
from xml.t_classifications_ipcr
where publication_id=xml.f_ucid2id('US-5551212-A');
  "<classification-ipcr ...",
  "<classification-ipcr ...",
  "<classification-ipcr ..."



Select an array of nodes based on element name:

select xml.f_getElementsByTagName('classification-ipcr', content)
from xml.t_classifications_ipcr
where publication_id=xml.f_ucid2id('US-5551212-A');
  "<classification-ipcr ...",
  "<classification-ipcr ...",
  "<classification-ipcr ..."



Select the name of the content node:

select xml.f_nodeName( content )
from xml.t_invention_title
where publication_id=xml.f_ucid2id('US-5551212-A');



Select all child nodes:

select xml.f_childNodes( content )
from xml.t_application_reference
where publication_id=xml.f_ucid2id('US-5551212-A');
 <document-id mxw-id="PAPP60364944" load-source="docdb" format="epo">
 <document-id mxw-id="PAPP84571955" load-source="patent-office" format="original">



Concatenate all child text nodes:

select xml.f_node2string( content::text, 1 )
from xml.t_abstract
where publication_id=xml.f_ucid2id('US-5551212-A');
 A package is formed by holding a nonextensible strip taut and wrapping it around the upper and/or lower periphery of the bundle and joining the ends of the strip segments e.g. by adhesive bonding. A projecting margin of the strip is folded to overlie the bundle or underlie the bundle so that an L-configuration is imparted to the frame formed around the bundle after the strip is applied thereto.

Use Case: Creating relational view of XML content

Although the underlying architecture of the CLAIMS Direct PostgreSQL database is a data warehouse, creating normalized relational views into the XML content is easily achievable using the functions described above. Our example will be to create a relational view of all patent citations. This view will let us SELECT based on criteria as well as being able to GROUP BY and ORDER BY. The first step is to define all the properties of a citation:

  • ref-ucid (cited document)
  • ref-publication-id
  • publication-date
  • country
  • kind
  • load-source
  • format
  • source (examiner or applicant)

Now we can create a function that returns a TABLE of these properties. We'll use publication_id as input to the function and create it in a private schema (mySchema).

create or replace function mySchema.f_rdb_citations( integer )
  returns table (
    publication_id integer,
    ref_publication_id integer,
    ref_ucid varchar(64),
    published date,
    country varchar(2),
    kind varchar(2),
    load_source varchar(32),
    format varchar(32),
    source varchar(32)
  ) as
declare v_content xml;
        v_node xml;
        v_doc_node xml;
-- example usage:
-- select * from mySchema.f_rdb_citations( xml.f_ucid2id('US-9240001-B2') );
  select x.publication_id, x.content into publication_id, v_content
    from xml.t_citations as x where x.publication_id=$1;
  if not found then
  end if;
  -- loop through all patent citations
  for each v_node in array
    ( select xml.f_findNodes('//patcit', v_content) ) loop
    select xml.f_getAttribute( 'ucid', v_node ) into ref_ucid;
    select x.publication_id into ref_publication_id
      from xml.t_patent_document_values as x
      where ucid=xml.f_getAttribute( 'ucid', v_node );
    select xml.f_getAttribute( 'load-source', v_node ) into load_source;
    select xml.f_getAttribute('name', xml.f_findNode('./sources/source', v_node ) ) into source;
    -- loop through all forms of the document-id(s)
    for each v_doc_node in array
      ( select xml.f_findNodes('./document-id', v_node) ) loop
      select xml.f_getAttribute( 'format', v_doc_node ) into format;
      select xml.f_textContent( './country', v_doc_node ) into country;
      select xml.f_textContent( './kind', v_doc_node ) into kind;
      select xml.f_textContent( './date', v_doc_node ) into published;
      return next;
    end loop;
  end loop;
  COST 100
  ROWS 1000;

Now, we test the results:

select * from mySchema.f_rdb_citations( xml.f_ucid2id('US-9240001-B2') ) limit 5;
 publication_id | ref_publication_id |     ref_ucid      | published  | country | kind | load_source | format | source
      160817649 |            5778519 | CA-2580978-A1     | 2006-03-30 | CA      | A1   | docdb       | epo    | APP
      160817649 |          127402642 | CN-201918032-U    | 2011-08-03 | CN      | U    | docdb       | epo    | APP
      160817649 |           75847915 | US-20020150866-A1 | 2002-10-17 | US      | A1   | docdb       | epo    | APP
      160817649 |           75947470 | US-20030060942-A1 | 2003-03-27 | US      | A1   | docdb       | epo    | APP
      160817649 |           76046244 | US-20030154010-A1 | 2003-08-14 | US      | A1   | docdb       | epo    | APP

Of course, GROUP BY and ORDER BY are available on the returned columns. In addition to SELECTing citations for a single document, we can also JOIN to table expressions, e.g., assume we are interested in the top 5 cited countries from US grant publication date 2016-05-03:

select count(*) as n, t2.country as ref_country
from xml.t_patent_document_values t1
  inner join mySchema.f_rdb_citations (t1.publication_id) as t2
    on (t1.publication_id=t2.publication_id)
where t1.country='US'
  and t1.published='2016-05-03'
group by t2.country
order by n desc
limit 5;
   n    | ref_country
 221800 | US
  14285 | WO
  13313 | JP
   7115 | EP
   2554 | CN