Chinese patents data coverage in PATSTAT

Following my recent post about chinese patents data, I had a a deeper look into SIPO (chinese patent office) and organized data by year (see table below) and compared them against PATSTAT 10/2009 data.
We can understand that, leaving aside designs, that data coverage is usually over 90% on applied patents.

Year	Total	Inv + utility	Invention	Utility Model	Design	PATSTAT	% coverage
1985	14.372	13.732	8.558	5.174	640	12.532	91,26
1986	18.509	17.682	8.009	9.673	827	15.992	90,44
1987	26.077	24.765	8.059	16.706	1.312	22.411	90,49
1988	34.011	32.052	9.652	22.400	1.959	27.388	85,45
1989	32.905	30.386	9.659	20.727	2.519	27.215	89,56
1990	41.469	37.752	10.137	27.615	3.717	32.090	85,00
1991	50.040	44.705	11.423	33.282	5.335	37.713	84,36
1992	67.135	58.778	14.409	44.369	8.357	48.832	83,08
1993	77.276	67.117	19.618	47.499	10.159	54.916	81,82
1994	77.735	64.578	19.067	45.511	13.157	57.315	88,75
1995	83.045	65.377	21.636	43.741	17.668	60.224	92,12
1996	102.735	78.121	28.517	49.604	24.614	67.837	86,84
1997	114.208	83.795	33.666	50.129	30.413	73.285	87,46
1998	121.989	87.357	35.960	51.397	34.632	79.939	91,51
1999	134.239	94.186	36.694	57.492	40.053	91.110	96,73
2000	170.682	120.562	51.747	68.815	50.120	112.283	93,13
2001	203.573	142.926	63.204	79.722	60.647	132.884	92,97
2002	252.631	173.371	80.232	93.139	79.260	165.357	95,38
2003	308.487	214.433	105.318	109.115	94.054	204.918	95,56
2004	353.807	242.958	130.133	112.825	110.849	234.252	96,42
2005	476.264	312.893	173.327	139.566	163.371	286.538	91,58
2006	573.178	371.856	210.490	161.366	201.322	337.228	90,69
2007	693.917	426.485	245.161	181.324	267.432	354.732	83,18
2008	828.328	515.424	289.838	225.586	312.904	255.837	49,64

The delusion comes from the quality of data coverage...

When examining the applicants data out of 776607 applicant none has a zip code or a city.
Only 3161 have an address (where address city zip are packed altogether) but all of them are non-chinese.
So no geographic analisys is possible.

One more thing to be careful about are counts by publication numbers: FI if you look for patent # CN1237182  you will find 2 different patents: the first with status A (unexamined) and the latter with status C (granted).

So we should clarify the meaning of Publication number. According to SIPO, there is more than one kind of publication numbers which are reflected by the Kind Code:

  ‘A’: Publication of Unexamined application for patent for invention

  ‘C’: Publication of Granted patent for Invention

‘Y’: Publication of Granted patent for Utility Model

‘D’: Publication of Granted patent for Design

Last but not least: IPCs.

Out of 3.270.000 published patents 1.136.000 have no indication of primary ipc class.

So we're going to use data straight from SIPO to build a chinese patents database...

(bytheway, if you are willing to search into sipo website, this is the link to english search page:
http://218.240.13.210/sipo_EN/search/tabSearch.do?method=init )

IPC: core vs advanced level in patstat

As very vell explained from EPO website (here) IPC classification contains over 70.000 entries, that give too much detail for most users.
That's the reason why aside from this advanced classification has been introduced a core classification containing "only" 20.000 entries.
The advanced level is a superset of the core level, ie it consists of all core level groups and many finer subdivisions, allows updates to be made every three months, which means it will grow considerably in the years ahead.
In PCT minimum documentation have an advanced-level IPC. They also have a core level IPC that is derived from the advanced level.

In patstat we can distinguish advanced and core levels by the content (A or C) of field IPC_CLASS_LEVEL in table tls209_appln_ipc.

I ran a query in order to understand how A & C IPCs are distributed across the various patent offices (see the SQL below) expecting a 1:1 distribution.

SELECT tls201_appln.APPLN_AUTH, tls209_appln_ipc.IPC_CLASS_LEVEL,  COUNT(tls209_appln_ipc.APPLN_ID)
FROM patstat.tls209_appln_ipc
INNER JOIN patstat.tls201_appln ON tls209_appln_ipc.APPLN_ID = tls201_appln.APPLN_ID
GROUP BY tls201_appln.APPLN_AUTH, tls209_appln_ipc.IPC_CLASS_LEVEL
ORDER BY tls201_appln.APPLN_AUTH, tls209_appln_ipc.IPC_CLASS_LEVEL;

The full results of the query may be downloaded here; bytheway I publish here a graph of patenting office with more than 850.000 C and A IPCs: in red you may find the count of CORE ipcs, in blu the ADVANCED ones.

At a glance we notice that, apart from Slovenia Patent office (SI), everywhere the figure of advanced IPCs is the higher, average above 10%.
The peak of difference is to be found in some african countries (34% Kenia, 36% Malawi) but also in Iceland advanced ipcs are 31% more than core ones. Same trend in US (11%), EP (15%) and  JP (12%).

European Classification scheme to be released in patstat april 2010

As announced at the Patent Statistics conference in Vienna 2009, PATSTAT should get an update by getting European Classification scheme (ECLA).

Recently Geert Boedt , from patstat helpdesk, announced in patstat forum this change will take palce in april 2010.

The way it will happen is through table TLS217_APPLN_I_CLS (contining previously national classifications like for MX, US...) to TLS217_APPLN_ECLA.

Four schemes are planned:

EC - known as ECLA, the European Classification scheme.

ICO- In Computer Only - an internal scheme used by the EPO for classifications which are planned for moving into ECLA at some stage

IDT - Indeling der Techniek , an old Dutch Patent Classification scheme.

ECNO - ECLA symbols which have been allocated to a document by a patent examiner who does not work for the EPO.

Validation of scientific articles with zotero

In order to complete the previous post, I uploaded a couple of videos; they show out how to search and import in zotero a scientific article found in google scholar, and then how to export it in Bibtex (and then straight into Jabref).

I start from an access DB containing a batch of records I had to clean & validate so I decided to use google scholar as parse & validation tool. Then Zotero & Jabref did the rest.

IMPORTING IN ZOTERO FROM GOOGLE SCHOLAR




EXPORTING FROM ZOTERO TO BIBTEX

Creating a scientific literature database with zotero and jabref

If we need to create a database of our own, for scientific literature, without going through the expensive services of Thomson, Elsevier or other data provider, there are some wonderful (and free of cost) tools who may help.

Let's figure we need to create/validate a list of scientific articles references; an easy way could be to use google scholar, by looking for the articole/journal title.
Ok, then?
Then we may use zotero, a firefox extension that help in collecting and managing data of articles and supports export in formats like Bibtex, refer, RIS and many others.
In other words, zotero is nothing but an xtml parser, able to recognize from the webpage what kind of publication (book, article, webpage, proceeding) is facing, and download the relevant informations.

After zotero, in order to handle better and have even more export forrmats (like MySQL fi) I suggest to use jabref, another free tool, that can be feeded from Zotero by exporting data in Bibtex.

With the two of them we can build and handle any dimension of scientific literature DB

EPO patenting activity by NUTS3

Recently I had to create a benchmark for the patenting activity on EPO patents (based on patstat 10/2009) by weighting each patent by inventors' nuts3.

The result was someway amazing 'cause it seems that the TOP25 (I'm publishing below here) has many EU areas in a higher position than the first US area. The first EU area is from Holland, not from Germany as we could expect from te total count of patents...

NUTS3	CY	COUNT	LOCATION
JPC13	JP	102503,082	TOKYO
JPF27	JP	32278,605	OSAKA
NL414	NL	22388,468	EINDHOVEN
JPE23	JP	22307,48	AICHI
FR101	FR	21689,393	PARIS
DE212	DE	16782,779	MUNCHEN
JPC11	JP	16390,328	SAITAMA
JPF28	JP	14888,869	KOBE
FR105	FR	14340,232	BOULOGNE BILLANCOURT
ITC45	IT	14030,154	MILANO
JPC08	JP	13297,203	YOKOHAMA
JPC12	JP	11934,707	CHIBA
25017	US	11879,31	FRAMINGHAM/ACTON
SE010	SE	11489,587	STOCKHOLM
KR013	KR	11158,808	SUWON-SI
JPC22	JP	11031,511	SHIZUOKA
CH040	CH	10752,436	ZURICH
36055	US	10563,428	ROCHESTER
DE300	DE	10329,677	BERLIN
06037	US	10057,301	REDONDO
JPF26	JP	9562,31	KYOTO
06085	US	9378,459	MOUNTAIN VIEW
JPD17	JP	9260,929	YOKOHAMA-SHI
DE115	DE	8883,498	LUDWIGSBURG
FR714	FR	8711,347	GRENOBLE
FR103	FR	8603,364	VERSAILLES

a VBA function counting chars

It may happen to find in some data sources more than one occurence of a data type inside the same field, separated by a special char.
But in many cases that  char may be not so special and be misguiding when trasposing the field.
Let's take as example Thomson SDC database, where in the alliance name you find the alliance participants names separated by the / char.
Obviously the / char may also be integral part of some participant name; so we need to count the occurrence of the / char in order to compare it to the number of partipants indicated in the same record in order to do futher cleaning, if needed.

So I wrote a small DBA function called CHARCOUNT, and I'm posting it here

Function charcount(inputstring As String, findstring As String)
' return the number of occurrences of findstring in inputstring

Dim a$, b$
Dim c, i As Integer

a$ = inputstring
b$ = findstring

c = 0
For i = 1 To Len(a$) - Len(b$) + 1
    If Mid(a$, i, Len(b$)) = b$ Then
        c = c + 1
    End If
Next

charcount = c

End Function
Function charcount(inputstring As String, findstring As String)
' return the number of occurrences of findstring in inputstring

Dim a$, b$
Dim c, i As Integer

a$ = inputstring
b$ = findstring

c = 0
For i = 1 To Len(a$) - Len(b$) + 1
    If Mid(a$, i, Len(b$)) = b$ Then
        c = c + 1
    End If
Next

charcount = c

End Function