AmberBlog

Australia may be better known as a land of beaches, kangaroos, and more recently coal and iron ore mines. But it is also the home of two of the top three patent filing inventors worldwide.  This is the surprising outcome from a list of leading patent global filers compiled on Wikipedia, which provides a list of all inventors known to have filed more than 300 patent families. 

The top ten inventors on this list are:

1) Heading the list of leading inventors is Australian Kia Silverbrook (1958 - ) , who led the development of the Memjet printer. Memjet claims to be the world's fastest printer, using 70,400 jets per printhead to shoot millions of drops per second from a full width printing head to full print colour pages at a rate of up to 60 pages per minute. As of 3 September 2012, Kia Silverbrook was listed as the inventor for 9,727 individual patents or patent applications (Espacenet), 4,457 being granted US patents. 757 of these patents were granted in 2011 alone.

According to Patentbuddy, the patent filed by Kia Silverbrook with the most forward citations is US6439908, Power supply for a four color modular printhead (filed in 2002), with 248 forward citations.

2) In second position is Shunpei Yamazaki (1942 - ) from Japan, who heads the research company Semiconductor Energy Laboratory , which is mainly focused on new display, solar cell and energy storage technologies. Yamazaki has a total of 11,399 patents or patent applications listed in Espacenet, with 2933 of these being granted US patents, including 163 patents granted in 2011. 

The patent filed by Semiconductor Energy Laboratory with the highest forward citation counts is US5,643,826  Method for manufacturing a semiconductor device (1994), with 1098 forward citations.

3)  Third in the list is another Australian, Paul Lapstun (19?? - ). Paul is a colleague of the prolific Kia Silverbrook, and has 3123 patents to his name worldwide, 1200 of these being granted US patents (268 granted in 2011).  The patent listing Paul Lapstun as an inventor with the highest forward citation count is  US6,720,985  Method and system for object selection (2004), with 140 forward citations. 

4) In fourth position is the famed American inventor Thomas Edison (1847-1931). Edison has been credited with 2,332 patents worldwide, 1093 patents in the US. Citation data in electronic form is not easily available, but among his famous of his patents is US223,898 Electric Lamp (1880) , which includes the following claim for an invention that has underpinned incandesent light bulbs every since:

2. The combination of carbon filaments with a receiver made entirely of glass and conductors passing through the glass, and from which receiver the air is exhausted, for the purposes set forth.

5)  In fifth position is the Canadian George Albert Lyon (1882-1961). George Lyon is credited with 993 patents, including the automobile bumper which is disclosed in US1325728 Automobile-buster (1917).

6) Closely behind (#6) is the fellow Canadian Leonard Forbes (1940 - ), with 999 granted US patents out of 1338 total patents, mainly in the semi-conductor area. Among these is US6150687 Memory cell having a vertical transistor with buried source/drain and dual gates (1997), with 245 forward citations.

7) Next on the list is the US florist Donald Weder (1947 - ). Weder has filed 1940 patents worldwide, of which 975 are granted US patents. These include US4733521 Cover forming apparatus (1986). This patent describes a method of forming covers for flowerpots, and has 709 forward citations.

8) Fellow American John F. O'Connor (1864 - 19??)  is in eighth position. In contrast to floral suppliers, O'Connor specialised in railway components, and filed 949 US patents during his lifetime. These include US982086 Fastening mechanism for refrigerator car doors (1910), which looks remarkably similar to closing mechanism still used on truck doors today.

9) In ninth position is the Indian born/US resident Gurtej Sandhu (1960 - ), who has 1568 patents worldwide, including 953 granted US patents. Sandhu works in the areas of semi-conductors, including US patent 5240552 Chemical mechanical planarization (CMP) of a semiconductor wafer using acoustical waves for in-situ end point detection (1993), with 316 forward citations.

10)  Rounding out the top ten is the American  Melvin De Groote (1896 - 1963), who filed 925 US patents in the area of chemical demulsifers. These include US patent 1844883 Process for preventing accumulation of solid matter from oil wells (1927), which has just one  claim: 

A process for preventing the accumulation of solid matter in an oil well or pipe line, which consists in introducing a minute quantity of carbon bisulphide into a substantially clean well or oil line that is producing or which contains non-cutting oil for the purpose of preventing the solid material in the liquid flowing through the well casing or oil line from adhering to and collecting on same.

So in total, this list of the ten leading inventors comprises four Americans, two inventors each from Australia and Canada, one Japanese and one Indian. The technologies that these inventors have worked on have ranged from simple to very advanced. Some of the inventors are from an earlier era, and some are still actively inventing today. Some of the older patented technologies are still in use today, such as the incandescent light bulb and railway doors.

Even the ten placed person on this list has almost 1000 patents to their name, and the rest have patent filing counts in their thousands. By any measure these are remarkable achievements, and should be recognised as such.

I have previously discussed the limitations of using the count of forward citations and some other parameters to predict patent quality. But some new evidence is suggesting that these limitations are stronger than we first thought.

Many analysts in the patent landscaping field use forward citation count, either alone or in combination with othe fields such as the size of the patent family, number of claims etc, to predict patent quality. But can forward citation count be trusted?

The argument for forward citation count as an important is that a higher number of forward citations suggest that the patent being analysed is important, because

•  applicants (or even the applicant for the patent being analysed) are also filing patents in the same area,
• the applicant or examiner thought that the patent being reviewed is similar to these later patents.

In fact, these are arguments for citation analysis in general, which is at the heart of Network Patent Analysis. But at Ambercite, we believe that you need to consider all of the citations in the patent network to understand the importance of individual patents, and not just the direct citations. By only relying on forward citations (or even only forward and backward citations) you may be missing out on a lot of valuable information in the rest of the network.

The evidence for this is coming from the area of Alzheimer's treatments. We have previously reported on a white paper we have published on the patents in this area. 48,000 patents in the area of Alzheimer's treatment were whittled down to a final dataset of 2153 patents, which formed into 23 clusters. The very highest rated patent of all was US7189819, which protects the pahse III trial Alzheimer's treatment bapineuzumab, being marketed by Pfizer . In 14th position was US7195761, which protects the Eli Lilly phase III trial drug solanezumab. What we have not reported to date is that another promising alzheimers drug, which is now in phase II trials, ended up with its main patent in around the 200th position (but for commercial reasons we are not able to say any more than this about this drug or patent).

When we wrote this white paper, we decided that one of the internal tests of whether NPA worked was whether it was able to rank highly the drugs being tested in phase I, II or III Alzheimer's trials. So these results were pleasing. But would we have got the same results if we had simply counted forward citations? Table 1 below helps to answer this. And in this table, I have also added data for these three patents for backward citation count, total citation count, and the number of members in the INPADOC patent family, which is also used by some analysts as predictor of patent quality. I have also 'normalised' forward citation count by the how long the patent application has been published for, in order to determine the number of forward citations per year, which is thought to be an important parameter by some.

Table 1: NPA and other rankings for known Alzheimer treatment candidate drugs

As you can see, relying on forward or backward citation count, or the number of family members, would have given a very different ranking of these obviously important patents. The importance of these patents would have been missed. However the special algorihms used by NPA was able to rank these Alzheimer's drug very highly due to where they sat within the total patent network.

But what would you have ended up with if you had used these more conventional measures? I have reversed the question, and in Table 2 below asked which patents would have come up top in using the above measures:

Table 2: Highest ranked patents in Alzheimer's treatment data set based on non-NPA ranking measures

This of course is a very different set of patents to what NPA showed was important for Alzheimer's treatments.

Which raises the questions:

• if you are relying on metrics that rely on forward citation count to identify high value patents, what key patents are you (or your clients) missing?
• What impact could this missing information make on your business decisions?
• What business risks are you taking due to this missing information?

Contact us for further information about how we provide an unique NPA patent quality ranking for your area of business.

While many people use forward citation count as a simple predictor of patent quality, the number of backward citations is also mentioned by some people as a predictor of patent quality. A commonly expressed hypothesis is that while lot of forward citations suggests a good patent, lot of backward citation probably suggests a weak patent. Suggested reasons for this might include that that a higher backward citations is a predictor of a weaker patent due to a) the examiner not liking the patent and trying harder to invalidate it or b) there is simply more relevant prior art, and so the likelihood of patent grant is lower. I myself was more or less accepting of this hypothesis until I realised that some of the very highest ranked patents in some of our NPA studies had high backward citations counts - despite some other evidence suggesting that these patents were truly important.

So I looked up the published literature on the effect of backward citation count. While lots of people were supporting this hypothesis because it sounded about right (it does, doesn't it?), I was looking for supporting data and not just subjective opinions. It turned out that the best data I could find was by Lanjouw and Schankermann, 2002, "Research Productivity and Patent Quality: Measurement with Multiple indicators", which correlated their measure of patent quality with various patent quality measures.

Lanjouw and Schankermann looked at a number of different indicators, including forward citation count which, not surprisingly, correlated with their measurement of patent quality. But so did backward citation count - in a positive fashion (see table 2 of the publication). In fact, in two of the seven technology areas studied (Biotech and Other Health), the influence of the backward citation count was greater than the number of forward citations (in the five years after patent filing date). But even in the remaining five technology areas, the effect of the number of backward citations was still significant, and in some areas not that much less than the forward citation effect.

So the opening hypothesis did not stack up to testing - but why? Why did a high backward patent count correlate with higher quality patents in this published study?

One possible explanation for this is that granted patent with a larger prior art base may disclose a broader and ultimately more valuable invention; 'standing on the shoulders of giants', as somebody once put it. 

There is something else to consider. Besides promoting Ambercite I manage a portfolio of (unrelated to Ambercite) patents, and work closely with good patent attorneys to do so. Like many patent applicants we have a policy of seeking the broadest possible patent claims (without being silly about it), particularly for what we think at the best inventions. Not surprisingly, sometimes the examiners push back at these broad claims by throwing lots of prior art at us, leading to the inevitable process of negotiation with the examiners until we end up with patentable claims, but as broad as possible to support our objectives. In other words, when assessing why some patents attract lots of prior art patents, this may be because some patent applicants try harder for what they think are their more important patents.  

I am not sure if other applicants have the same approach to patent prosecution as us (I would be interested in feedback on this) - but if so, this may only further confuse the relationship between patent quality and backward citation count. 

What do other people think of the likely relationship between patent quality and backward citation count, and factors driving this?

Patent quality is often assessed from patent data by counting the number of forward citations a patent has, and perhaps augmenting this data with other available statistical data such as the number of family members a patent has, and whether the patent has been litigated or not. While such analysis is increasingly possible from available patent data, are these necessarily the right measures? Just because something can be measured, does it have predictive value?

The thinking behind counting the number of forward citations from a given patent is that each forward citation shows that somebody else (or the patentee itself) is filing a patent for a similar invention, suggesting that the patent being anlaysed may have some value. This does make sense, and for the purpose of today's blog can be regarded as a basic network measure, in reference to the fact that most patents are connected to other patents by citation linkages, leading to a network of patents.

In addition, we define a set of prosecution measures (because these measures all relate to decisions made by the patent owner or their attorney about how to prosecute their patents):

• the number of family members,
• if the patent or any family members have been abandoned,
• the patent coverage in key market,
• length of the patent claims
• number of patent claims
• and other patent family measures.

The thinking behind the value of prosecution measures is that that the owner should be best qualified to judge the quality of the patents. Decisions made by patent owners during patent prosecution can provides a useful signal about how they percieve the quality of the patent.

A third category of patent quality predictor is what can be defined as litigation measures. These refer to whether the patent has been litigated, either during its prosecution or after it has been granted. And we can also define a fourth category, internal measures. This can include categories such as the number of inventors, if inventors come from different nationalities, whether the applicant is a big or a small company, the number and breadth of International Patent Classification (IPC) patent codes, all of which are thought by some to be predictors of patent quality. Internal measures' can be distinguished from prosecution measures in that internal measures are largely set by the parameters of the invention and can't be controlled by the patent owner - a patent owner can't, for example, in ordinary circumstances control the number of international inventors for a given invention.

The strengths and weaknesses of these different measures can be contrasted below. 

Of the above measures, the basic network measure of forward citation count is probably the strongest measure, even allowing for the its reliance on patent age (some analysts adjust the forward citation according to the patent age to compensate for this. Other analysts adjust the forward citation count to allow for different practices in different patent offices).

But what if you could draw upon the whole of the patent network when assessing patents? Let us take the example of a patent with 15 forward citations and 10 backward citations. Altogether this is 25 pieces of data used to assess this quality (assuming that the backward citation count combined with the forward citation count to assess the quality of the patent). But each of these citations in turn is connected to other patents, leading to what can be a vast network, as shown in the figure below. Logically, as in all areas of business, some patents will be more valuable than others, and these patents should be identifiable as those with the strongest influence to other patents in the same area of the network. The figure below shows the patents forming into a cluster, with the cluster showing a discrete area of technology, in the case the details of the composition of the product being analysed.

Figure:  Example of a patent cluster, as created using Network Patent Analysis

Network Patent Analysis (NPA) can rank patents based on their influence in what can be vast networks. Up to one million citations linkages can be used to form these networks, and to rank patents, making the ranking a lot more statistically robust than a simple forward citation count (because one million data points are being analysed in this example as opposed to 25). Even recently filed patents can be recognised as high value patents, due to their network similarity to other highly ranked patents.

NPA even picks up on the prosecution measure of large patent families - this will tend to push up the individual rankings of the family members. And this is appropriate, as large patent famiies indicate a large investment in the technology, and inventions that can be hard to go around due to what can be a number of variations claimed in the different family members

In the terminology I have used above, NPA can be regarded as an full network measure, because it draws upon the whole of the network rather than just a small part of the network, as in a forward or backward citation count. And because of the power of collective wisdom, we believe that NPA can provide a quality of patent assessment superior to many other measures.

Accordingly, we can now add a new row to the above table:

Ready to move your patent analysis beyond counting forward citations? Please contact us, and we can discuss how we have helped clients large and small gain a competitive advantage by an advanced understanding of their patent landscape.