AmberBlog

Recent and valuable discussions on Linked In regarding patent quality has reminded us that timing can be all important when assessing patent quality. The main elements of timing in relation to patent analysis are:

• Timing with respect to the expiry date of the patent. Patents have a maximum (except for some pharma patents) term of 20 years, but it is still possible to assert patents after they have expired, providing the assertion relates to commercial activity by the alleged infringer prior to the expiry date (Obviously an injunction becomes a moot point after expiry).
• Timing with respect to the commercial success of the patented technology. Some technologies take years to reach significant commercial volume. As just one example, Paice Corporation, the developer of the hybrid car patent that has been successfully asserted against Toyota and Ford, filed this patent in 1992, which happened to coincide with a long term low in oil prices. By coincidence, Toyota started developing their hybrid technology at about the same time, but it was 1997 before the first Japanese hybrid Toyota was sold, and 2001 before the Prius was available outside of Japan. In very recent times hybrid drive trains have been commercialised by a number of manufacturers. However Paicie may miss out on the bulk of this boom, with their key US5343970 patent due to expire in 2012 (although Paice have other patents they may assert).

The first of these elements should be straight forward for any patent analysis, and can be determined by comparing the expiry dates of the patents being analysed to the commercialisation timeframe of the patented technology.

The second factor can be analysed as well. Advanced patent analysis techniques such as NPA can be used to group patents into clusters of patents of related subject matter. Patents cost money to file, so a large amount of activity in a given area means that the area has  value. If lots of patents are being filed in a cluster at around the same time, this means that that area become commercially interesting for the patents to become commercially valuable. Since clusters in an NPA analysis can be dominated by highly ranked patents, we can start to undertand when the key technologies in each area have been developed, and this should correlate with a commercial or technology boom in this area.

The value of this type of analysis becomes very clear when we consider a comparison of the patent timelines for the top ranked patents in a few different technology areas we have looked, see the figure below. In this figure, we have identified when the highest rated patents in a given analysis were filed, and grouped these filing dates (after 1990) into 5 year groups. It is easy to seen when each of these area has attracted the most attention. 

This figure shows that:

• The three clusters from the smartphone report occupied three of out the top four positions. This is not surprising when we consider the recent innovation and legal developments in this area.
• Hybrid car patents were second on this list. Hybrid cars have become a lot more popular in recent years, and more importantly a range of car manufacturers are now introducing hybrid cars to help meet fuel efficiency targets set by governments.
• There were a lot of recent patents for Alzheimer's treatments. Although Alzheimer's has been known for over a century, there has been a large amount of medical research done in the last decade.
• The ICT project we looked at was for a very new area of technology, and hence there was nothing filed of consequence prior to 1995. However activity had appeared to have slowed down in this particular area of technology in the recent years. This shows that this particular technology had a relatively narrow peak of activity.
• The mining project had a number of older patents, but then an increasing number of newer patents. The reason for this is there was a new development in this particular tecnnology, and this has catalysed more recent activity.
• The heavy engineering project had a similar story to the mining project, but the new development peaked in the period between 1995 to 1999, when a lot of the key patents were filed.
• In contrast, the food technology project was in a very mature area, which was not surprising given the specific subject matter (which I can't discuss in detail, but I can say it was for a type of food that many of us eat almost every day).

Having said all of that, this comparison of technology timelines should be considered an example only of what is possible, as all of this data was collected in slightly different ways (apart from the three smartphone patent clusters). Nonetheless, this gives a good example of what might be discoverable from a NPA patent timeline analysis.

Patent thickets are a hot topic at the moment, but are often defined in a conceptual rather than objective sense. For example, Carl Shapiro of Berkely defined a patent thicket in 1991 as:

An overlapping set of patent rights requiring those seeking to commercialise new technology obtain licenses from multiple patentees.

These days, we might use a slightly amended definition:

An overlapping set of patent rights requiring those seeking to commercialise new technology obtain licenses from one or multiple patentees.

Many of us would probably claim to be able to recognise a patent thicket when we see one, but it is also useful to be able to objectively quantify the presence and strength of a patent thicket. Network Patent Analysis (NPA) is ideally suited to this, due to its algorithms that can precisely cluster similar patents together, independently of whether these patents share keywords or IPC codes. Other algorithms are able to define the relationship strength between each of the individual patents in the cluster, and the ranking of these patents within the cluster.

A typical cluster is illustrated below - this shows the leading patents and their owners in the mobile data access cluster of our smartphone report.

Ranking scores in turn can be used to assign 'patent points' to each patent in the cluster, so in a cluster of 5 patents, the top ranked patent has 5 points, and the last ranked patent 1 point. This in turn allows the comparison of individual patent owners within larger clusters in a more meaningful way than just counting patent portfolios.

So how can we apply these algorithms to patent thickets? A patent cluster can be regarded as a type of patent thicket. By looking at the average relationship strength within these clusters, we can estimate the degree of overlap, and therefore come up with a figure for 'Cluster thicket density'.

And by looking at the relative 'patent point' scores for different patent owners, we can start to understand the relative dominance of these clusters by individual patent owners. And since patent thickets are a legal concept, we might limit such analysis to the patents in the cluster that are less than 20 years old, since patents older than this will be expired (although a patent historian might include this in their analysis).

In the figure below, we have looked at cluster densities for the three largest clusters for a range of projects, being the published smartphone and Azheimer's reports, and three unpublished projects. These value range from 24.3 for the largest cluster (Peptides and antibodies) down to 1.6 for a mining project we have done. 

We were initially surprised to see that smartphone technologies did not have the highest cluster density. However what has made the smartphone patent wars so complex is that companies are litigating over a wide range of technologies that make up a smartphone. In comparison, a smaller range of technologies might be found within a drug, but the potential very high value of this drug might encouraging a plethora of patent filing around these key techologies.

We can also look further at the details of these clusters. In the table below, we have only looked at the largest cluster for each technology, but it is self-evident that the analysis could be extended into any of these clusters within the project.

It should be noted too that the size of the clusters, as we have defined them, depends heavily on one of our settings in our process (which defines how many patents we show in the final patent landscape map). If we use different setting we would end up with a different cluster size, and so this figure is given for completeness only. However the other parameters should be far less dependant on this process setting.

So as you can see, there are a range of results. Cluster size ranges from 65 to 941 patents (but note the dependence on NPA process settings). The proportion of 'patent points' owned by the leading patent owner in the top cluster ranges from 45% for an ICT project, to just 6.2% for a mechanical engineering project. There does not appear to be any direct relationship between cluster size, and cluster density. Similarily, there is no relationship between the proportion of the cluster owned by the leading patent owner, and other parameters. In short, every cluster (patent thicket) is different and needs be considered as such.

But in summary, yes it is possible to quantify a patent thicket (otherwise known as cluster), in any of technology, and to determine how strong the leading patent owner is within this cluster.

Cover of William Shakespeare

"What's in a name? That which we call a rose

 By any other name would smell as sweet"

Wrote William Shakespeare in one of his most well known play's "Romeo and Juliet" many centuries ago. But does the same apply to patent searching?

Names, and particularly technical words are particularly important when searching patents. For this reason a whole field had developed to improve searching using keywords, sometimes referred to as 'semantic searching". Keyword based searching does have a couple of important natural limitations to deal with though:

• Different patent applicants (or their attorneys) use different technical words for the same concept. For example, if we take the easy to understand area of patents for cardboard boxes, different applicants can use the technical words 'box', 'carton', 'container', 'package' and 'carrier' all to describe the same concept.
• A technical word or combination of technical words can include a wide variety of inventions. Returning to the world of cardboard boxes, there are thousands of different patents that cover different variations of cardboard boxes. However a patent attorney when drafting such patents is limited in the range of technical words they can use to distinguish one invention from another - for example 'top', base', 'side', 'flap' etc (and in fact every technical area has a limited set of technical words that are combined in different ways to describe different inventions). So a given patent search using a given combination of technical words can bring up a wide range of inventions.

Combining a keyword search with a International Patent Classification (IPC) or US patent classification codes can improve matters, but again limitations appear:

• A given IPC patent code can cover a range of a inventions within a given technical area.
• IPC codes can be imperfectly assigned, particularly when codes are translated from one classification system to another, for example from USPTO patent codes to IPC codes, or vice-versa.

But of course, we all know this. Anybody who has ever reviewed a list of patent search results would have waded through a whole set of mostly irrelevant patents before finding the patents they were looking for. Patent examiners can spend hours seaching for relevant patents, even using some very sophisticated patent searching tools.

This is an alternative though, and more and more people are using citation analysis to augment their keyword searches. During patent examination either patent applicants or examiner identiify similar patents, and the resulting patent citations are publicly available. The advantage of citation analysis is that links can be made between patents based on the inherant subject of the patents, i.e independently of the technical word used. A patent examiner instinctively knows that a 'box' is a 'carton' is a 'container', to give but one example. Similarly, patent citations are generally made to patents disclosing similar inventions, but not to every patent in the technical field. So inherently, using patent citations to identify similar inventions has the potential to bring a lot of precision to the patent searching task.   

However, even citation searching has its limitations:

• Some individual patents can have hundreds of patent citations, creating an additional large group of patents to wade through in order to find the patents you are looking for.
• Sometimes patent examiners, particularly US patent examiners, can identify an invention which is similar to the patent being examined, but which might be a missing a key feature (hence the patent being examined is novel over the prior art patent). The examinar then identifies a second patent disclosing the missing feature from what can be a different technical field, and then states words to the effect 'if you combine these two otherwise unrelated patents, the invention being claimed is not inventive". Besides being frustrating to patent applicants, this type of prior art analysis also creates what can be a citation link to a very different invention, in what can be a different technical field.    

Network Patent Analysis (NPA) uses a different approach to analysing citation connection. It looks for supporting evidence (similar citation links) to weight the importance of a given citation link. Weighting of citation links is then used to help identify clusters of patents of very similar subject matter, with the relative importance of these patents emphasised. Weakly connected patents (due to being in another subject matter area) are ignored. Results are presented in a visually insightful manner, as shown in the figure below. The visual cluster can even be used to suggest similar patents that are not directly citationally connected to the patent in question.

Figure: Example of cluster analysis from an upcoming NPA white paper on Alzheimer's patents. Another examples of cluster analyis, applied to smartphone patents, is found here.

This relieves the patent analyst, attorney or lawyer of the burden of having to wade through what can be hundreds or thousands of unrelated or only loosely patents. Instead they can concentrate on the most similar and most influential patents in the field, saving time and money, and reducing the risk, of missing key patents.  Or in other words, NPA brings a new level of both:

• precision (focusing in particular inventions) and
• generality (by bringing in inventions that can have different keywords or patent codes, and allowing objective analysis of tens of thousands of patents).

into patent searching.

Or in other words, we believe that the words of Shakespeare can be applied to patents - a patent by another (technical) name can smell as sweet... 

It is well known that there has been an increase in patent filings in many countries. But what can these patent filings tell us about the rate of technology change, as opposed to just the amount of patent activity?

One method of answering this is to look at the age of the backward citations. Patent examiners or applicants refer to patents in their search reports that they think are relevant to the patent being examined, and so we should be able to measure the rate of change based on how far back their citations go. We can measure this for either a single or group of patents in two different ways:

• By looking at the average age (difference in years between the patent and backward citations) of the backward citations
• By looking at the oldest backward citations, which tell us just how long patents thought to be relevent to the patent being examined have been around for.

From a statistical viewpoint, the second measure is flawed as just a single, potentially careless reference to a very old patent can distort the data. In fact we are mainly interested in the age of the majority of the citations - we may not need to know about the age of the very oldest citations to gain a reasonable idea of the rate of change. In practice, the age of the youngest 90% of the patent citations should give a reasonable idea of how long a particular technology has been around for (as only 10% of backward citations are older than this). We can call this measure 'Patent Turnover", being the maximum age of the youngest 90% of backwards citations.

We would also want to be a little selective about which backward citations we use to calculate these figures. As an example, imagine the development of a new drug diagnosis test with application for a wide variety of therapies, with patents filed for this new diagnosis. This development in turn may catalyse a number of new patents, including potentially in the area of Alzheimer's treatments (as just one example). The patent examiner or applicant reviewing this new Alzheimer's patent (that builds on this drug diagonsis invention) may quite rightly cite this drug diagnosis paten in their search report. But is the original drug diagnosis patent an Alzheimer's patent? We would argue not - it falls within another technology field, namely that of drug diagnosis, and so should be distinguished from any technology analysis of Alzheimer's patents.

Hence any analysis of the ages of the backward citations in any technology field should only consider the age of cited patents within the field itself - otherwise we are measuring the rate of change of other fields as well. Luckily, NPA lends itself very well to this type of analysis, as it is can be used to cluster patents with other patents within the same technology field, but to ignore patents cited outside of the nominated field.

Anyway, enough background -  so what is the rate of change for a range of different technical fields? We have taken this data in Figure 1 from a variety of projects. The 'n' values for each type of technology refer to the number of relevant backward citations (to other patents in the same field) and not to the number of patents. These values are affected by the way each project was set up, and should not be interpreted in any other way than to confirm that there is sufficient data for the calculated values to statistically robust. 

 Figure 1 tells us:

• The average age of backward citations ranges from about 6.5 years (an ICT project and for Alzheimer's drugs) up to over 15 years for a mining industry project
• The "Patent Turnover" (age of youngest 90% of the patents) is around twice the average age of the citations.
• Within the area of smartphones, patent associated with mobile data access (to emails etc) had a lower citation age than patents associated with touchscreens. In simple terms, touchscreens appear to have been around for longer than mobile data access. 

The varying Patent Turnover figures would suggest that companies may need to innovate at a faster rate within the ICT and pharmaceutical industries when compared to mechanical engineering and mining. Accordingly, a technology more than 10 years in the ICT industry may be obsolete, but technologies in the mechanical engineering and mining industries may be commercially relevant for much longer. These trends may not surprise many, but it is pleasing to see the patent data confirming what we probably all suspected.

We can also compare backward citation age gaps for different patent owners for a given technology, in this case for four leading owners in the mobile data access patent cluster in our previously published smartphone white paper, see Figure 2 (the number of backward citations used to calculate the Motorola values are less than ideal, and this may have affected this result). 

So what do these owner results mean? There are different potential interpretations. We do know that Palm (remember the Palm Pilot?) and RIM (makers of Blackberry) were among the first companies to offer portable devices that could be used to read your email. It is possible that the examiners for the earlier patents filed by these companies could not find too many (then) recently filed prior art patents, and so had to go further back into the patent database to find relevent prior art. In contrast, later entrants to this market had more (then) recently filed patents to choose from. In this case, we could say that the early patents in this field were 'game changing' patents, Figure 3.

There is potentiallly a lot more that can be gleaned from this type of analysis, and we will continue to develop analysis tools to better interpret the wealth of data within the world's patent databases.  

.

As discussed here, NPA analyses vast networks of patents to rank and group patents. While we have previously shown patent rankings in a simple fashion (1st, 2nd, 3rd, 4th, etc), we can also calculate patent rankings in terms of their relative strength. Just like with tennis pro earnings and patent portfolios, we can 'normalise' these values so that the highest ranked patent has a value of 100. We have done this for the three public domain NPA patent studies we have published, namely for hybrid cars, smartphones and Alzheimer's treatments (will be published soon).

Figure 1 shows that the drop-off in relative values was slower for patents than for tennis pros (and therefore patent portfolios). The three patent distributions appeared to have similar curves, suggesting a similar distribution of relative values. 

Practically, we can interpret Figure 1 (and its underlying data) to show that:

• The 10th ranked patent is worth between around 1/3 to 1/2 of the top ranked patent (35% to 52% in our data)
• The 50th ranked patent is worth around 1/6th of the top ranked patent (13 to 18%)
• The 100th ranked patent is worth around 1/12th of the top ranked patent (8 to 10%)
• The 200th ranked patent is worth around 1/20th (4% to 6%)

This drop-off in relative value was actually slower than we expected, which shows the value of analysing patent data rather than speculating.

We can also use this data to show the rise in cumulative patent value as we move from the highest ranked patent. Figure 2. What is astounding however is that regardless of the size of the datasets, 58,000 down to 7,000, the curves are very similar.

There are slight differences between the curves, but I don't think that three curves are enough to draw any firm conclusions about these differences. Again, we can use this data to draw some practical conclusions regarding patent value:

• The top 100 patents represent between 43 to 53% of the total predicted value of the patent dataset
• The 200th patent takes you to around 60% of the highest value
• About 80% of patent value in a given dataset, regardless of its size (!), appears to come from the top 500 patents
• About 90% of patent value is given by the top 900 patents
• At about 1350 patents, we have covered around 95% of the patent value, with the curve flattening off significantly
• 2000 patents should pick up around 98% of the value in the dataset 

In other words, if you want to be on top of the critical patents in a given field, you need to analyse a targeted dataset of 500 to 2000 patents, depending on the level of understanding you require. While this may seem like a lot of work, the ability to group, rank and visualize patent connections can make this task a lot more manageable, particularly if combined with more traditional patent analysis methods. 

Mike Lloyd and Doris Spielthenner