Category: Amberblog

Cluster vs Semantic searching – ‘Head to head’

Here at Ambercite we have developed a means of searching for similar based on applying advanced algorithms to patent citation data, which we call Cluster Searching.

But what does everybody else do?

The two most other forms of patent searching are keyword searching and semantic search. An earlier and much read post discusses the underlying reasons for the natural limitations of keyword and semantic approaches. But enough of the theory – how do they compare in practice? Is it time for a ‘match up’? In Australia we might say a ‘Head to Head’ competition. 

In such a match, the choice of patents analysed can make a big difference. It is possible to cherry pick patents to prove or disprove any hypothesis, or make any approach look great. In an ideal world we might randomly pick say 100 patents, but you probably don’t have time to read the results from such a study, and I certainly don’t the time to report them.

An alternative approach I have used before is to pick ‘round number’ patents, where the subject matter is determined by the allocation of patent numbers by the USPTO. In this case, the last big round number granted by the USPTO was US9,000,000 for a Windscreen Washer Conditioner – which claims an aspect of using rainwater to refill your car windscreen washer. This is in the scheme of things a patent that should be easy to search using any technique, with relative simple and commonly used languate

So having set up a contest ‘who can find the most similar patents’, we need some contestants. It can be bad form in marketing to directly disparage your competitors (not that they really are competitors – we tend to think of them instead as complementary techniques as we will discuss below), so I will avoid naming them. Instead I will list them as follows

  1. Ambercite Cluster Searching
  2. A semantic patent search – that can search for similar patents based on patent numbers
  3. A very commonly free patent engine – that has a ‘similar documents’ function –
  4. A commonly used subscription search engine – that has a built in semantic search function

And the test? For each of these search engines, I will run a search using US9,000,000 or its abstract, and list in order the most similar patents listed.

The results will be provided below. But before we look at the results, we should take a closer look at U9,000,000

The abstract for this patent is:

A system and method of collecting and conditioning rainwater and other moisture, such as dew, from a windshield of a vehicle and utilizing the collected fluid to replenish the fluids in the windshield washer reservoir. A collection funnel is positioned on a vehicle in order to collect rainwater and other moisture. Rainwater and other fluids from the collection funnel are directed to a conditioning cartridge where the water is de-ionized and windshield washer fluid is added. The cartridges are designed to be single replaceable units. The mixed fluid from the mixing cartridge is directed to the pre-existing windshield washer reservoir.

So we could say that the key elements could include

  • Collection of rain water for a car windscreen washer
  • Conditioning of the fluid, and addition of washer fluid
  • A conditioning cartridge to do the above

So we want any patents found to disclose as many of these elements. Since we want to rank the different systems, we might assign a point for each element found.

 

Result set 1 – Ambercite Cluster Searching

The query looked as simple as this:

us9000000-query.jpg

 

The five highest ranked patents (in terms of similarity) were (after removal of US5669988, which was in the same family as US60224803)

Patent

Patent title

Collection of rainwater for car windscreen washing?

Chemical treatment of the water?

Cartridge system for chemicals?

Points

US3738575

Automatic windshield washer reservoir filling and mixing means

yes

yes

no

2

US6024803

Replenishment of vehicle windshield washer solvent using rainwater

yes

no

no

1

US6266842

Windshield cleaning device with liquid collection

yes

yes

no

2

US2770017

Windshield washer system

yes

no

no

1

US20120055857

Water Purifier

no

no

no

0

Total points

6

So as you can see, in this case, just for the top five patents, no novelty knockout was found, and one patent was listed that disclosed none of these elements. So as we can see patent searching is an imperfect art – but we all know that. And of course there was a reason why this patent was granted.

 

Result set 2 – First semantic search

This particular semantic search engine provided search results based on US900000. The top five ranked results are listed below

Patent

Patent title

Collection of rainwater for car windscreen washing?

Chemical treatment of the water?

Cartridge system for chemicals?

Points

US5261254

Self replenishing windshield washer system

no

yes

no

1

WO1996040544

Replenishment of vehicle windshield washer solvent using rainwater

yes

no

no

1

US2015027158

System for transporting condensed water from a vehicle’s air conditioner to its windshield washer reservoir

no

no

no

0

US6266842

Windshield cleaning device with liquid collection

yes

yes

no

2

WO2014198962

Reservoir for windshield washer fluid, support intended for a windshield washer fluid container and comprising said reservoir, and device comprising said support

no

no

no

0

Total points

4

As you can see, only one of the patents (US6266842) made it to two points – and this patent was also listed by Cluster Searching. This technique earned a total of four points. 

 

Result set 3 – Similar patents list in a well known free search engine

This particular search engine does not produce a ranked list of results – instead it produced a list of about 25 patents, sorted in date order

In the list below, we have ranked the patents by reverse filing date – with the latest filing date top of the list

 

Patent

Patent title

Collection of rainwater for car windscreen washing?

Chemical treatment of the water?

Cartridge system for chemicals?

Points

US6042023

Automatic deicing unit

no

no

no

0

US5103878

Apparatus and method for flushing and draining the coolant system of a vehicle

no

no

no

0

US4090668

Windshield washer and deicer

no

no

no

0

US5946767

Pipe cleaning vehicle

no

no

no

0

US5946763

Module for a motor vehicle

yes

yes

no

2

Total points

2

So not a great match here – only two point in total. But will the Cluster Searching ‘champ’ make it through to the final round?

 

Result set 4 – Similar patents list in a semantic search function in a well known subscription search engine

This particular search function requires a block of text rather than a patent number, so I based this system on the abstract from US900000,  as given above. At the top of the list of the results returned were US9,000,000 and a family member – but we have ignored these two patents in the list below.

Patent

Patent title

Collection of rainwater for car windscreen washing?

Chemical treatment of the water?

Cartridge system for chemicals?

Points

US7673814

Vehicle windshield cleaning system

no

no

no

0

US20160193986A

Vehicle surface wash apparatus with heated wash fluid

no

no

no

0

US20140352095

Plastic heat-conducting component for a system for supplying and/or distributing a window-washing liquid for a motor vehicle

no

no

no

0

US20100140378

Automotive Fluid Distribution System

no

no

no

0

EP2196367

Washing liquid tank, in particular for windows or projectors of a vehicle

no

no

no

0

Total points

0

This result honestly surprised me. It makes sense for the US9,000,000 and its family member to come at the top of the list – but the rest of the patents were a long way off.

 

What about a conventional patent search – what would this find?

We can make a reasonable guess of the answer to this one – by looking at the list of prior art citations listed by the examiner.

Patent

Patent title

Collection of rainwater for car windscreen washing?

Chemical treatment of the water?

Cartridge system for chemicals?

Points

US20040112411

Method and apparatus for cleaning a photoactive and/or hydrophilic surface

no

yes

no

1

US6089620

Rapid connection and disconnection device for electropumps and conduits for motor vehicle wind-shield washing systems

no

no

no

0

DE4436023

Method and device for the automatic filling up of the supply container of the washer system of motor vehicles with rainwater

yes

no

no

1

US5347661

Water conditioner dispensing apparatus

no

no

no

0

DE4101820

Windshield washer with rainwater reservoir – uses rainwater collector and reservoir as backup to main water reservoir

yes

no

no

0

US2703127

Fluid reservoir for vehicle windshield clearing systems

no

no

no

0

Total points

3

This is not a completely fair comparison as the examiner might found other relevant patents and not listed them, or tried to form an obviousness argument by combining separate documents. Regardless, it does make an interesting comparison to the other results.

 

The verdict

I will have to concede that this is not a knockout blow for Cluster Searching (for knockout blows see here and here) – but regardless it was a clear points decision, as shown in the table below

Technique

Total points

Cluster Searching

6

Semantic search engine

4

Similar to function in well known search engine

2

Similar patents list in well known subscription engine

0

Examiner listed patents (6 patents listed)

3

This is not a big surprise for us, as by applying sophisticated algorithms to both forward and backward citation links provided by experts such as examiner and applicants, we would expect Cluster Searching to be the better system.

The other thing to note is the range of results. Only one patent was duplicated in more than one list. We have send this before in an earlier comparison of cluster searching to other patent search techniques.

 


But this is an exclusive decision?

Not at all. Cluster Searching is certainly fast and friendly, and a good starting point for a patent search.  But if my client was facing a million dollar patent lawsuit, or making a million dollar investment, I would use every technique available to me – particularly because each method appears to give a different set of results.

And this is why we are happy to promote Cluster Searching as a second opinion engine. We want our clients to find the best possible patents, and are happy to provide part of the answer to this – and work in with the other parts. 

 

Want to try Cluster Searching for yourself?

Please contact us about a personal demonstration and free trial. We also offer free trial searches (we do the work for you, to prove that it works) for patent litigators. 
 

Read More

Patent invalidation searching in seconds – case study on patent asserted against Toyota, Volkswagen and other global car companies

How long does a patent invalidation search take?

If you were to use a traditional approach, it could be quite a while. You need to work out the keywords for the search, and make sure that you don’t miss any relevant keywords. You might also look up and apply a class code filter to the results, an additional step which could take additional time. 

You would then run this search through a conventional patent search database, and then comes the fun part – sorting the relevant from what could be hundreds of irrelevant patents. Realistically the ratio of relevant to total hits is going to be very low – and you will probably end up finding all of the patents already listed as patent citations.

Many many hours can lost in this way.

But there is another approach, and it can produce results in seconds – all the time while avoiding the risk of missing relevant patents due to inconsistent keywords. Best of all, this process builds on all the prior art searches already done in the same area of technology:  crowdsourcing on steroids, if you will.

This approaches is the Ambercite Cluster Searching patent search tool. To show how this works, we are going to use the example of US patent 8155342, filed by Blitzsafe in June 2006 for a Multimedia device integration system. This  essentially claims the concept of choosing songs streaming from your phone say, by using a controller in your car stereo via a bluetooth connection.

US8155342_image.jpg

Does your car do that? These days it is increasingly difficult to find a new car that doesn’t. Which is probably why the owner of this patent is busy suing the likes of Toyota, Volkswagen, Honda, Hyundai and Nissan for patent infringement. Who in turn are busy launching PTAB submission about why the patent is invalid. And no doubt running many patent invalidation searches.

So what would such a search look like in Cluster Searching? There are two main ways we can run an invalidation search for a known patent:

  • The really simple way – simply enter the patent number itself into Cluster Searching
  • The slightly less simple way – enter either the known prior art citations, or perhaps the most relevant of the prior art citations (as judged by the searcher or perhaps by the examiner).

In this case study, we are going to go for the really simple way. US8155342 has a filing date of June 2006 and a lot of listed prior art patents. Because there is potentially a lot of money at stake in this litigation, we might ask for the most similar 500 patents. The result search query looks like this:

US8155342_query.jpg


The results start off like this (double click on this image to get a full size version)

US8155342_top-few-results.jpg

 

This table shows a ranked list (by similarity) of the similar patents predicted by our algorithms. Perhaps not surprising, at the top of the list is a slightly earlier patent (US2005023434) filed by the same inventor. But as with any set of Cluster Searching results, the full list contains patents filed by a long list of applicants.

While this list provides the key details of the owner, name and filling date of the patents, pressing the blue circle in each row (in the live version) provides much more information about each patent listed, including the full specification in many cases.

 

Known vs Unknown citations

One of the key columns in the results in the second last column, headed “Citation’. This indicates whether a listed patent is either:

  • “Known”, i.e already known as a citation, discoverable using conventional patent search databases
  • “Unknown” (as citations). These are patents predicted to be similar but which have not yet been listed as citations.

Accordingly, ‘Unknown’ citations can be using in re-examination processes such as the US PTAB hearings, because they have not been cited by the patent examiners or the applicants.

But are they relevant? The column heading includes a simple filter, and so we can set this filter so that only unknown patents are shown. The top few results looks like this:

US8155342_top-few-unknown-results.jpg


Are the patents on this list relevant? They are all in the same general area – and some quite similar, for example (these patents are found a little further down the list)

  • US20050044574, filed by Harman International in June 2004 for a Audio/visual server. T discloses a music server that can communicate with a car stereo – and a portable digital music player is a music server. 
  • US20050286481, filed by Apple in June 2004 for a Method for data transmission utlizing a portable multimedia device, and which includes the ability to recieve track information and control the music

And are no doubt many more relevant patents – but these two examples should be enough to show that there is value in these results.

These results also confirm a point made earlier – searching for these sorts of patents can be hard because of the wide range of keywords used for similar concepts. The Harman patent application, for example, uses the terms ‘server’ and ‘computer’ to describe what Apple call a ‘portable multi-media device’ – but ultimately it is much the same thing.

 

Summary

This simple example shows how easy it is to search for previously uncited prior art that can be used to help invalidate a patent. These searches can be based on the number of the patent being invalidated, and rather than relying on your own assumptions about keywords, you can instead can draw upon the collective wisdom of all of the prior art reports in this area to do the work for you.

And does it work? In the words of our users:

We would like to mention how great we think the AmberScope search tool performs. We had the opportunity to test it now already for a case we are working on and it performs brilliantly.” – Dr. Patrick Daum, M.Sc., B.Sc, European Patent Attorney

I represent [X] in their case against [Y].  I was thoroughly impressed by your analysis of the [Y] patents”  – US patent attorney

I was looking for a prior art patent to invalidate a granted patent that was blocking a client’s product coming to market. I had tried a conventional patent search, and even a conventional citation search, but without luck. So I ran the search in AmberScope – and by exploring the visual links was able to find a prior art patent that essentially discloses the blocking patent. My client is now in a much better position to bring a product to market. Once again, AmberScope has found prior art missed by other searching techniques.” Gennaro Simonetta – Patent attorney, IP consultant – Griffith Hack

I recently used AmberScope for a novelty search and got a knock-out hit in less than half an hour. Even though, we generally use other platforms for searches I have to admit once you use Amberscope, you can’t go back to anything else. Keep up the good work!” –  A.M, Australian patent attorney

 

Want to try this for yourself?

Please contact us about a personal demonstration and free trial. We also offer free trial searches (we do the work for you, to prove that it works) for patent litigators. 

Read More

Yahoo is selling 2600 patents – what can we learn about them?

Yahoo is putting up a portfolio of 2659 of its patents up for sale, and is accepting bids at the moment. Price expectations are up to 1 billion dollars.

2659 is a lot of patents by any measure, and this would stretch the resources of any patent analyst. However here at Ambercite, we have developed some powerful tools such as Cluster Searching to help these analysts, and so we were curious to explore this portfolio ourselves. In particular, we were wondering:

  • Of these 2659 patents, which were the best – and what sort of subject matter did they claim?
  • Which companies owned the most similar and important patents?
  • If a prospective buyer missed out on this portfolio – what are the similar patents they might consider buying?

 

How we did it

We used our Cluster Searching patent searching tool, using the following process:

  1. We obtained the full list of patents from an assignment filed by Yahoo at the USPTO to move these patents to new company set up to own these patents, Excalibur IP LLC.
  2. We entered these patent numbers 200 at a time into Cluster Searching, and downloaded reports of their AmberScore values. AmberScore is our prediction of patent importance based on its citation density
  3. We enter a list of the 200 patents with the highest AmberScore values into Cluster Searching, and ran a query for the most 2000 similar patents to these 200 patents, that were filed after June 1996 – and we were excluded the Excalibur patents from these results.
  4. We identified the owners of these patents with the highest total licensing Potential
  5. We also reviewed the individual patents with the highest individual potential., that were not owned by major companies and so potentially available for acquisition by other companies who might want to get into the same space.

This took about 30 minutes or so

 

What are the best patents in the list of patents to be sold?

This was assessed by our metric AmberScore, which considers both forward and backward citations, and related citations. AmberScore has been normalised so that the average for granted US patents is 1.0 – so anything above 1 is above. In any reasonably sized portfolio we would expect a range of AmberScore values, and there is nothing wrong with this.

Regardless, the top scoring patents are likely to be much more valuable than lower scoring patents. The top ten list is shown below.  Perhaps not surprisingly, the list is dominated by patents linked to searching and internet advertising.

Number

Patent title

Year

AmberScore

US6269361

System and method for influencing a position on a search result list generated by a computer network search engine

1999

84.8

US5983227

Dynamic page generator

1997

35.8

US6151624

Navigating network resources based on metadata

1998

28.5

US6078866

Internet site searching and listing service based on monetary ranking of site listings

1998

26.7

US7269590

Method and system for customizing views of information associated with a social network user

2004

19.5

US6907566

Method and system for optimum placement of advertisements on a webpage

1999

18.5

US6826572

System and method allowing advertisers to manage search listings in a pay for placement search system using grouping

2001

17.8

US6360215

Method and apparatus for retrieving documents based on information other than document content

1998

16.9

US6112203

Method for ranking documents in a hyperlinked environment using connectivity and selective content analysis

1998

16.7

US8352331

Relationship discovery engine

2001

16.7

 

We can also consider the full range of AmberScore values, as shown in the table below. The above table showed that only 4 patents had AmberScore value greater than 20, and only a further 34 had AmberScore values greater than 10.

Amberscore value range

Number of patents in this group

>20

4

10 to 20

34

5 to 10

93

2 to 5

264

1 to 2

284

0.5 to 1

323

Less than 0.5

1657

 

As discussed above this sort of distribution is not surprising, and we have seen similar distributions for the full portfolio of patents filed by Apple, Microsoft and Google   – but does it does suggest two implications:

  • The vast majority of patents in this portfolio, like any large set of patents, may have limited values
  • Conversely, only a few patent will have high values – which places an responsibility on any likely purchaser to identify the high value patents and focus on these.
  • The history of purchases of large patent portfolios should confirm this thesis. The Nortel patent portfolio was famously bought for 4.5 billion in 2011, and then realized only about 1 billion in value. Around about this time Ambercite released an analysis of smartphone patents, and Nortel failed to make the list of the top ten patent owners in this space. So personally I was not surprised to see these patent realize less value than expected at the time.

 

Who owns the most similar patent portfolios?

This can be assessed by looking for similar and important patents, in comparison to the top 200 patents as assessed by AmberScore. This can be determined by our Licensing Potential metric, which predicts the potential for a licensing relationship for a later filed patents. A higher value suggests a greater potential, and if a patent owner has more than one patent in this list, these Licensing Potential values can be added together.

So who were the companies with the greatest Licensing Potential?

The top 10 companies are shown below. Perhaps not surprisingly Google heads this list. Yahoo itself was second on the list – note that these are patents not being sold as part of the Excalibur transaction.

Yahoo_graph_pic.jpg


Perhaps more surprising is the role of Millennial Media. Not all of us have heard of them, but Millennial Media describes itself as “the leading mobile ad marketplace, making mobile simple for the world’s top brands, app developers, and mobile web publishers.” Millenial Media was acquired by AOL in late 2015. 

The rest of the owners are well known in the field. But this does raise the question – of the patents in the list of similar patents with the highest Licensing Potential, which of these are owned by smaller companies who might be more open to selling them? This is covered in the next section.

 

If a prospective buyer missed out on this portfolio – what are the similar patents they might consider buying?

We answered this question by considering the similar patents with the highest Licensing Potential, and which were not owned by major IT companies. As you would expect the full list is a lot longer than this, but this gives a taste of what is possible.

Number

Owner

Patent title

Filing year

Licensing Potential score

US6285987

BEH INVESTMENTS

Internet advertising system

1997

185

US6324519

EXPANSE NETWORKS

Advertisement auction system

1999

169

US6029195

HERZ F S M

System for customized electronic identification of desirable objects

1997

159

US6421675

GLOBALBRAINNET

Search engine

1998

155

US5862325

INTERMIND CORP

Computer-based communication system and method using metadata defining a control structure

1996

13

US5855008

CYBERGOLD

Attention brokerage

1997

12

US6199099

ACCENTURE

System, method and article of manufacture for a mobile communication network utilizing a distributed communication network

1999

120

US6012053

LYCOS

Computer system with user-controlled relevance ranking of search results

1997

116

US5920859

IDD ENTERPRISES LP

Hypertext document retrieval system and method

1997

112

US6253188

PRODIGY SERVICES INC

Automated interactive classified ad system for the internet

1996

107

 

We expect that these patents could be of interest to companies looking to build up an ownership position in this space, but without the capital to buy the whole of the Excalibur portfolio. There are a wide range of owners on this full list, providing plenty of options for potential acquirers.

 

Like to know more?

Please contact us for further information, or for details on how you can conduct your own analysis on your own set of patents

 

 

Read More

Cluster Searching strikes again – a real world case study

Besides working with Ambercite, I do some IP management work with clients from time to time. As part of this, we were working with a client a couple of weeks ago who was worried about an Australian patent. This patent had been granted in Australia, and a very similar family member had been granted in the US – with the granted US patent having 30 prior art patent citations, and some non-patent literature. 

In Australia it can be very difficult to invalidate a patent for anything other than lack of novelty. So we needed a single document that disclosed all of the key features of the granted patent.

So what do we do? As you would expect we did the usual:

  • Looked at the cited prior art
  • Looked at a list of prior art documents found by the client
  • Looked at all of the documents cited against the US family member, and the file wrapper for the US family member

And none of these documents fully disclosed the patent we were trying to invalidate.

Naturally, we also of course involved Cluster Searching in the search. We did this in two ways:

  • A general search, in which we added all of the known prior art into a single Cluster Search, and reviewed the results
  • A narrow search, in which looked at the closest prior art documents we found in the any of the previous activity – and entered these close patents into Cluster Searching, one at a time, and had a look to see what we could find.

After this, we performed a list and preliminary ranking of the documents found, and asked a patent attorney familar with the case to provide an opinion on what we thought were the most relevant patent

He wrote back and said that one of the patents we found disclosed the main claims in the patent we were trying to invalidate.

 

a1sx2_Success_500_Success.jpg

 

Why was Cluster Searching successful when other approaches had failed?

The novelty destroying document was for a process that that a different commercial objective. However in achieving this different commercial objective, this earlier patent had disclosed every essential step of the patent we had been trying to invalidate.

We have seen this before, many manyu times. Conventional patent searching based on keywords is always an inexact process. All searchers are forced to make assumptions on terminology and keywords to make the search process manageable. Often this does work – I was involved in a patentability search this week based on a keyword search, and unfortunately for another client was able to find the exact process we were looking for.

But many other times keyword (and class code) searching by itself does not work – which is why we continue to recommend to all clients to use Cluster Searching to provide a second opinion to their conventional searching processes. Which is exactly what I did above – combined Cluster with conventional patent searching.

This can be confirmed by our many testimonials, including these two:

In 33% of our test cases with AmberScope we were able to find ‘X’ citations, i.e. patents highly relevant to novelty, that we did not find through any other search process we applied“. Patent office.

I was able to find a patent I had not seen before, even after 20 years in the technical area” – G.S, Patent Attorney

 

Do you want to find patents missed by others?

If so, please contact us and we will be happy to set up you with trial access. We also offer free sample seaches for patent litigators looking to invalidate problem patents on behalf of their clients or companies. 

Photo courtesy of www.gotcredit.com – used with permission.

Read More

日本語(Google翻訳)で読む