I'm updating this article as info comes in. The first update concerned a typo, very subtly [NOT ;) ] pointed out on Slashdot. In general, updates are integrated into the text here and significant ones are announced on my blog. So, you may want to check my blog if you're interested in the subject. Also, we have set up a mailing list for discussion of the Pinpoint patents. If you're interested in the list send an email to majordomo@transpose.com with the following in the body:

subscribe cfpatents

Update: I'm adding comments in an appendix at the bottom that relate to particular Pinpoint patents. The first concerns 6,029,195. 7/22/03: I've added a comment to the appendix on 5,754,939.

A lawsuit has been issued against Amazon involving their collaborative filtering (a.k.a. "recommendation engine") technology:

Pinpoint, a company that says it holds several e-commerce patents, sued Amazon.com on Thursday for allegedly violating its business-method patents. The suit, filed in the United States District Court in Chicago, also named six of Amazon's affiliated retail partners, including Borders Group, Target, Toys "R" Us and CDNow. The lawsuit claims that Amazon violated Pinpoint's personalization technology patents--a service popularized by Amazon that allows e-commerce retailers to recommend related products or newer items to shoppers based on their purchase history at the site. [News.com]

The present article will discuss those patents, including relevant prior art from the 1980's, some of which originated with the present writer.

Here is a page which lists Pinpoint's patents. I am not sure yet which ones the suit is about because none of the news reports I have seen mention them by name or number, but the candidates seem to be 5,754,938, 5,754,939, 5,835,087, and 6,029,195. The earliest one was filed on November 29, 1994 and the rest are continuations-in-part derived ultimately from that one. Further details can be obtained by by purchasing "file wrappers" of the prosecution history of the patents. I have placed an order for one, but it will take some time to arrive.

All in all, there are a couple hundred pages of specification, drawings, and claims here, and we don't yet know exactly which parts of this material are going to be used against Amazon. (This page will be updates as such details emerge.) Update: of course the claims are the key, but you often have to read the specification, and the prosecution history, to understand what the claims really mean. And we don't know which claims they are trying to apply.

But Amazon has been doing collaborative filtering for a number of years. In fact I had a few conversations with Jeff Bezos about it around 1995 because I thought Amazon might be interested in licensing my own CF technology, but not much later they made a deal with Net Perceptions to supply CF. If anyone has any dates on when specific recommendation-related features were added to the Amazon site, and/or the sources of those features, please let me know. I will be collecting information on the case and posting updates here. My impression is that Amazon is now building its own CF technology in-house. Update: One of the people on commenting on this piece on Slashdot is Patrick Tuft, who apparently is responsible for Amazon's "customers who shopped for this item also shopped for ..." technology. I wonder if his bit is specifically targeted in the lawsuit.

In order to be useful against Amazon in U.S. court of law, the patentees must have an earliest provable date of invention earlier than Amazon's, or earlier than that of whoever may have licensed the technology to Amazon.

The concept of "earliest provable date of invention" is a bit complicated. In the U.S., it isn't the date on which the application was filed. It goes back, under ideal circumstances, to the date the invention was conceived of. But several things have to line up properly. After concieving of the idea, the inventor must have been diligent in "reducing the invention to practice." This term refers either to "actual reduction to practice," which means building and testing the invention, or "constructive reduction to practice" which takes the form of simply filing a patent application.

It is possible to keep an invention secret for years after an actual reduction to practice and before filing a patent application, and still keep the date of conception as the date of the invention, as long as the patent is applied for within one year of the invention becoming public.

Given all the above, it may be impossible to determine from publicly available documents, including the file wrapper, what actual date of invention Pinpoint will claim. That may only by determined in court, when the parties involved in a lawsuit put all their cards on the table in order to make their cases.

However, there is some prior art that is very highly likely to come before any date of invention they will be able to claim. First of all, there is my own 212-ROMANCE project from the mid-1980's. That project used techniques that seem fundamentally not unlike some of the basic principles used in the Pinpoint patents. For instance consider claim 1 of the '195 patent, the last part of which describes:

scanning bulletin board postings to existing target object bulletin boards,
identifying groups of user identifications whose associated users have common interests,
matching users with other like inclined users to create a new target object bulletin board.

It is not a stretch to consider a list of recorded personals ads, accessed via telephone, to be a bulletin board with the individual ads playing the role of the postings. 212-ROMANCE monitored people's responses to the recorded ads as they listened to them on the phone. For example, they could hit particular touchtone keys to skip over the rest of an ad, or immediately respond to it, or store it away for later listening. Based on that kind of input, and considering to the logon ID's of the users as "user identifications," the 212-ROMANCE software performed the function of "identifying groups of user identifications whose associated users have common interests" where those interests involved preferences for individual personals ads. Then, the software performed the function of "matching users with other like inclined users to create a new target object bulletin board" where the new bulletin board was a customized list of personals ads played in the likely order of preference of the user listening to the ads.

Since I have been involved in the collaborative filtering field since that time, and have been aware that people might try to get broad patents that could potentially preclude me from doing my work, I have kept the relevant source code around ever since as proof of my date of invention. On a bookshelf in an adjoining room from where I am typing this, I have ancient paper printouts and 8-inch CP/M floppies in a manila folder containing that code. They date from 1986.Update: some folks on Slashdot pointed out that 8-inch disks have a short half-life. Rest assured I sent them to a conversion house some years ago and now have 3.5-inch floppies, DAT tapes, and probably have it somewhere on the laptop I'm typing from.

This prior art may have already proven useful. John Hey filed patent applications in 1987 and 1989 that basically did the same thing as described above but using different mathematical calculations for determining which users "have common interests." These patents, 4,870,579 and 4,996,642, were subsequently purchased by a company called LikeMinds, Inc. The CTO of LikeMinds posted a message on the Collaborative Filtering Mailing List in 1997 which mentioned LikeMinds' supposed "broad patent" in the field of collaborative filtering. Some perceived this as a veiled threat to eventually sue infringers. (It wouldn't do them much good to have purchased, rather than merely licensed, the patent if they had no ambition to ever use it offensively).

So I mentioned my prior art in a reply on the same list, and a fairly lengthy private email exchange ensued between me and the CTO in which he grilled me for details about 212-ROMANCE, and I had the distinct impression he was trying to find nontrival ways to differentiate the Hey patents from 212-ROMANCE. I further had the impression that he felt that he had failed in that effort. Eventually, the email exchange fizzled out, and I never again saw any public mention of those patents.

Update: Within hours of posting this piece I received an email from the CTO mentioned above, and based on the fact that he's a human being no different from me I feel a little bad about making the above info public... So I removed his name from the text so it doesn't become one of the top contexts in which Google searches find him. This isn't a personal issue, I'm just trying to make a point about the value of prior art.

With respect to the Pinpoint case, the Hey patents are yet another source of prior art; almost certainly the 1994 original patent application behind the stream of continuations was not based on prior art dating back to before 1987 or even 1989. Update: The Hey patent could be easier prior art to make use of in court, because no one can possibly argue that the filing date wasn't the one on the patent. In contrast, someone might argue that someone doctored the file creation dates on my disks, or argue that that feature wasn't in the live 212-ROMANCE at the time it was in the software (admittedly an improbable argument). The latter is relevant because it doesn't matter what software someone had in his computer at a given date if it wasn't publicly used. It's the public use that makes it prior art. (Actually, the CTO mentioned above pointed out that he found in his own research that there's yet another complication -- if it was publicly used, but the technology was a trade secret, that may not count -- the case law isn't fully resolved on the issue. 212-ROMANCE was not a trade secret, however.)

One fact that I find highly interesting is that the Pinpoint patents don't list the Hey patents as references. This leads me to the conjecture that the patent examiner didn't find them in his prior art search. If that is the case, the broad aspects of the Pinpoint claims may fall apart as soon as Hey's patents are mentioned in court.

Now, I am not saying that these items of prior art would completely invalidate the Pinpoint patents. But that's not necessary in any case. They have a right to patent whatever unique (and unobvious) ideas they have come up with. And they have plenty of technology described in those hundreds of pages; some of it is, no doubt, unique. But based on what I've seen of the patents so far, and based on my knowledge of CF technology in general, it does not seem likely in the light of the prior art that these patents would be defensibly broad enough to include Amazon's recommendations technology. I am basing that tentative judgement on the visible aspects of Amazon's technology, which could be accomplished any number of ways under the covers; there is no reason I know of at this point to assume that the technical details of Amazon's technology are the same. Rather, it would have to be the business method itself that is infringing -- the business method of making recommendations. But that was done nearly a decade before the Pinpoint patents were applied for.

I should mention a major caveat with respect to the above discussion. At present we do not know exactly what infringement Pinpoint is claiming against Amazon. So, we cannot rule out the possibility that Amazon will be found to have infringed. However, if the supposed infringement is based on a very broad interpretation of the invention described in the specification of the original 1994 patent application, then the prior art mentioned here may prove to be useful.

The fear, of course, is that Pinpoint will go on to sue other companies in the CF space. Luckily, Amazon has been sued first, and has deep enough pockets to defend itself quite effectively. Hopefully, the Amazon team will be very aggressive and seek to have the Pinpoint patents overturned as being overly broad.

However, there is a danger we should be aware of. Pinpoint may have the following strategy in mind: Amazon can afford to pay a relatively small license fee without breaking into a sweat, so maybe it will just do so rather than fight. The financial details, of course, would not be made public. Then, the precedent of having appeared to be able to force a settlement out of Amazon would give Pinpoint a psychological edge when they subsequently go after smaller practicioners in the CF field, who do not have Amazon's resources to defend themselves.

From a purely business sense, Amazon would have little reason not to go along with the strategy above. Alternatively it could go a step further and simply purchase Pinpoint and its patent portfolio. I don't need to descibe the danger of a company with Amazon's resources having such patents in its back pocket. Even if the prior art mentioned above would be a powerful weapon in combating lawsuits based on those patents, small companies may simply not have the financial resources to avoid simply being overwhelmed by Amazon's legal resources.

Update: someone on Slashdot mentioned Ringo. The problem with Ringo as prior art is that Upendra Shardanand's paper, like the earliest Pinpoint patent, is also from late 1994, and as discussed above, the filing date is not the same as the invention date. The invention date will be earlier, and that is what counts in U.S. patent law. So, at this point the question of whether Ringo can serve as prior art is unknowable.

Full disclosure: I have a couple of patents of my own in the CF space. However, these patents are for highly specific mathematical techniques and are of no interest to anyone using different mathematical techniques, of which many are freely available. Update: there have been some comments of Slashdot about these patents. One person read the Abstract and thought it looked like one of those ridiculously broad mindless patents the PTO keeps awarding. I admit that that the Abstract does look exactly like that -- I had other things I needed to do and let the lawyer put in some meaningless gobbledegook because it had no legal relevance, and I didn't care and had no reason to think the patents would ever be examined by anyone without a strong interest in the very specific underlying mathematics. These are EXTREMELY narrow patents when you look at the claims and specifications. They will not effect anybody's work unless they want to use my particular brand of arcane statistical mathematics, which took a very long time and a lot of effort for me to work out. BTW, if you have an interest in looking at them for some reason, only look at the later one, beginning at column 20, line 59. The rest was written while I was still trying to figure stuff out, but stuff starting in column 20 is pretty good IMHO. :) Also, suggestions were made that a) these patents of mine could serve as prior art to Pinpoint (unlikely due to being later than the root of Pinpoint's continuations) and b) that my 1980's prior art should invalidate my own patents (not true because my later patents are very narrow and do involve substantial mathematical improvements over my 1980's work).

Comments on specific Pinpoint patents

So far I've only had a chance to look at one of the patents in a non-trivial way, still withot reading the whole thing. I may never read the whole things, in fact, because I have no interest in these patents beyond figuring out what they are trying to protect.

5,754,939: SYSTEM FOR GENERATION OF USER PROFILES FOR A SYSTEM FOR CUSTOMIZED ELECTRONIC IDENTIFICATION OF DESIRABLE OBJECTS

As far as I can tell, Claim 1 of this patent, filed for in 1995 exactly mimics 212-ROMANCE's collaborative filtering feature from 9 years earlier, and also mimics an embodiment described Hey's 1989 patent (search for the words "remote user" to find the network-related embodiment). In other words, a lawyer should begin preparing a case to put this patent to rest. It's very hard to imagine how Pinpoin could defend it in light of this very solid prior art.

6,029,195: System for Customized Electronic Identification of Desirable Objects

The most striking thing about the claims of this patent is that they relate to "bulletin boards" or "target object bulletin boards." At first I thought that perhaps they had redefined the term "bulletin board" in the specification to mean something broader. But if you search through the spec in your browser, you won't find any use of the term that is nonstandard, with the sole exception of extending it to audio BBS such as in the text " In the case of a spoken rather than written message..."

The specification, when taken separately from the claims, is definitely broader than bulletin boards alone, so I suspect much of the same spec will be reused in others of this set of found patents, but with different claims, targeted at different subject areas.

But with regard to this particular patent, it looks like a recommendation system for music or movies would be uneffected.

Now of course Amazon has "bulletin boards" of reviews associated with every item. At this point, it looks to me that it may be possible that their customer-written review feature could infringe, depending on the techicalities of how Amazon does their processing internally.

This patent was filed for on December 5, 1997. If anyone knows how that date relates to the first appearance of customer-written reviews on Amazon, let me know. Although I must add that the 1997 date is not necessarily of great importance, because if the basic ideas are there in the 1994 application of which the present one is a continuation, they may be able to win a judgement against Amazon based on the earlier time period.

However, especially considering the fact that the patent explicitly extends the concept of bulletin boards to include audio versions, this is a case where 212-ROMANCE may provide enough prior art that the patents broadest claims could be invalidated.

		© Copyright 2006 Gary Robinson. Last update: 1/30/06; 2:48:13 PM.