Patents and Semantic Search: No Good, No Good

March 31, 2016

I have been working on a profile of Palantir (open source information only, however) for my forthcoming Dark Web Notebook. I bumbled into a video from an outfit called ClearstoneIP. I noted that ClearstoneIP’s video showed how one could select from a classification system. With every click,the result set changed. For some types of searching, a user may find the point-and-click approach helpful. However, there are other ways to root through what appears to be patent applications. There are the very expensive methods happily provided by Reed Elsevier and Thomson Reuters, two find outfits. And then there are less expensive methods like Alphabet Google’s odd ball patent search system or the quite functional FreePatentsOnline service. In between, you and I have many options.

None of them is a slam dunk. When I was working through the publicly accessible Palantir Technologies’ patents, I had to fall back on my very old-fashioned method. I tracked down a PDF, printed it out, and read it. Believe me, gentle reader, this is not the most fun I have ever had. In contrast to the early Google patents, Palantir’s documents lack the detailed “background of the invention” information which the salad days’ Googlers cheerfully presented. Palantir’s write ups are slogs. Perhaps the firm’s attorneys were born with dour brain circuitry.

I did a side jaunt and came across a white paper from ClearstoneIP called “Why Semantic Searching Fails for Freedom-to-Operate (FTO).”i The 12 page write up is from a company called ClearstoneIP, which is a patent analysis company. The firm’s 12 pager is about patent searching. The company, according to its Web site is a “paradigm shifter.” The company describes itself this way:

ClearstoneIP is a California-based company built to provide industry leaders and innovators with a truly revolutionary platform for conducting product clearance, freedom to operate, and patent infringement-based analyses. ClearstoneIP was founded by a team of forward-thinking patent attorneys and software developers who believe that barriers to innovation can be overcome with innovation itself.

The “freedom to operate” phrase is a bit of legal jargon which I don’t understand. I am, thank goodness, not an attorney.

The firm’s search method makes much of the ontology, taxonomy, classification approach to information access. Hence, the reason my exploration of Palantir’s dynamic ontology with objects tossed ClearstoneIP into one of my search result sets.

The white paper is interesting if one works around the legal mumbo jumbo. The company’s approach is remarkable and invokes some of my caution light words; for example:

  • “Not all patent searches are the same.”, page two
  • “This all leads to the question…”, page seven
  • “…there is never a single “right” way to do so.”, page eight
  • “And if an analyst were to try to capture all of the ways…”, page eight
  • “to capture all potentially relevant patents…”, page nine.

The absolutist approach to argument is fascinating.

Okay, what’s the ClearstoneIP search system doing? Well, it seems to me that it is taking a path to consider some of the subtlties in patent claims’ statements. The approach is very different from that taken by Brainware and its tri-gram technology. Now that Lexmark owns Brainware, the application of the Brainware system to patent searching has fallen off my radar. Brainware relied on patterns; ClearstoneIP uses the ontology-classification approach.

Both are useful in identifying patents related to a particular subject.

What is interesting in the write up is its approach to “semantics.” I highlighted in billable hour green:

Anticipating all the ways in which a product can be described is serious guesswork.

Yep, but isn’t that the role of a human with relevant training and expertise becomes important? The white paper takes the approach that semantic search fails for the ClearstoneIP method dubbed FTO or freedom to operate information access.

The white paper asserted:

Semantic

Semantic searching is the primary focus of this discussion, as it is the most evolved.

ClearstoneIP defines semantic search in this way:

Semantic patent searching generally refers to automatically enhancing a text -based query to better represent its underlying meaning, thereby better identifying conceptually related references.

I think the definition of semantic is designed to strike directly at the heart of the methods offered to lawyers with paying customers by Lexis-type and Westlaw-type systems. Lawyers to be usually have access to the commercial-type services when in law school. In the legal market, there are quite a few outfits trying to provide better, faster, and sometimes less expensive ways to make sense of the Miltonesque prose popular among the patent crowd.

The white paper, in a lawyerly way, the approach of semantic search systems. Note that the “narrowing” to the concerns of attorneys engaged in patent work is in the background even though the description seems to be painted in broad strokes:

This process generally includes: (1) supplementing terms of a text-based query with their synonyms; and (2) assessing the proximity of resulting patents to the determined underlying meaning of the text – based query. Semantic platforms are often touted as critical add-ons to natural language searching. They are said to account for discrepancies in word form and lexicography between the text of queries and patent disclosure.

The white paper offers this conclusion about semantic search:

it [semantic search] is surprisingly ineffective for FTO.

Seems reasonable, right? Semantic search assumes a “paradigm.” In my experience, taxonomies, classification schema, and ontologies perform the same intellectual trick. The idea is to put something into a cubby. Organizing information makes manifest what something is and where it fits in a mental construct.

But these semantic systems do a lousy job figuring out what’s in the Claims section of a patent. That’s a flaw which is a direct consequence of the lingo lawyers use to frame the claims themselves.

Search systems use many different methods to pigeonhole a statement. The “aboutness” of a statement or a claim is a sticky wicket. As I have written in many articles, books, and blog posts, finding on point information is very difficult. Progress has been made when one wants a pizza. Less progress has been made in finding the colleagues of the bad actors in Brussels.

Palantir requires that those adding content to the Gotham data management system add tags from a “dynamic ontology.” In addition to what the human has to do, the Gotham system generates additional metadata automatically. Other systems use mostly automatic systems which are dependent on a traditional controlled term list. Others just use algorithms to do the trick. The systems which are making friends with users strike a balance; that is, using human input directly or indirectly and some administrator only knowledgebases, dictionaries, synonym lists, etc.

ClearstoneIP keeps its eye on its FTO ball, which is understandable. The white paper asserts:

The point here is that semantic platforms can deliver effective results for patentability searches at a reasonable cost but, when it comes to FTO searching, the effectiveness of the platforms is limited even at great cost.

Okay, I understand. ClearstoneIP includes a diagram which drives home how its FTO approach soars over the competitors’ systems:

image

ClearstoneIP, © 2016

My reaction to the white paper is that for decades I have evaluated and used information access systems. None of the systems is without serious flaws. That includes the clever n gram-based systems, the smart systems from dozens of outfits, the constantly reinvented keyword centric systems from the Lexis-type and Westlaw-type vendor, even the simplistic methods offered by free online patent search systems like Pat2PDF.org.

What seems to be reality of the legal landscape is:

  1. Patent experts use a range of systems. With lots of budget, many fee and for fee systems will be used. The name of the game is meeting the client needs and obviously billing the client for time.
  2. No patent search system to which I have been exposed does an effective job of thinking like an very good patent attorney. I know that the notion of artificial intelligence is the hot trend, but the reality is that seemingly smart software usually cheats by formulating queries based on analysis of user behavior, facts like geographic location, and who pays to get their pizza joint “found.”
  3. A patent search system, in order to be useful for the type of work I do, has to index germane content generated in the course of the patent process. Comprehensiveness is simply not part of the patent search systems’ modus operandi. If there’s a B, where’s the A? If there is a germane letter about a patent, where the heck is it?

I am not on the “side” of the taxonomy-centric approach. I am not on the side of the crazy semantic methods. I am not on the side of the keyword approach when inventors use different names on different patents, Babak Parviz aliases included. I am not in favor of any one system.

How do I think patent search is evolving? ClearstoneIP has it sort of right. Attorneys have to tag what is needed. The hitch in the git along has been partially resolved by Palantir’’-type systems; that is, the ontology has to be dynamic and available to anyone authorized to use a collection in real time.

But for lawyers there is one added necessity which will not leave us any time soon. Lawyers bill; hence, whatever is output from an information access system has to be read, annotated, and considered by a semi-capable human.

What’s the future of patent search? My view is that there will be new systems. The one constant is that, by definition, a lawyer cannot trust the outputs. The way to deal with this is to pay a patent attorney to read patent documents.

In short, like the person looking for information in the scriptoria at the Alexandria Library, the task ends up as a manual one. Perhaps there will be a friendly Boston Dynamics librarian available to do the work some day. For now, search systems won’t do the job because attorneys cannot trust an algorithm when the likelihood of missing something exists.

Oh, I almost forget. Attorneys have to get paid via that billable time thing.

Stephen E Arnold, March 30, 2016

Comments

Comments are closed.

  • Archives

  • Recent Posts

  • Meta