Patentsurf Discussion:

A comment from by Harrsha Reddy on the Intellectual Property Practitioners blog led me to explore the Patentsurf web tool. The comment: "Nice tool for invalidity search or to find the closest prior art, good part of this tool is that you need to put the patent number of your interest and this will give all closest patent with out writing the keywords" - Reddy

Patentsurf is linked to our members only PIUG Wiki list of Patent Database Resources: I had not tried this web search engine and decided to throw in a basic term that should do well as a natural language query: "milk processing" which covers everything from collecting milk to the pasteurization and bottling process. I did not add a particular patent classification or truncate terms.

Upon receiving the result set, I clicked on the term WIKI: It appears to have taken a Wikipedia record and extracted basic "verbs" from the various dairy definition. One could learn about the milking process by reading the accompanying WIKI article or choose to hold or search the verb terms on the resulting list. I looked at the Wiki definition of automatic milking and decided to select those terms related to "valve or actuator", collect a list of related terms and then manipulate them further.

I extracted from WIKI SURF list #1 the following terms. "alfa laval", Tesla, valve. milk. actuator, "hydraulic vacuum", "hydropneumatic suction", dairy, teat, regulator – and pasted these terms into a new PatentSurf search window. The resulting list of patents was not comprehensive, but the iterative process only retrieved patents related to the mammal milking or animal milking process. (Not vacuum cleaners or auto parts).

By repeating this process, one could very quickly collect list of terms, and more importantly identify terms or patents that prove useful or highly relevant.

In this case, I selected assignee names and dairy terms that were most closely aligned with valve actuators that controlled or released suction, one could obtain further lists of patents. This is where the process ended. I did not see a clear method of bucketing these patents after the browsing process for further manipulation.

While I see many promising features, additional testing is required to determine whether or not the PatentSurf product is a good tool for validity searching. I have not pursued this NLP browser further to determine it's effectiveness or comprehensiveness. It does, however, open up a world of vocabulary and patent numbers that can be further explored on traditional online databases. Since it is presently free, it can be used as a pre-processing tool to investigate terminology and develop vocabulary. Additionally, I would like to see a clear method of collecting or "bucketing" the patents to be reused without manually cutting and pasting patent numbers into a new search.

A second feature unrelated to natural language query searching that would make the tool more robust is an automatic "cite forward" or "cite backward" reiteration process. Once all patents are collected into a bucket or patent set, the cite forward process could be invoked. Alternatively if one is able to cite forward and backward, then narrow by IPC, USPC or EP patent classes then one should be able to develop a promising search technique using a web based patent search tool.

While we tend to use databases that are highly indexed and annotated, perhaps others will find these tools useful

I am interested in hearing from other PIUG members and knowing about their reactions or their experience with this product - especially those analysts who work with mechanical device patents – in any field of engineering or medicine.

Sherri Voebel
Patentskill LLC


  1. Hey Sherri,  thanks for posting this review.  I think you're right that the tool lacks features to support the full patent search workflow, such as "bucketing" the patents.    I had the opportunity to speak briefly with Mark of LittleEarth, Inc., the producer of the tool, and I think he did say that it's available for free as a showcase of their technology rather than a totally comprehensive and "finished" search product, so I think your concerns about coverage are quite valid. However I think it could of course be useful as a supplement to traditional searching in its current form.

    I did a quick write up of some of the features of patentsurf a little while ago for Intellogist, if you'd like to take a look it's available at   One of the things I did note at that time was that I preferred viewing the "Grid" results list.  I'm interested to hear other feedback from members as well.

  2. For PatentSurfP, As of 23 Jun 2010,

    1. Let NPD = Number of patent documents: 7730547 + 3590520 +1 = 4140028
    2. Let NR = Number of (patent document) relationships: 8,279,151,593,455
    3. Let UWC = No. of unique words in the corpus (where stopwords, post-process stemming & numbers are excluded)

    Let's try to determine the formula used in the computation of NR?

    • First attempt: NR = NPD * UWC (where only patents titles/abstracts/claims are used in UWC)
    • Second attempt: NR = NPD * UWC^2 (where only patents abstracts are used in UWC)

    Other attempts?

    1. Thanks. Rex.

      A lot of work available using various tools and internal/external ontologies such as licensed MESH, licensed METAWISE (Biowisdom),chemical nomenclature and smile strings (IBM). I'm especially interested in 1) advances in NLP projects and method of mining against patents in biomedical and bioengineering patents and literature 2) Tools for automation of the process of ontology curation and methods of ontology creation from various patent classifications. 3) The relationships between "unique set" or "small set" ontologies in patent landscaping.

      Sounds like this is your area. How did you determine Unique Word Count (UWC)? I think other PIUG members working in this space may also be interested.

      1. Hi Sherri,

        UWC is one of the starting point in Natural Language Processing (NLP)... PatentSurf NLP is most probably more complex than UWC, taking into consideration NL grammer or even semantics. May find my lectures relating to AI/NLP for UOL to be of interest:

        I found PatentSurf to be interesting, novel and with room for many useful improvements. There seems to be indications that PatentSurf prefers to keep the tech trade secret rather than patent it.

        PS: I still miss MAPit dearly - one of the best NLP patent search tool in the 90s and I'm still hoping that we have yet another patent search tool that is close to Map-it by Manning & Napier Information Services - in the 5 stages of grief-loss, am probably in the bargaining stage as far as MAPit is concerned.

        1. Thanks for the links to your lectures. I'm not certain what happened to Dr. Liddy and the TEXTWISE group out of Syracuse. For those of you who don't remember MAPIT, CINDOR & DR. LINK:

          1. The anonymous article you cited is very interesting - it sure does show how ephemeral NLP projects are.  SmartPatents, which was also discussed in the article, is still around, but after some financial hard times it exists as Thomson Reuters's Aureka. 

  3. Hello and thanks for posting information about this tool.  I actually used it for about an hour this morning.  As Kristin states above, I agree that at this point, it is a good "supplement" to more traditional searching.  You can easily search on a topic/patent # and review for relevancy, but that's about as far as you can use the tool since there is no place to save your results (at least none that I could find).  It is extremely easy to use and overall, I think it's a good tool for when management knocks on your door and asks you to do a "quick search" on something.