octimine technologies

Prior Art Search

Constantin BahlmannConstantin Bahlmann

Written by Dr. Marc Feron, former EPO patent examiner.

Always try to query with documents, or preferably agglomeration of documents, which globally look like what you are looking for.

Step 1: Document identification

Find the application or patent publication number to be invalidated, for instance: JP2009291310 or DE102006047370A1, get a facsimile/PDF copy of the original publication from Espacenet, the EPO’s patent information service, read it, look at the figures and get a feeling for the technical problem being solved, and the main features of the solution proposed by reading the abstract, the discussion of prior art in the first pages of the application and the description of the front page image. (Espacenet provides a free instant machine translation service into any language in case the document is in a language unfamiliar to you, e.g. Japanese).

Older Asian patents are sometimes only available in facsimile in Espacenet, but even these can be converted to machine-readable language by free OCR software.

Please also make a note of the patent classification codes (IPC, CPC, USPC, FI and FT) characterising both the technical problem and its solution, as provided by the bibliographical database of Espacenet. Espacenet often provides more detailed, updated and more reliable patent classification codes than the codes printed on the front page, especially for US patents. Do not restrict the search to these classes right away, though, because a surprisingly large number of patent documents are not classified where they should be! This is a substantial advantage of octimine over Boolean searches, especially over search methods relying on patent classification based restrictions!

Step 2: Octisearch!

Input that patent number in the Octisearch field, then click on search. Please don’t use filters at this stage, not even a date nor an IPC class: the goal of the present step is to find more very relevant documents about the same idea, even if they were published too late or misclassified.

  1. Octimine searches in English, but will try to find an English language family member if your patent number corresponds to a document in a foreign language. If it does find one (octimine knows that JP2009291310 is in the same family as US2009/0295200 and may assume that the latter is a usable English translation of the former), you’re all set to do a so-called „single patent number search.“ Wait a few seconds for octimine to display its results after working its semantic similarity magic, and go on to step 3 below.
  2. If the specified patent number has no English-language family member, like JP 2003- 299551A, octimine will pretend not to know its number and unfortunately won’t tell you how to handle the situation but don’t despair! You still can do a quick and precise octimine invalidation search on all 90 million patent documents of the Espacenet database maintainedby the European Patent Office, in any language (yes, even Chinese, Portuguese or Hindi language patent documents), with the following trick.

    First, get the original language text of its description and claims from Espacenet, the EPO’s patent information service. Second, use Espacenet‘s "Patent Translate", a free online automatic translation service for patents to get a free machine translation into English and then do an octimine text search by pasting the translation of the description into the Octisearch query space!

    Octimine will provide excellent results on such machine translations even when they look as horrible as machine translation usually does, so do not spend any time or money on a better translation. To get more focus on the invention, it may be more effective to feed octimine only with the description of a particular embodiment (usually the embodiment selected for the front page image.

    In case Espacenet is unavailable, you can also use Google Translate, or the free machine translation service of the relevant national patent information authority (J-Plat Pat for Japanese patents, which also provides info about the excellent Japanese FI and F-Term patent classification, for instance) to do an octisearch on its translated text, because octimine works very well even on very poor translations!

Step 3: Analyze the octimine results!

  1. Expand & explore the fields „Citations and references“ and „Similar documents“ - click "+"
  2. Look at the abstracts and figures in the results list, without paying attention to their dates for now:
    • Are they apparently dealing with the same problem and solution?
    • Are they more or less what you were looking for?
    • If relevant, are their IPC classes where you were expecting them? Check the IPC patent classification on the WIPO website to find out how to interpret the IPC codes, to decide whether those new codes may be worth exploring.
  3. Make a note of the documents that seem most interesting, especially those which were published before the priority date, as an invalidating document (proving obviousness or lack of novelty) may already be found among them! Jackpot already?
  4. Also make a note of later documents which are technically very similar to the patent to be invalidated, as they can provide richer vocabulary and more options for the semantic search: this is one of the reasons why octimine offers searching multiple patent numbers in a single query: if they are similar, this offers greater semantic certainty that a similar document will be found.
  5. If the results seem disappointing, go to Step 5 to learn how to use them anyway!

Step 4: (Re)Iterate your octisearch !

How? With multiple patent numbers or different text sections!

The previous step(s) probably yielded several very relevant patent publication numbers that are very similar to your original document: combine them with the publication number of your original document, separating them with commas. The idea here is to have a larger pool of vocabulary and semantic descriptions for that particular invention we’re looking for, to ensure that no stone is left unturned, and no document is missed. Then re-run octimine on that combination.

If you feel octimine has missed one important aspect or feature of the invention, it may be worthwhile to perform an octimine text search on the description fragment where that missing feature was best described, and possibly add equivalent passages from other documents describing that same aspect which was mostly missed to find relevant patent numbers with that feature.

By reiterating step 4 and trying different combinations of documents or texts, you should soon get to a point where you have identified all documents that are most relevant to seeding your search.

If there is a very specific word (never a general one!) or acronym that is very specific of the technical field of interest, like MIDI in electronic music or LTE in mobile telephony, then it may be wise to include it as limitation in one of the iterations, to try to uncover more relevant documents specific to that topic.

Step 5: Present your results

To present the results of a prior art search to your clients, you can decide on the 5 most relevant documents D1-D5, by order of relevance, having at least collectively all the features of the claims of the document to invalidate, and present their disclosure in tabular form, with one line for each claim feature, on column for each document, and in each intersection of a line with a column, a citation of where the feature of that line is disclosed in that particular document, or a O if it is not disclosed.

Claim 1 D1 D2 D3 D4 D5
Feature a       o
Feature b   o   o o
Feature c       o o
Feature d     o    


Claim 2 D1 D2 D3 D4 D5
Feature e o     o
Feature f   o   o

You should also indicate which documents are logically combinable with one another according to the problem-solution approach and by which reasoning they might be combinable.

Step 6: Diagnose and Fix your Octisearch!

Normally, an Octisearch provides consistently good results, especially in cases the “similarity average” curve on the “Top IPC codes” diagram exhibits just one or just a few strong peaks (warning: the most relevant IPC peak for the invention is not always the highest!) But a flatter similarity average, or one with too many peaks can be really bad news for the relevance of the results. If the results are disappointing, you can try one or more of the following:

  1. Try limiting the search with one of the IPC classes written down from Espacenet at Step 1 or one of the peaks of the “similarity average” curve: you should find at least a few halfway relevant documents that way, which will allow you to iterate as per Step 4. The classification code meanings can be retrieved at “classification search” in Espacenet or at the WIPO web site. It may be wise to restrict an octisearch only at the IPC class level (e.g. B62 for cycles) subclass level or main group level.
  2. Try to add to the octisearch text query one or more scientific papers or standards describing the invention or one of it more elusive aspects in very narrow technical terms.
  3. The “similarity average” on the diagram called “Top IPC codes” (the wavy line, not the bars) is actually your best indicator of the suitability of your query for an octimine semantic search. The better clue to the truly most relevant class is a peak of the similarity average (but that peak will not necessarily be the highest peak; in fact it is often a minor peak).

DO’s and DON’Ts with octimine

Do pay attention to the peaks of the similarity average (in the Top IPC Codes diagram), their shape, their number, and the meanings of the corresponding IPC codes (Do go to the WIPO web site to find out what they mean!). If the “similarity average” line of your octisearch has rather well defined peaks, and not too many of them, then your query is ideally suited for semantic search because the technical fields it describes have been very well recognized. Restricting your octimine search to those codes (for instance simply by clicking on them in the Top IPC Codes diagram) is likely to put a smile on your face! , because inventions are classified in patent classification systems such as the IPC in terms of the technical problem they solve and its solution as technical contribution to the prior art, so that in theory, if you find the right IPC code, you also find the place where all the most relevant prior art is located.



Do experiment with segmenting your octisearch text to search with increased focus on this or that feature. In particular, a restriction to the description of just one embodiment, or just one figure, may work wonders in a long text describing 76 very different embodiments.



Don’t restrict your initial searches (whether in time, nor in terms, nor in patent classification classes): you might miss an important document which may have been misfiled, or a later document which may be of no use as prior art, but ideal as a source of vocabulary for semantically finding (in a multiple patent number Octisearch) a very relevant prior document: any document that explains precisely an important feature or group of features of the invention is good to have for an octisearch.



Don’t be tempted to octisearch the terms of the claims: these terms are often much too general to be useful in a semantic search. Do be tempted to use very specific terms instead.



Don’t be tempted to use only three or four words in octisearch, unless they are really very, very specific, and you are prepared to limit the search by the correct IPC code.



Don’t pay attention to the rather misleading bar graph in the “top IPC codes” diagram, and do not restrict the search according to the tallest bars, because in most cases they are a statistical illusion.



Don’t spend money on expensive translations if you search foreign language texts: octimine works just as well with raw machine translations. See the detailed prior art search example of JP 2003- 299551A, a patent application unknown to Google because it has no non-japanese equivalent, which any non-japanese speaker can search and invalidate fairly quickly in octimine.



Don’t be tempted to use octimine as you would a classical boolean search engine: octimine thrives on very specific technical terms and is most likely to lamentably fail on catch-all terms. Patent searchers need to throw away their old boolean search reflexes, which are downright harmful and counter-productive in an octimine search and acquire a new way of thinking to make the most of octimine (how can I provide octimine with the fullest bags of words most representative of the invention): changing your boolean search habits (a search string for searching a few relatively broad search terms) to adapt to octimine (lots of very specific words) may actually be the only difficult adjustment necessary to very effectively search with octimine.


References: Traditional patent searching approaches

“Patent Searching Made Easy: How to do Patent Searches on the Internet and in the Library 6th Edition” by David Hitchcock, published by Nolo, 950 Parker Street, Berkeley, California 94710

Start your free patent search now.

+49 89 444 78663 0 | contact@octimine.com | © octimine technologies GmbH, 2017