The team is joined by: GuestKats Mirko Brüß, Rosie Burbidge, Nedim Malovic, Frantzeska Papadopoulou, Mathilde Pavis, and Eibhlin Vardy; by InternKats Ieva Giedrimaite, Rose Hughes, and Cecilia Sbrolli; and by Verónica Rodríguez Arguijo (TechieKat), Hayleigh Bosher (Book Review Editor), and Tian Lu (Asia Correspondent).

Monday, 13 November 2017

Looking for answers on ANSERA

Lemme ask some questions...
ANSERA (ANother Search ERA) is a semi-automatic search engine for patent information used internally by the European Patent Office in its search efforts. The purpose of (semi)automatic search is to automate as far as possible the search process and eliminate all non-value added steps for examiners at the beginning of the search workflow. The goal is to deliver highly relevant prior art documents before the examiner starts working on a patent application.

The publicly available information on ANSERA is surprisingly sparse. We know from an IT Roadmap Update and Plans submitted by the President of the EPO to the Administrative Council on 6 June 2014 that it was completed in 2014. From the same document, we learn that ANSERA enables the following types of searches:
  • Figure searches using reference sign relative positions in a drawing,
  • searches using concepts, where results are ranked according to the number of common concepts (or features) as well as according to their occurrences in the full text; and
  • "find similar" searches, based on a selected text of a patent application. It further allows result filtering, e.g. by limiting on classes.
We know a little bit about the technology underlying ANSERA (or its remake?) from a June 2017 interview with Bastiaan, a Senior Software Engineer with Competa who is working on-site at the EPO. Bastiaan is working on a remake of ANSERA, and he has the following to say on its technology:
At the back Ansera is driven by an Elastic Search implementation and a persistency database. It also communicates with some other services for acquiring full-texts and performing translations on the fly. The only way your patent application will be granted is if no prior art exists for claims stated in the patent application.  Ansera enables the user to create concepts using a custom query language. A concept can range from anything like a simple term to a complex equation.
After the user defines the concepts, these concepts are used in various search types. One of the search types is called Figure Search. With this search one can visually place concepts on a figure and the figures collection will be queried for figures containing these concepts. The relative distance between these concepts is important for yielding better search results. The way it’s set up is very sophisticated.
Merpel being a curious cat, on 22 October 2017 she posed a number of questions to the EPO through the official query form on the EPO’s website:
  1. What are the key capabilities of ANSERA? How do they differ from other patent search software? 
  2. What technologies drive ANSERA? 
  3. Are there additional features planned for ANSERA?  
  4. What are the next steps in (semi) automatic search at the EPO? 
  5. Are there any plans to make ANSERA available to the public/users? If so, when and under what conditions? (free, pay, login) 
  6. How satisfied is the EPO with the results generated by ANSERA? Are there any areas where improvement is required, and if so, where?
  7. Is it correct that new examiners are only trained in search using ANSERA?
The enquiry received the number 542532. Merpel was promised her questions would be forwarded to the “department in charge”. Not having heard from the “department in charge” (or anybody else at the EPO), Merpel sent a follow-up enquiry on 1 November 2017 (Enquiry no. 544546). Merpel was informed that “[a] reminder has now been sent to the department dealing with your enquiry”.

It now being the 13 November 2017 - more than three weeks after the initial enquiry - and not having heard from the EPO, Merpel puts the above questions to the readership of IPKat. She is particularly interested in hearing about your experiences using ANSERA if you are an examiner with the EPO.

Please do comment, but keep the Comment Moderation Policy in mind – use a pseudonym, and be civil. If you prefer to send an email, Merpel can be reached at


Anonymous said...

Shocking that the EPO don't reply instantly. Luckily the Kats have access to a broad community for obtaining information for their work.

Kant said...

"The purpose of (semi)automatic search is to automate as far as possible the search process and eliminate all non-value added steps for examiners at the beginning of the search workflow."

Or rather, the purpose of semi-automatic search is to de-skill the task of patent searching so as to enable the highly skilled and experienced examiners to be replaced by unskilled workers on short term contracts.

Anonymous said...

Sounds like it should be re-named "NO ANSERA"....

Anonymous said...


Some may view your post as cynical, but when you view ANY business operation - and most all business operations are based on that very same concept - why is it that you think that examiners should be excused from the (seemingly) natural market forces that affect everyone else?

Proof of the pudding said...

@Anon 13:47

Exactly which "market forces" do you believe are at play with regard to the recruitment of EPO examiners?

The EPO is an international organisation that has a monopoly on dishing out (EPC-wide) monopolies. This means that the EPO does not have any relevant "competition". It is therefore completely inappropriate to apply "free market" concepts to such an organisation.

Of course, should its "users" so demand, then it might be appropriate for the EPO to look for ways of reducing costs (and hence reducing fees). But even then, one has to balance any drive to reduce costs against other demands that the "users" of the system may have.

At this point, it is important to remember that the "users" of the system include 3rd parties whose freedom to operate will be curtailed by the monopolies that the EPO grants.

It is therefore inconceivable that a majority of the EPO's "users" would ever be in favour of any cost-cutting that compromised the ability of the EPO to conduct high quality examination. From this perspective, it hard to see what justification there could possibly be for adopting recruitment practices that are aimed at "de-skilling" the EPO's entire examiner base.

Anonymous said...


You misunderstand the aim of my earlier comment.

It is not that the EPO "has competition" - or not.
It is nothing whatsoever to do with the "users" of the system (therein lies nothing but dust-kicking).

It is that the EPO may apply what is no more than standard business protocols to its own work.

Or do you think that such is somehow off limits? Under what basis would this power to set as "off limits" come from?

Proof of the pudding said...


I think that you are rather missing the point.

There is no "standard business protocol" when it comes to a patent office. This is because there is no "business" to speak of.

Patent offices exist for the sole purpose of being the first (and most important) gatekeeper to a state-sanctioned monopoly. A patent office therefore only serves its purpose if it applies adequately (but not overly) stringent criteria to the grant of a monopoly.

So yes, it is "off limits" to consider adopting practices that are liable to render the patent office not fit for purpose.

It is interesting to note that the European Medicines Agency values its "skilled" staff so highly that it believes that the new location for the Agency should only be selected from the cities that staff surveys show would provide a high retention rate.

The EMA is much like the EPO in that it examines applications to check that they meet suitable standards. So if retaining skilled staff is such a high priority for the EMA, why should it be any different for the EPO? Improvements in efficiency are one thing, but my view is that any "improvements" that would render the EPO incapable of performing its function should never even be contemplated.

Kuifje said...


You're a bit harsh. Of course it is not off-limits for EPO to consider changing their practices.

Or would you want them to still use index cards and miles of bound volumes of old applications?

There's nothing wrong with improving efficiency. They can consider, and test, and evualuate all they want and only keep the good stuff.

I do agree that this should not reduce quality. And with the current EPO management that is indeed a worry. Quality has dropped drastically the last 3 or 4 years.

EPC article 1: A system of law, common to the Contracting States for the grant of patents for invention is established by this Convention.

For the grant of patents, not for their refusal! For the grant of patents, not neccessarily for the grant of high quality patents!

Quality should be assured by a constantly vigilant AC, alas, that is lacking.

Cynic said...

Back at Merpel’s questions...
1. Don’t know. And I’ve done the training...

Seriously, it is another technique and maybe it does work but it works in parallel with all my experience and doesn’t easily combine with it. It’s a bit like speaking Spanish for years and then one day being told that you would be better in Flemish. Why? Nobody really explains and you have no time to learn it. So you just ignore it. If you only learn Flemish and never learn Spanish (and they stop any Spanish classes), stats will always show Flemish is more popular except with the old fossils.

In practice there is no comparison of quality differences being made, as far as I know.

AICaramba said...

"AI renders knowledge workers redundant."

We'll see this headline a few more times over the coming years. But it's not always realistic about the outcomes. Patent search is very amenable to moving many tasks to AI implementations. one well, it should free examiners up to spend more time on examinations, if costs are to be kept at the current level, or to reduce costs for users, if examiner workload is to be kept constant. Both seem like positive outcomes.

Anonymous said...

Curious thst IPkat is interestedin EPO again. Merpel should however know EPO examiners are not allowed to comment on the Office.

Glad to be out of the madhouse said...

In my humble experience, little separates AI from NI (Natural Idiocy). It's true that there have been impressive improvements in some fields, like image recognition and automated translation, but even in those fields those improvements have only been achieved by force-feeding computers with phenomenal amounts of information, and I doubt that such an investment is yet economical or even possible in such a specialised field as patent searching. In any case, it still remains very much a matter of GIGO (garbage in - garbage out) and, AFAIK, in low tech fields like mechanics, the EPO databases are still quite corrupted by crappy OCR scans of old prior art (which can still be very much relevant in those fields: I've known cases where the killer prior art was over a century old).

I'm also slightly perplexed by the assumption that AI is put to better use in search than in examination: maybe AI would be less subject to hindsight bias than a human examiner when applying the Problem-Solution Approach, or determining whether something could be "directly and unambiguously derived" from a disclosure, don't you think?

Anyway, before starting to wonder whether android examiners would dream of electric mousetraps, perhaps we should employ a dose of realism and ask ourselves whether EPO management isn't falling into an old pattern of seeking salvation in fashionable tech they fail to understand, underestimating the challenges to implement it, and spending valuable resources into external consultancies in exchange of underwhelming results...

Proof of the pudding said...


It seems that you and I are in agreement. If you re-read my comment, it is clear that I only placed "off limits" those practices that are liable to render the patent office not fit for purpose.

I am in no way suggesting that investing in new technologies or adopting new ways of working is a bad thing. Instead, what I am saying is that all modernisation / efficiency drives must not render the EPO no longer fit for purpose.

On this point, I am afraid that I share Kant's concerns that "the purpose of semi-automatic search is to de-skill the task of patent searching so as to enable the highly skilled and experienced examiners to be replaced by unskilled workers on short term contracts". I would be delighted to be proved wrong. However, even if there is no diabolical plot behind the reforms are being, I suspect that Glad to be out of the madhouse is correct to wonder whether the EPO management is falling into the trap of pressing ahead with unproven technology without fully appreciating the possible implications.

Anonymous said...

I believe that many of you may be right and some not, but putting that aside, do you really think that the reduction of cost will affect applicants? Really? I believe that it won't happen at all. EPO management might lower some fees, while raising some others...
This ANSERA search seems like what third party search provides are already offering (e.g. Questel). If that's the case, I do believe that the quality of the Examination will go down dramatically, as it is currently happening. I believe that if you do not have a good prior art search, citing related prior art, you cannot expect to have a good Examination report at all!
So, while we may be getting ANSERA searches, we may also be getting more grants, and as a result, more oppositions and appeals. Thus, at the end this means more fees for the EPO and for the AC.

AnsQuera said...

My comment y'day was too long, and I did not have the time to cut it sensibly down to the 4000 character limit.
So new try for the most relevant parts:

The old EPOQUE databases have long since reached their design limits.
They have improved on that, but some limits will remain.
ANSERA is based on modern database structure, which is much more flexible.
ANSERA was and is therefore ncessary.

Re. search approach: EPOQUE with INTERNAL and XFULL is very classification oriented.
In ANSERA, classification symbol limitations often do not work, or are even working faultily.

Re. question 7: this is a perception bias. Newcomers still do get training in EPOQUE. But....
- tools training for newcomers has reduced from 7 weeks to 2 weeks, and encompasses more tools than before.
The correct teaching is left to the tutor. Who has to meet high preocution values, and has NOT received training how to train.
so newcomers get to lock themselves up and have to find something relevant.
Understanding the classification systematic isn't easy, and prone to errors, leading to searches in irrelevant fields. If done correctly, your documents would all be highly relevant.
With ANSERA you get a google effect: many seemingly relevant documents, but very hard to filter down to the most relevant ones. Just like google gives some 20 relevant result pages. So you'd have to scroll through MANY more documents to see all relevant ones. Which ANSERA is rather unsuitable for, and transfering large amounts to the better document tool is still impractical.
So people look at the top 70 documents or so, according to some random evaluation algorithm.

And because finding documents loosely relevant is easy with ANSERA, non-engineers think it is the better tool. A real engineer gets frustrated, because getting the relevant documents out of the heap of close fields is impossible. But if you got the mention of how to use classification symbol searches only once within a haystack of other information in a new environment, retrieving this how-to when you need to is not something your brain will remember, so you pick some documents out of the large stack, and miss THE relevant classification symbol completely.
And since Quality Nominees get only 2 hours to do this check, including understanding the application, evaluating if the cited documents have been evaluated correctly (X/Y; technical features), checking clarity issues, ..., no time remains to read up on the seach strategy, or even do a quick and dirty search in the correct classification symbol. And without better documents, there is no reason to mark a low quality search in the quality recording tool.

To recap: ANSERA itself is a wonderful tool. If the broken features get repaired.
They are making great progress.
But neither ANSERA itself nor EPOQUE are automated.

The useability of the automated pre-search (of which one element uses ANSERA) is extremely dependent on field and the specfic application you want to search. It made progress, then they had new versions which ais, for my field, a setback. Other fields improved.

Re question 5: I presume sometime external users of EPOQUE/INTERNAL/XFULL will also get access to ANSERA. But that access is very limited to patent offices, or the cut-down version EspaceNet offers.

Look at things said...

The introduction of ANSERA reminds me somehow of the introduction of BEST. There were heavy investments in IT material, so it had to be shown that the investment was worth it. And it ended with BEST. It is not that it is bad as such, but it is not a reason for halving the training time as has happened.

It seems that the same mistake is repeated with ANSERA. Even in a worse way.

Doing a computer supported search requires a much more important intellectual input, than doing it in paper This is forgotten at the higher levels of the EPO.

The net result is the "optimum" BEST effect. Shoddy search=quick grant at the end. So production figures are higher and higher, but quality is lower and lower.... Is this really to the benefit of the users? I have some doubts.

When looking at case law, on appeal after refusal, the BA are quoting regularly new prio art not found during the original search. It might not be the role of tone BA to provide better prio art, but it is nevertheless symptomatic. And this only happening after a refusal. I dare imagine how many crappy searches have ended with a grant.....

This comment is not to be used by Techrights, be it directly or indirectly!

Cynic said...

Fair comments. My experiences are reflected in your analysis. The new system is a bit of a black box which gives me results - but are they the best? In areas where classification systems are necessary and searching means looking at a lot of similar documents, finding small details isn’t easy.. Details of gear boxes which have gears, springs, fixed gears, internal gears etc. generate very general terms which may be closely located within almost every document. Finding the right conglomeration isn’t easy with a simple full text search for terms.
A difficulty is that the Epoque system still gives very good results and probably better than Ansera - for the moment.

Anonymous said...

I take it that comments about the comment moderation system are automatically eliminated....

Anonymous said...

Well , it may help not to mix technologies, functionalities and examiner targets. Ansera uses off the shelf technologies for performing searches, nothing that Google has not been doing for ages or any modern data retreival system. Not sure the EPO can seriously pretend to be a reference in searches while keeping technology of the 80’s. What is different in Ansera from say Google, is that it supports the query language unique to the EPO and yes , surprise surprise, searching with classes. I think that by now everybody knows that ANYTHING is used as an excuse for higher targets.

Morepatents said...

Things are getting really weird at the EPO lately. The production objectives for 2018 are up another +10% over 2017. To help the examiners achieve that target, 2 days public holidays are cut, classification investments and the training and travel budget for experienced examiners are down to zero and directors have been instructed to reject requests for parental leave. Note that the directors who objected last year are simply out of a job. Also: examiners will be moved from single rooms to double occupancy and the rest of the building shall be rented out. One can only wonder what is in for 2019.

Subscribe to the IPKat's posts by email here

Just pop your email address into the box and click 'Subscribe':