Tuesday, 21 July 2015

T 306/10 - Recommender systems not patentable


Can an improved recommender system be inventive? Claim 1 of the main request concerned a 'method of discovering relationships between items'. In response to a query item, user logs are found in which the query occurs. Based on the logs a result item is then identified. 

One problem with such recommenders is that best sellers tend to occur in many user logs. Thus, such an algorithm would tend to recommend best sellers in response to any query. The invention in this application solved this problem by identifying an item which is over-represented compared to all user logs. An auxiliary request also outputs the identified result item as a recommendation.

However, the board will have none of it. Improvement in the selection of an item for recommendation is regarded as subjective, and thus not technical, and thus not patentable. A lot of non-trivial research is done on recommender systems, however; much of it involving mathematics (see this example). Following this decision, none of it can be protected.

The decision uses the notorious general-purpose computer as closest prior art. Some other boards seem to dislike this and have held that notorious should be interpreted narrowly (T698/11; T690/06; T359/11). In this regard it would be interesting if the amendments suggested in an obiter dictum (R.4.8) were carried out, to see if that could at least have swayed the board to adopt a more realistic starting point for inventive step.



 


Reasons for the Decision


3. Background of the invention

3.1 The present invention is concerned with the discovery of relationships between items on the basis of item selections of a plurality of users in the context of making user-specific recommendations.

3.2 The background section of the description explains inter alia that it is known for online commerce sites to keep track of user purchases and, on the basis of such purchases, to make recommendations of products and services likely to be of interest to a particular user. Such recommendations may be based on an analysis of the purchases of other users who have purchased the same products and services. This technique is said to lead to inaccurate results as relatively few data points may be available. A typical user may make four or five purchases annually from any particular online store, which is insufficient to develop a reasonably accurate user profile in a relatively short period of time. In addition, some purchases may be gifts, and may thus fail to accurately reflect personal preferences of the purchaser. Distortions may furthermore result from the fact that the merchant may not be able to easily determine whether the purchaser was satisfied with the product.

3.3 The background section further explains that a commonly used technique for making recommendations based on data analysis performed on observed user behaviour is to observe that people who buy a particular product X also tend to be more likely to buy a particular product Y. Thus, the system may suggest, to a user who is observed purchasing (or browsing) product X, that he or she may also be interested in product Y. This technique is said to often lead to inaccurate results, in particular when the observed purchase is a relatively rare product. Relationships between such products tend to be overstated, since relatively few data points are available for both the purchased product and the suggested product. In addition, certain products, such as best-sellers, tend to appeal to virtually all consumers, so that co-occurrence is seen between a best-seller and nearly every other product.

4. Main request - inventive step

4.1 Claim 1 of the main request is directed to a computer-implemented method of discovering relationships between items. It starts with receiving, for a plurality of users, a set of "item selections", the item selection being "detected from observed behavior of each of a plurality of users". For each user, identifiers for the items selected by that user are stored in a "user log". Upon receipt of a query including at least one item identifier, a "score" is assigned to each user log. The Board understands this score to be a function of the number of occurrences of the at least one item identifier in the user log. Based on these scores, a subset of user logs is identified. The Board understands this step as being to select, for example, the ten user logs with the highest scores. From this subset of user logs at least one "result item" is identified, the result item being an item which "occurs more frequently in the subset of user logs than expected based on the occurrence of the item in the entire set of user logs".

4.2 The Board interprets the feature "receiving item selections detected from observed behavior of each of a plurality of users" merely as a step of receiving item selections of each of a plurality of users. Whether a particular item selection is "detected from observed user behaviour" is not a property of the item selection, and the claim does not include separate steps of observing user behaviour and detecting item selections from the observed behaviour.

4.3 The subject-matter of claim 1 hence essentially amounts to a computer-implemented method of discovering relationships between items on the basis of item selections using an abstract mathematical algorithm.

A mathematical algorithm contributes to the technical character of a computer-implemented method only in so far as it serves a technical purpose (see decision T 1784/06 of 21 September 2012, reasons 3.1.1).

4.4 In the statement of grounds of appeal, the appellant essentially argued that the method of claim 1 solved the technical problem of obtaining more relevant results in response to a query and of saving a user time when performing a search. Besides its use in direct marketing, it could also be used in non-business-related fields such as making better recommendations of content, e.g. songs, for a personalised radio station.

However, the purpose of improving search results is not expressed in claim 1. The appellant's arguments therefore cannot convince.

4.5 In fact, claim 1 is not limited to any specific purpose other than the identification of at least one "result item" that bears a particular statistical relationship in terms of item selections to at least one "query item". In the Board's view,

4.6 Since the mathematical algorithm does not contribute to the technical character of the claimed method, an inventive step can be present only in its technical implementation. However, the claim in this respect merely specifies that the method is "computer-implemented". While the claim does use wording such as "generating", "receiving" and "user log", these terms, although reinforcing the point that the algorithm is computer-implemented, do not imply any specific implementation details. The Board further has no doubt that the skilled person, who in this case is a computer programmer, would have no difficulty in implementing the steps of claim 1.

4.7 The conclusion is therefore that the subject-matter of claim 1 lacks an inventive step within the meaning of Articles 52(1) and 56 EPC over a notorious general-purpose computer.

4.8 For the sake of completeness, the Board notes that it would have reached the same conclusion had claim 1 been amended to include separate steps of observing user behaviour and detecting item selections from the observed behaviour (see point 4.24.2 above), as such steps are known in the art (see points 3.2 and 3.3 ) and do not lend the mathematical algorithm of claim 1 a technical character.

5. First auxiliary request - inventive step

5.1 Claim 1 of the first auxiliary request differs from claim 1 of the main request essentially in that the identified result item is outputted as a recommendation.

5.2 In the Board's view, the selection of an item, for example a song, for recommendation to a user does not qualify as a technical purpose. From a technical point of view it is irrelevant what songs are recommended to a user. While making "good" or "bad" recommendations may lead to different user reactions and thereby, in the end, to different technical results (the user might for example play more or fewer songs, or issue more or fewer search queries in order to find other songs), such results do not qualify as a technical effect of the recommendations, as they depend on subjective choices made by the user (cf. decision T 1741/08 of 2 August 2012, reasons 2.1.6).

5.3 It follows that the amendments to claim 1 cannot overcome the objection of lack of inventive step raised in respect of the main request. Hence, the subject-matter of claim 1 of the first auxiliary request likewise lacks an inventive step (Articles 52(1) and 56 EPC).

6. Conclusion

Since neither of the requests on file is allowable, the appeal is to be dismissed.


This decision T 306/10 (pdf) has European Case Law Identifier:  ECLI:EP:BA:2015:T030610.20150204. The file wrapper can be found here. Photo "Last.fm Wallpaper"  by Rodrigo Galindez (Generated using lastfm.alekc.org/) obtained via Flickr under CC BY 2.0 license (no changes made).

 

2 comments :

  1. Romano Beitsma23 July 2015 at 09:26

    Interesting post BUT the statement "Following this decision, none of it [= recommender systems] can be protected" seems a bit bold.

    First, we are talking about the EPO, so about protection in Europe (only). What can and cannot be protected in the US, China and Japan is not the issue here.

    Second, this is a Board of Appeal decision in 2015, made by a particular Board. Another Board may come to another conclusion, and two years down the road even the same Board may take a different stance. As long as we don't have a definitive decision by the Enlarged Board of Appeal, the case law can (and probably will) be fluid.

    Third, this decision applies to a certain type of improvement in a certain type of recommender systems. Perhaps it proves to be possible to make a technical (that is, non-subjective) and therefore patentable improvement to a certain type of recommender system.

    ReplyDelete
    Replies
    1. You are quite right of course. Thanks for adding this additional comment.

      Delete

Statcounter