Classification of search engine resources using deep learning: non-technical

The application underlying the present decision relates to spam classification of search engine resources. However, the European Patent Office refused to grant a patent since the distinguishing features relating to the allegedly non-technical spam classification would not produce a technical effect. Here are the practical takeaways of the decision T 0874/19 (Classifying resources using a deep network/GOOGLE) dated July 6, 2022 of Technical Board of Appeal 3.5.07:

Key takeaways

The invention

According to the present application, internet search engines aim at identifying resources relevant to a user’s needs or interests. A user submits a search query and receives a set of search results, each of which identify a corresponding search enginge resource (cf. application as filed, page 1, background section).

In order to avoid manipulation of these search results by a spammer, the present application discloses an approach that classifies search engine resources as either a spam resource (i.e., the resource belonging to the “spam” cateogry) or as a not spam resource (i.e., the resource beloning to the “not spam” category) (cf. application as filed, page 5). For this purpose, a score representing a predicted likelihood is generated for each category.

Fig. 1 of WO 2014/160282 A1

Here is how the invention is defined in claim 1 of the main request:

Is it technical?

According to the Board, in accordance with the Appellant’s view, D1 forms the closest prior art for the subject-matter of claim 1.

D1 discloses the idea of combining deep-learning and large-margin discriminative learning called “DNet-kNN” based on a deep encoder network. The approach of D1 is applied to classify newsgroup text data into four categories (“computer”, “recreation”, “science” and “talks”).

Compared to D1, the Board identified the following distinguishing features over D1:

a respective category score (226) for each category in a pre-determined set of categories,

wherein each of the respective category scores measure a predicted likelihood that the resource belongs to the corresponding category.

Regarding the claimed categories the Board noted that these may be:

a “spam” category and a “not spam” category

a “not spam” category and a category for each type of spam, such as “content spam”, “link spam” and “cloaking spam”, etc.

a category for each resource type of a group of resource types, including “news resources”, “blog resources”, “forum resources”, “shopping resources”, “product resources” and “political resources”

However, according to the Board, it was not apparent what further technical effect the distinguishing features relating to these categories could produce nor what objective technical problem is to be solved.

The appellant argued that following decision T 1227/05 (Circuit simulation I/Infineon Technologies) the claimed system could at least be regarded as simulating a hardware circuit that classifies inputs and thus has a technical purpose.

However, the Board stated:

The board considers that decision T 1227/05 cannot be followed as argued by the appellant in view of the recent decision G1/19, Reasons 133.

As a result, the Board stated that claim 1 of the main request does not provide any technical effect which goes beyond its straightforward implementation in one or more computers.

Therefore, the Board dismissed the appeal due to lack of inventive step.

More information

You can read the whole decision here: T 0874/19 (Classifying resources using a deep network/GOOGLE)

Please share this article if you enjoyed it!