Information
This decision is only available in German.
T 1632/19 09-02-2023
Download and more information:
ASCRIBING ACTIONABLE ATTRIBUTES TO DATA THAT DESCRIBES A PERSONAL IDENTITY
Inventive step - main request
Inventive step - (no)
Amendments after summons
Amendments - not all justified by possible exceptional circumstances
Amendments - raised new objections
Amendments - not admitted
I. The appeal is against the decision of the examining division refusing the European patent application No. 11 769 597 (published as WO 2011/130526 A1) on the ground that the sole request before it lacked clarity (Article 84 EPC) and did not involve an inventive step (Articles 52(1) and 56 EPC).
II. Reference is made to the following document, cited in the decision under appeal:
D1: US 2009/0164454 A1
III. At the end of the oral proceedings before the board, the appellant's (applicant's) requests were that the decision under appeal be set aside and that a patent be granted on the basis of the claims of the main request, filed with the statement setting out the grounds of appeal, or of the auxiliary request, filed with the letter dated 8 February 2023, or that the case be remitted to the examining division for further prosecution. The appellant further requested that the expert declaration filed with the letter dated 8 February 2023 be admitted into the proceedings.
IV. Claim 1 of the main request, reads as follows:
A method comprising:
(a) receiving an inquiry (103) to initiate a search for data concerning a specific individual;
(b) extracting, from said inquiry (103), an attribute that is relevant to a search of a reference database (110), wherein said inquiry (103) includes an inquiry value for said attribute;
(c) obtaining, from a first table, a weight that is indicative of aa [sic] relative impact of said attribute in finding a match to said inquiry in said reference database;
(d) obtaining, from a second table, a number of records in said reference database that have said inquiry value for said attribute;
(e) modifying said weight based on said number of records in said reference database that have said inquiry value for said attribute, thus yielding a predictive weight for said attribute, where K = W x F, where K is said predictive weight, W is said weight, and F is said total number of records;
(f) setting K1 = K, wherein said attribute is a first attribute;
(g) repeating operations (b) through (e) for a second attribute through an Nth attribute, thus yielding K2 through KN ;
(h) establishing a function that characterizes a relationship among said field1 through said fieldN as an indication of a degree of relationship of said field1 through said fieldN to said inquiry, wherein said establishing is performed prior to retrieving candidates for said match from said reference database;
(i) searching said reference database for said match, wherein said searching comprises:
retrieving from said reference database, based on said function, candidates having attribute values that indicate likely matches to said inquiry;
determining a best candidate from said candidates; and
returning said best candidate as said match;
(j) outputting said match; and
(k) outputting an indicator of a level of confidence that said match is an appropriate match to said inquiry.
V. Claim 1 of the auxiliary request is worded as follows (added features with respect to claim 1 of the main request underlined by the board; numbering (i)-(v) by the board):
A method comprising:
(a) receiving an inquiry (103) to initiate a search for data concerning a specific individual; and wherein process (120) analyzes data to identify specific data fields that are associated to attributes in order to enhance the opportunity to identify matches from reference database (110) by using those data fields singularly or in pluralities, and
wherein said attributes can be learned by updating with data from data fields (140) to include attributes that have not previously been recognized, to which predictive weighting and other information is defined to recognize not previously recognized attributes for future execution of said process (120) for said inquiry (103);
(b) extracting, from said inquiry (103), an attribute that is relevant to a search of a reference database (110), wherein said inquiry (103) includes an inquiry value for said attribute;
(c) obtaining, from a first table, a weight that is indicative of a relative impact of said attribute in finding a match to said inquiry in said reference database;
(d) obtaining, from a second table, a number of records in said reference database that have said inquiry value for said attribute;
(e) modifying said weight based on said number of records in said reference database that have said inquiry value for said attribute, thus yielding a predictive weight for said attribute, where K = W x F, where K is said predictive weight, W is said weight, and F is said total number of records;
(f) setting K1 = K, wherein said attribute is a first attribute;
(g) repeating operations (b) through (e) for a second attribute through an Nth attribute, thus yielding K2 through KN ;
(h) establishing a function that characterizes a relationship among said field1 through said fieldN as an indication of a degree of relationship of said field1 through said fieldN to said inquiry, wherein said establishing is performed prior to retrieving candidates for said match from said reference database; and
(i) wherein said function (150) and/or f(x) is a listing of weighted attributes, where the weight of a particular attribute or combinations of attributes is indicative of a predictiveness, and therefore an importance, of that attribute or combinations of attributes in identifying an appropriate match to a record in said reference database (110); and
(ii) wherein said predictiveness is a value of said function (150) and/or f(x) equal to a frequency of attribute multiplied by a weight and/or relative impact given to said attribute, which equals a value of K and/or said predictive weight of said attribute and/or attributes; and
(iii) wherein said value of K is a representation of said established function that characterises a relationship among said field1 through said fieldN; and
(iv) wherein said lower value of K would make for a more effective and efficient search to be performed to find and provide the desired results; and
(v) wherein said lower value of K would make for a less effective and efficient search if said function (150) and/or f(x) was limited in the listing of weighted attributes and therefore number relative to predictiveness;
(i) searching said reference database for said match, wherein said searching comprises:
retrieving from said reference database, based on said function, candidates having attribute values that indicate likely matches to said inquiry;
determining a best candidate from said candidates; and
returning said best candidate as said match;
(j) outputting said match; and
(k) outputting an indicator of a level of confidence that said match is an appropriate match to said inquiry; and
wherein said indicator is feedback expressed in three components: (1) Confidence Code, which indicates the relative degree of confidence in similarity between data fields (140) and candidates in reference database (110); (2) Match Grade String, which indicates the degree of similarity between attributes of data fields (140) and candidates in reference database (110); and (3) Match Data Profile which indicates the type of data in reference database (110) that was used in the match experience.
VI. The appellant argued essentially that the defined function was directing the search algorithm to specific addresses in the computer's memory and, hence, it produced a technical effect. It should therefore not be considered a non-technical mathematical method. Regarding the admission of the auxiliary request, the appellant argued that the board had raised new objections for the first time in its preliminary opinion and it should be given an opportunity to respond. There were, thus, exceptional circumstances which justified the admission of the auxiliary request.
1. The claimed invention
1.1 The claimed invention relates to searching an individual in a database. A user enters an inquiry comprising specific values for the attributes to be used in the search (e.g. name, address, telephone number, etc.; the attribute is e.g. the family name and the value is "Smith").
1.2 Before the search is carried out, the attributes are extracted from the inquiry and a weight is obtained from a table, the weight being indicative of the relative impact of the attribute in the search. For example, a first name would have a lower impact than the last name, because it is more probable that many people have the same first name than the same last name. From a second table the number of records in the database that have the same value for said attribute (e.g. how many individuals with last name "Smith" are in the database) is obtained. A predictive weight for the attribute is calculated as product of the obtained weight and number of records.
1.3 After such predictive weights are calculated for all the attributes in the inquiry, a function is established based on the results, and this function is used for the search in the database.
2. Main Request, inventive step
2.1 It was common ground that D1 represented the closest prior art.
D1 describes a system and a method for searching venues such as restaurants. The search is based on attributes such as venue-type, cuisine-type, price range, payment method, etc. (see paragraphs [0018] to [0034]). D1 uses in addition factors (corresponding to the weights of the present application) such as importance factor, uniqueness factor, prominence factor, etc. These factors are applied to the attributes (see paragraphs [0038] to [0040]). The "importance factor" is used to specify the relative importance of an attribute among a set of attributes ([paragraph 0041]). The "uniqueness factor" is a measure of uniqueness of a particular attribute relative to other attributes. The uniqueness factor is venue-independent, but attribute-specific. For example 5,000 venues may exist in a database but only 500 of those venues may offer the "Karaoke" attribute. Accordingly ... the "Karaoke" attribute may be weighted by a uniqueness factor ... [1-500/5000]; or 0.9 ([paragraph [0042]).
The "importance factor" is taken thus to correspond to the weight that is indicative of a relative impact of said attribute in finding a match (i.e. "W", see features (c) and (e)) of claim 1, while the "uniqueness factor" to the number of records in said reference database that have said inquiry value for said attribute (i.e. "F", see features (d) and (e)).
According to D1, the search is carried out first and then the results are weighted with the factors (see paragraphs [0046] and [0047]).
It was thus common ground that the difference between claim 1 of the main request and D1 lied essentially in the definition of the function and the modified weights and that the (modified) weights were applied before the search is performed in the database. In D1, the weights were applied after the search in the database has been carried out, in order to "filter" the results.
2.2 In the decision under appeal, the examining division held that the features relative to the calculation of the modified weights and the establishment of the function before the search was carried out related to a mathematical method which involved no technical considerations nor produced any technical effect. According to established case law and practice it would have been given to the skilled person for implementation in the method and system of D1, which they would have done without exercising any inventive skill (see points 14.8 to 14.12 of the Grounds).
2.3 According to the appellant, the features distinguishing the claimed method from D1 provided the technical effects that (see statement of the grounds of appeal, page 4, second paragraph):
- there was a nexus between the operations such that (i) obtaining the weight, modifying the weight, and establishing the function were performed before the searching, and therefore (ii) the searching was performed after the obtaining, modifying, and establishing; and
- by the modification step, the computer was technically controlled by being directed to a certain memory location.
The appellant argued further that the obtaining, modifying and establishing operations facilitate[d] the act of searching. The resulting method of searching for a record in a database that best matche[d] a given inquiry therefore identifie[d] and select[ed] match results in a more efficient and effective manner compared with the specific prior art method as disclosed by D1 (see ibid., page 4, third paragraph).
2.4 Regarding the order of the operations, the board agrees that indeed there is a difference between the claimed method and D1.
From the point of view of the user, however, there is no technical effect from the different order of operation. The user enters an inquiry based on specific attributes and receives the results of the search, which have been weighted using predetermined weights/factors. Whether the weighting is done before the search is executed or after has no effect on the final results of the search. The user receives the weighted results in both cases. So, the different order of carrying out the searching and the calculating/establishing the function does not provide any technical effect itself.
2.5 The next question is whether the use of the predictive weights and the function according to claim 1 of the main request produce any technical effect.
2.5.1 As a first remark, the board notes that neither the claims nor the application as a whole provide any information about how the search is carried out using the function established based on the calculated predictive (i.e. modified) weights. Claim 1 mentions only "retrieving from said reference database, based on said function, candidates ..." (see feature (i)).
According to the claims and the description, the attributes used for the search are weighted using the predictive weights (K). According to the board's understanding, in case of using e.g. the attribute "First name" and the value "John" this would mean that instead of the inquiry being "First name=John" it would be K times "First name=John", where K is a number (K=WxF), so for example 2500 times "First name= John". The application does not provide any information, how this search query is to be understood.
2.5.2 During the oral proceedings, the appellant explained that K was to be taken as an indication of the importance of the search term, and that it could be defined, for example, that if K were under a predetermined threshold (e.g. 150) the specific attribute/search term would not be taken into account. The search result would then be based only on attributes, which according to their corresponding predictive weights were considered important. Exceptions could be made when there was only one or very few attributes used in the search function and omitting one of them would render the search impossible or meaningless.
2.5.3 There is nothing in the application, however, that could support or hint at this explanation by the appellant. Moreover, it cannot be considered that such an understanding of the search function would be part of common general knowledge. In the board's view, the skilled person remains with no real information as to how to use the defined function in carrying out a search.
2.6 Moreover, the information provided in the application is contradictory or at least not clear. In paragraphs [0034] to [0047] the application describes how the predictive weights for the various attributes are calculated. This calculation is a multiplication of a predetermined weight for each attribute (see Table 4) with the frequency of the specific value of the attribute in the database (see Table 5). The function f(x) is a summation of the products of the respective multiplications for each attribute in the inquiry. It is to be understood that the higher the product/predictive weight ("K"), the more impact the specific attribute has in the summation and hence to the established function.
2.6.1 Paragraph [0048], however, provides contradicting information on this point. Comparing an example based on the first name "John" (which has a relatively low weight of 0.25 and a high frequency of 10.000; predictive weight K=0.25x10.000=2500) with the mobile number "1234567890" (which has a higher weight of 1 and a frequency of 1; predictive weighting K=1x1=1), the application concludes that it appears that in f(x), first name "John" has a greater predictive weighing that mobile phone number "1234567890". However, based on the actual logic that is executed, the lower f(x) may be more predictive than the higher f(x).
2.6.2 The board notes that the application does not give any information regarding when it would be that the lower f(x) may indeed have a higher predictive weighting than the higher f(x). During the oral proceedings, the appellant explained that the establishment and the use of the function is left to the skilled person. The skilled person would select which attributes to include and which not and would also decide whether a low or a high f(x) is more predictive, and assess the results accordingly.
2.6.3 The board cannot accept this argument. Claim 1 defines that a function has to be established using the calculated predictive weights and this function is used in carrying out the search. It cannot be left to the skilled person to decide on a case by case basis, and without predetermined, unambiguous criteria when a low predictive weight is more important than a high predictive weight. Otherwise, the whole concept of establishing a function using calculated predictive weights would not have any purpose at all.
2.6.4 The board's conclusion is, thus, that the skilled person does not get any concrete information about how the function f(x) is actually established and used in the search. The appellant's assertions that the claimed method facilitates the search or that it provides for a more efficient and effective search are thus not supported by the application and remain mere allegations without proof, which the board cannot accept.
2.7 Regarding the second technical effect identified by the appellant (see point 2.3 above), the appellant, referring to the submitted expert's opinion, reiterated its argument that by the modification step, the computer was technically controlled by being directed to a certain memory location. Hence, there was a technical effect from the calculation of the predictive weights, and the related features should not be considered a mathematical method as such.
2.7.1 The board, for the sake of discussion, admits the expert opinion into the proceedings. It notes, however, that the opinion is rather general, stating that the whole idea of the Search ... is to direct the attention of an algorithm into a large corpus of data. In this particular case, that corpus of data is held in memory, therefore, it is intuitively obvious for the skilled person that the search, by the claimed features relating to the recalculation of the weights before the Search is carried out, is directed to a specific address in the memory (see last paragraph of the expert's opinion filed with the letter dated 8 February 2023; highlighting omitted by the board).
2.7.2 If it were to be accepted that any search instruction/query directs the corresponding search algorithm to certain data, stored in specific memory locations, then any search query would produce a technical effect. Irrespective of whether it is true or not, this would be valid for the search in D1, as well.
However, the relevant question is rather whether the calculation of the predictive weights and the use of the function f(x) as defined in claim 1 directs the algorithm to a different memory location. In other words, whether the recalculation of the weights and the use of the function changes/affects the memory location where the algorithm is directed to by the search instructions. As explained above, the application does not provide any information about how the function and the recalculated ways are (to be) used in the search or how they might affect the execution of the corresponding search algorithm. The expert's opinion - taken into account although it was late filed - does not provide any relevant explanations, either. Hence, the answer to the question, has to be negative.
2.8 Summarising, the board agrees with the examining division that the features distinguishing the claimed method from D1 are not technical, and would be given to the skilled person for implementation. As purely mathematical features, their implementation is considered obvious.
The board's conclusion is that the subject-matter of claim 1 does not involve an inventive step (Articles 52(1) and 56 EPC).
3. Auxiliary request
3.1 The auxiliary request constitutes an amendment of the appellant's case which was filed after the board had issued the summons to oral proceedings. Whether it is taken into account is to be decided under Articles 13(1) and (2) RPBA 2020.
3.2 The appellant argued that the auxiliary request was filed to address the board's objection that the application did not give any information regarding when it would be that the lower f(x) may indeed have a higher predictive weighting than the higher f(x). This objection was raised for the first time in the board's preliminary opinion and constituted exceptional circumstances in the sense of Article 13(2) RPBA 2020. Since the auxiliary request addressed this objection, it should be admitted into the proceedings.
3.2.1 Without going into the discussion on whether or not the identified objection was raised for the first time by the board and whether or not it constitutes exceptional circumstances, the board notes that claim 1 of the auxiliary request includes extensive amendments which do not relate to that particular objection. For example the features added in (a) and (k) (see point V. above) do not relate to that objection. Also the features added in (h), points (i), (ii) and (iii) do not relate to that objection, either.
Hence, even if it were accepted that there were indeed exceptional circumstances, these circumstances would not justify the filing of all the amendments included in claim 1 of the auxiliary request.
3.3 The board refers also to the criteria under Article 13(1) RPBA 2020 and notes that amended claim 1 raises prima facie several new objections, such as:
3.3.1 Point (ii) in feature (h) corresponds essentially to feature (e). Both features define the same subject, i.e. that K=WxF, albeit in different wording. This is against the requirements of clarity and conciseness (Article 84 EPC).
3.3.2 Point (iii) in feature (h) defines that "wherein said value of K is a representation of said established function that characterises a relationship among said field1 through said fieldN" (emphasis by the board).
It is not clear to what "that" refers to.
- If it refers to "said value of K" then it is in contradiction to the first sentence of feature (h), according to which the function (and not K) characterizes a relationship between said field1 through said fieldN;
- If it refers to the "established function" then it repeats the definition of the first sentence of feature (h) in slightly different wording.
In either case, this feature does not fulfil the requirements of clarity and/or conciseness (Article 84 EPC).
3.3.3 Points (iv) and (v) in feature (h) are not directly and unambiguously disclosed in the originally filed application. In particular, paragraph [0048], which describes an example of when a lower value of K may be more predictive than a higher one, does not provide the sufficient support for this feature, as it discloses no general criteria.
3.4 Regarding clarity, the appellant argued that even if features were repeated in the claim, they were clear to the skilled person. Any small differences in the wording between the newly added paragraphs and the existing ones in feature (h) served to better define K. In addition, the skilled person would understand that K and the function had the same importance.
The board does not follow these arguments. Defining the same feature twice in the same claim with different wording does not help to define the feature better. On the contrary, it creates ambiguity to the skilled person as to what exactly this feature is supposed to mean. Even if the individual definitions were considered to be clear themselves, this does not change the fact that there are two different definitions for the same feature in the same claim rendering the claim not clear and not concise in the sense of Article 84 EPC.
3.5 Regarding the disclosure in the originally filed application, the appellant acknowledged that there was not literal disclosure of points (iv) and (v). It argued however that for the skilled person who knew that logic had to be programmed according to personal wishes they were implicit in paragraphs [0048] and [0049]. In any case, it was the skilled person that would decide when a low value for K would be more or less effective.
The board does not accept this argument, either. Paragraphs [0048] and [0049] describe an example when a low value of K may be more predictive than a high value. There is no hint of any general criteria that could be directly and unambiguously derived from those paragraphs and, more specifically, not the criteria defined in points (iv) and (v).
3.6 Summarising, the board does not take the auxiliary request into account under Articles 13(1) and (2) RPBA 2020 because not all the included amendments would be justified by exceptional circumstances (assuming there were such circumstances) and because it gives rise to several new objections.
4. Regarding the request for remittal of the case to the examining division for further prosecution, the appellant did not put forward any reasons for such a remittal.
As the board does not see any such reasons, either, and since there is no allowable request on file on the basis of which the application could be prosecuted further, the board decides the case on its merits, pursuant to the powers conferred by Article 111(1) EPC.
5. Since there is no allowable request on file, the appeal must fail.
For these reasons it is decided that:
The appeal is dismissed.