T 1416/06 (Text mining/BOEING) of 24.4.2009

Summary of Facts and Submissions

I. This appeal is against the decision of the examining division to refuse European patent application No. 00932572.1.

II. The following document will be referred to:

D2: M.W. Berry et al., "Low-Rank Orthogonal Decompositions for Information Retrieval Applications", Numerical Linear Algebra with Applications, I(1), 1996, 1-27.

III. According to the decision appealed the then main and first auxiliary requests contained undisclosed subject-matter. The invention according to the second auxiliary request was not inventive, and claim 1 of the third auxiliary request was not clear.

IV. With the statement setting out the grounds of appeal dated 17 July 2006, the appellant requested that the decision be set aside and a patent be granted based on the claims according to the main request or auxiliary requests A to G, all filed with the same letter. The main request and auxiliary request C corresponded in essence to the second and third auxiliary requests before the examining division, respectively. Auxiliary requests D-G were filed as a matter of precaution in case the preceding requests were not found to meet the requirements of Article 123(2) EPC.

V. In a communication accompanying the summons to oral proceedings requested by the appellant as an auxiliary measure, the Board stated that it had doubts whether the characterising features solved a technical problem. Providing "verbal semantics", ie "words or terms" for describing the meaning of a dimension seemed basically to be a presentation of information. The verbal semantics was directed to the mind of a user. It was therefore not clear that there was a technical effect indicating the presence of technical means for solving a technical problem. As to the auxiliary requests the Board stated that it was not apparent that any part of the invention according to the independent claims of these requests contributed to the solution of a technical problem. The steps were of a mathematical nature, serving to process information with the purpose of improving its presentation or visualisation to the user. Whether or not this was new and original could not play any role as long as no technical effect was achieved. As to the skilled person it was clear that mathematical skills were crucial. A mathematician was however not a technically skilled person but could at most be part of a technical team.

VI. Oral proceedings were held on 24 April 2009. In the early hours of the same day the representative had sent a fax stating that the appellant had decided not to attend the oral proceedings. The oral proceedings were held in the appellant's absence.

VII. The Board verified that the appellant had requested in writing that the decision under appeal be set aside and a patent be granted on the basis of the main request or auxiliary requests A to C, filed with the statement setting out the grounds of appeal, or on the basis of auxiliary requests D to G corresponding to the foregoing requests but including in the independent claims all the features of dependent claims 5-12 as originally filed.

VIII. Claim 1 according to the main request reads:

"Method of representing a document collection in a computer system, wherein the document collection comprises a plurality of documents, with each document comprising a plurality of terms, using a matrix decomposition based on a distribution of the frequency of occurrences of each of the terms in each of the documents, the method comprising:

(a) constructing a term frequency matrix, having a dimension represented by the documents and another dimension represented by the terms, wherein each element of the term frequency matrix is the frequency of occurrence of one of the terms in one of the documents;

(b) determining a projection type;

(c) determining a lower dimensional subspace;

(d) determining a number of matrix dimensions to use;

(e) generating an original term subspace by projecting

the projection type into the lower dimensional subspace,

wherein the step of projecting comprises performing a

truncated two-sided orthogonal decomposition of the term frequency matrix according to the determined number of dimensions, the decomposition identifying significant features using three matrices: a term basis matrix (Uc), a weight matrix (Rc,Lc) and a document basis matrix (Vc), wherein the term basis matrix and the document basis matrix both have orthonormal columns; characterized by

(f) providing verbal semantics for at least one dimension of the subspace, wherein providing verbal semantics for the dimension comprises:

(f1) identifying a column in the term basis matrix that represents a desired dimension;

(f2) identifying a plurality of terms in the column with the largest absolute values for positive elements;

(f3) identifying a plurality of terms in the column with the largest absolute values for negative elements; and

(f4) returning the identified positive elements and the identified negative elements as a contrast set describing the meaning of position along the dimension".

IX. According to auxiliary request A the preamble of claim 1 is the same as for the main request whereas the characterising part reads:

"characterized by

(f) providing a legend for at least one dimension of the subspace, wherein providing a legend comprises:

(f1) identifying a column in the term basis matrix {or term frequency matrix ?} [sic] that represents a desired dimension;

(f2) identifying a plurality of terms in the column with the largest absolute values for positive elements;

(f3) identifying a plurality of terms in the column with the largest absolute values for negative elements; and

(f4) returning the terms with the identified positive elements and the identified negative elements;

characterized by [sic]

(f5) generating a legend for the dimension, the legend consisting of one or more terms of the identified positive elements attached to one direction of the dimension and one or more terms of the identified negative elements attached to the other direction of the dimension".

X. According to auxiliary request B the following features are added to claim 1 of the main request:

"(g) performing information visualization, wherein performing information visualization comprises:

(g1) in response to a user request for dimensions of visualization, determining a number of dimensions;

(g2) computing the requested dimensions;

(g3) determining whether the user has requested fewer than the determined number of dimensions;

(g4) if the user has requested fewer than the determined number of dimensions, using a set of default

dimensions;

(g5) determining if the dimensions are orthogonalized;

(h) if the dimensions are not orthogonal, taking an appropriate action based on a user preference, wherein the user preference comprises one of:

(i) orthogonalizing the dimensions; and

(ii) providing an indication that the dimensions are not orthogonal;

(i) projecting the documents onto the dimensions;

(j) generating labels for the dimensions from the returned positive and negative elements;

(k) displaying a plurality of indicators, wherein each indicator represents a document, on a plurality of labeled axes corresponding to the dimensions".

XI. According to auxiliary request C the preamble of claim 1 is the same as for the main request whereas the characterising part reads:

"characterized by

(f) providing verbal semantics for at least one dimension of the subspace;

(g) updating the original term subspace as the document collection changes, wherein said updating the original term subspace comprises:

- identifying a plurality of new documents;

- identifying a plurality of new terms in the new

documents;

- constructing a new term frequency matrix that

represents the new documents;

- projecting the new term frequency matrix on the original term subspace, the term basis matrix;

- computing a residual;

- augmenting the existing term subspace with the

residual;

- expanding the original term subspace, wherein expanding the original term subspace comprises expanding the document basis matrix by adding a small identity matrix; and

- re-identifying significant features in the

subspace".

XII. No sets of claims according to auxiliary requests D-G were filed. The appellant merely indicated in the statement setting out the grounds of appeal that the claims according to these requests were to correspond to the claims of the main request and auxiliary requests A-C, respectively, with the independent claims additionally including all features of claims 5-12 as originally filed.

XIII. At the end of the oral proceedings the Board announced its decision.

Reasons for the Decision

All requests

1. Construction of claim 1

Claim 1 is directed to a method of "representing a document collection in a computer system". On the basis of the description (see eg p. 10, l. 10-14 and fig. 2) this formulation is taken to mean that the method is performed by means of a computer system, not that the collection is stored in a computer system.

2. Exclusion under Article 52(2) EPC

With the above interpretation of claim 1 the method is an invention within the meaning of Article 52(1) EPC since it comprises a computer system.

3. Field of technology

3.1 Claim 1 of all requests is directed to a method of representing a document collection. The method is to a large extent defined in terms of equations. The purpose of the method is to present the information in a way that can be more easily understood or evaluated by a user. A fundamental question in this context is whether the invention - apart from its being implemented on a computer - is within a field of technology. At the bottom of the method is a mathematical technique known as orthogonal decomposition. This technique is generally applied to large matrices and, like many mathematical functions, can be represented graphically. It is typical for mathematical representations that they involve pure numbers, ie abstract data, having no physical connotation. In the present invention the representations are of documents and the terms used in the documents. Thus, although the data have a certain "meaning", they remain abstract. They can hardly be regarded as forming a physical entity, nor does the method result in a change in the data but merely in their representation (cf T 208/84 "Computer-related invention/VICOM", OJ EPO 1987,14, point 5 of the Reasons). It could therefore be argued that the invention - again apart from its implementation - is essentially a mathematical method pursuant to Article 52(2)(a) EPC, resulting in a presentation of information pursuant to Article 52(2)(d) EPC.

3.2 In the following these general concerns will however not be pursued since there are more specific reasons for not allowing the appeal.

Main request

4. The prior art

The appellant acknowledges (statement setting out the grounds of appeal, point 4.2, first paragraph) that D2 discloses the preamble of claim 1. The examining division held that in addition it implicitly disclosed feature (f) of providing verbal semantics except for the identification of a plurality of terms in the column with the largest absolute values for positive and negative values (decision under appeal, point 3.3). The Board agrees with this view since in the diagram shown in fig. 4 of D2 all terms are indicated, implying that the corresponding columns have been identified and the respective elements returned.

5. Novelty

According to the invention a plurality of terms in the column with the largest absolute values for positive and negative values are identified. These terms are used to describe the meaning of respective dimensions. In D2 all terms are identified, not just the largest ones. It could be argued that the claim feature is not actually a distinction since according to D2 the largest values are indeed identified (as a sub-group of all values). However, to the benefit of the appellant it is assumed that a means for identifying the largest values (and discarding smaller ones) is implied. This would be a new feature (Article 54(1)(2) EPC 1973).

6. Inventive step

The invention serves to indicate to the user what terms a document is likely to contain (see the paragraph bridging pp. 21 and 22 of the description). The solution concerns the legends (labels consisting of one or more words) describing the axes (dimensions) so that documents near one end of the dimension tend to contain the words at that end or words correlated with those words in the document set. As most people are aware, diagram axes are usually labelled. The question in the present case is therefore whether it was obvious to label the axes with the most dominant term or terms. Apart from any personal preferences of the users the answer appears to depend on the size of the database. A data base visualisation may comprise several hundred dimensions (as indicated on the first page of D2). The axis labels could not indicate all associated terms but a selection must be performed, and common sense dictates that the dominant terms should be selected. Thus the subject-matter of claim 1 does not involve an inventive step (Article 56 EPC 1973)).

Auxiliary request A

7. Inventive step

Ignoring the comment "{or term frequency matrix ?}" in claim 1, the Board finds that the only difference between auxiliary request A and the main request is that a legend containing the axis information is generated. This feature has already been considered above. Thus also this subject-matter does not involve an inventive step (Article 56 EPC 1973).

Auxiliary request B

8. Clarity, support in the description

Claim 1 has been supplemented by features of original claim 23 including the features "in response to a user request for dimensions of visualization, determining a number of dimensions" and "determining whether the user has requested fewer than the determined number of dimensions". These features are contradictory.

The claim also reiterates from claim 23 the feature that if a user has requested fewer than the determined number of dimensions, a set of default dimensions is used. This could be taken to mean that the default dimensions (entirely) replace the requested dimensions. On p. 21, l. 11-13 of the description, however, it is explained that the user request is "filled out with default axes", not that it is replaced.

Therefore the claim is not clear and supported by the description in the sense of Article 84 EPC 1973. In the following it will be interpreted in accordance with the description.

9. Inventive step

The additional features in claim 1 permit the user himself to define dimensions of visualisation. If necessary the dimensions are orthogonalised. Axis labels are generated.

The wish to be able to study particular aspects (terms) of the data collection follows directly from a user's particular interests, which are known to be subjective and therefore cannot render the problem formulation inventive. Given that such interests might exist, it was clear that the system should be designed such that the displayed axes are selectable. This implies choosing a particular subspace of the total term space, ie a particular projection, a technique that is well understood. The choice offered to the user between orthogonalisation or no orthogonalisation merely takes into account the fact that different users have different preferences.

Thus the subject-matter of claim does not involve an inventive step (Article 56 EPC 1973), and the request is refused.

Auxiliary request C

10. Clarity

The examining division decided that the features "small identity matrix" and "significant features" were not clear (decision under appeal, point 4.1). The Board agrees. In addition, the feature "computing a residual" leaves open what this residual is and how it is computed. The corresponding part of the description (fig. 7 and p. 19, top) contains no equations and no further explanations. Also this feature must thus be regarded as not clear (Article 84 EPC 1973). Therefore auxiliary request C is refused.

Auxiliary requests D-G

11. These requests correspond to the four foregoing requests but include in the independent claims all the features of dependent claims 5-12 as originally filed. The appellant filed these requests in case the Board would be of the opinion that the foregoing requests were not allowable under Article 123(2) EPC. It was not argued that the additional features rendered the claimed subject-matter inventive, nor is this apparent to the Board. Thus these four last requests are also refused (Article 56 EPC 1973).

Non-attendence at oral proceedings

12. The appellant's fax indicating his intention not to attend the oral proceedings was received by the EPO at 4.29 hrs on the day of the oral proceedings so that the Board could not have been made aware of it before the oral proceedings were opened. The Board therefore had to wait and see whether somebody representing the appellant might have been delayed and eventually had to ask the registrar of the Board to make inquiries by phone calls. This situation is clearly undesirable and should be avoided by providing information of non-attendance in due time.

ORDER

For these reasons it is decided that:

The appeal is dismissed.