DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 24 February 2026 has been entered.
All previous objections and rejections directed to the Applicant’s disclosure and claims not discussed in this Office Action have been withdrawn by the Examiner.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 07 January 2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Amendments and Arguments
The Applicant’s arguments with respect to the claims have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. New reference Jahjah et al. is introduced to address online resources and misclassification communication to a user. New reference Kelly et al. is introduced to address clustering of multiple embedded fields.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 22-42 is/are rejected under 35 U.S.C. 103 as being unpatentable over CN 115599910, hereinafter referred to as Liu et al., in view of US 12002276, hereinafter referred to as Kelly et al.
Regarding claim 22 (New), Jahjah et al. discloses a method for identifying miscategorized transactions in accounts associated with an online resource (“The infrastructure modeling hub services (e.g., iModelHub™ services) 130 may interact with a number of other services in the cloud, that perform information management and support functions. For example, information management services (not shown) may manage asset data, project data, reality data, Internet of Things (IoT) data, codes, and other features. One such service may be a design validation cloud service 136 that evaluates the impact of design changes on performance of the infrastructure model, including project schedule, cost, and safety compliance. The design validation cloud service 136 may include a misclassification identification service 138 that is capable of automatically identifying elements of an infrastructure model that have misclassified, so they may be reviewed and corrected, thereby allowing the design validation cloud service 136 to provide better evaluations,” Jahjah et al., para [0033]. Here, “project” is interpreted as account and “service in the cloud” as online.), the method performed by a computing device associated with the online resource and comprising:
receiving, over a communications network coupled to the online resource, a transaction
for an account associated with a user (“FIG. 1 is a high-level block diagram of at least a portion of an example software architecture that may implement techniques for automatic identification of misclassified elements. The architecture may be divided into client-side software 110 executing on one or more computing devices arranged locally (collectively “client devices”), and cloud-based services software 112 executing on one or more remote computing devices (“cloud computing devices”) accessible over the Internet,” Jahjah et al., para [0025]. And, “When a client 120 desires to make changes to the infrastructure model, it may use the database system to preform primitive database operations, such as inserts, updates and deletes, on rows of tables of its local copy. The client 120 records these primitive database operations and eventually bundles them to create a local changeset 162. At this stage, the local changeset 162 represents pending changes to the infrastructure model, that are reflected locally on the client 120, but that have not yet been accepted to be shared with other clients. Subsequently, the client 120 may push the local changeset 162 back to infrastructure model hub services 130 to be added to the set of accepted changesets 160 in a repository 140-144 ,” Jahjah et al., para [0032]. Here, the changesets represent a transaction for an account (i.e., project) associated with a user.);
identifying the received transaction as miscategorized based on the comparison indicating
that the distance exceeds the threshold (“At step 950, the misclassification identification service 138 groups elements based on the unsupervised features and determines elements with that are far (in terms of statistical distance) from their group's center taking into account variance of elements in that group, and identifies these as misclassified elements,” Jahjah et al., para [0084]. See also, Jahjah et al., para [0011]. The examiner notes that “far from their group’s center” indicates that the distance exceed a threshold.); and
transmitting, to the user over the communications network, an indication that the received
transaction is miscategorized (“At step 960, the misclassification identification service 138 displays indications of misclassified elements of the infrastructure model in its user interface,” Jahjah et al., para [0086].).
Jahjah et al., though, does not disclose generating, by a first natural language processing (NLP) model, a first embedding of the received transaction based on text extracted from a first field of the received transaction; assigning the received transaction to a cluster of previous transactions based on a comparison of the first embedding of the received transaction with corresponding first embeddings of the previous transactions; generating, by a second NLP model, a second embedding of the received transaction based on text extracted from a second field of the received transaction; generating, by the second NLP model, second embeddings of the previous transactions based on text extracted from corresponding second fields of the previous transactions; determining a center of the cluster based on the second embeddings of the previous transactions; determining a distance between the second embedding of the received transaction and the center of the cluster; and comparing the distance to a threshold.
Kelly et al. is cited to disclose generating, by a first natural language processing (NLP) model, a first embedding of the received transaction based on text extracted from a first field of the received transaction (“As illustrated in FIG. 2, different characteristics of each page, such as natural language characters (e.g., sentences, page numbers) and symbols from each pace are extracted, tokenized, and converted into corresponding feature vectors that are embedded in the feature space via the document processing model(s) 208. For example, regarding the page 202, the symbol 202-1, the natural language text 202-2, and the page number 202-3 (and/or the X, Y coordinate positioning of this information (e.g., via a bounding box)) are encoded and aggregated (e.g., via a dot product), via the document processing model(s) 208, to the corresponding embedded feature vector 206A. The embedded feature vector 206A represents each detected feature of the corresponding page 202, except that the detected features are machine-readable (e.g., contains integers, as opposed to natural language characters). Such features may be indicated in terms of distance (e.g., Euclidian distance) or feature vector values, which indicate the relationships between features of a single document. In this way, each embedded feature vector represents intra-value relationships between each detected feature of each document. Likewise, the data (i.e., 204-1, 204-2, and 204-3) within pages 204 and data (i.e., 206-1, 206-2, and 206-3) of page 206 are likewise processed via the document processing model(s) 208 to derive the embedded features 204A (representing page 204) and 206A (representing page 206) respectively,” Kelly et al., col. 11, lines 21-46.);
assigning the received transaction to a cluster of previous transactions based on a
comparison of the first embedding of the received transaction with corresponding first
embeddings of the previous transactions (“FIG. 5 is a schematic diagram of an example visualization of feature space that illustrates various feature vectors representing one or more pages that have been clustered or classified, according to some embodiments,” Kelly et al., col. 1, lines 59-62.);
generating, by a second NLP model, a second embedding of the received transaction
based on text extracted from a second field of the received transaction (Kelly et al., col. 11, lines 21-46.);
generating, by the second NLP model, second embeddings of the previous transactions based on text extracted from corresponding second fields of the previous transactions (Kelly et al., col. 11, lines 21-46.);
determining a center of the cluster based on the second embeddings of the previous
transactions (Kelly et al., fig. 5, shows a circle around each cluster which indicates which data points belong to that cluster. The circle provides a radius/threshold from a center of the cluster. See also Kelly et al., col. 12, lines 39-54.);
determining a distance between the second embedding of the received transaction and the
center of the cluster (Kelly et al., fig. 5, shows a circle around each cluster which indicates which data points belong to that cluster. The circle provides a radius/threshold from the center of the cluster. See also Kelly et al., col. 12, lines 39-54.); and
comparing the distance to a threshold (Kelly et al., fig. 5, shows a circle around each cluster which indicates which data points belong to that cluster. The circle provides a radius/threshold from the center of the cluster. A data point is compared to this threshold to determine whether or not it belongs to that cluster. See also Kelly et al., col. 12, lines 39-54.). Kelly et al. benefits Jahjah et al. by extending the techniques of Jahjah et al. to document distinguishing. Therefore, it would be obvious for one skilled in the art to combine the teachings of Jahjah et al. and Kelly et al. to apply the identification of misclassified elements as taught by Jahjah et al. to document distinguishing.
As to claim 33, system claim 33 and method claim 22 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 33 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 23 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 22, wherein the first field of the received transaction includes a description of the received transaction, and the second field of the received transaction includes a vendor associated with the received transaction (Kelly et al., col. 19, lines 48-67. Here, 615A is a description of the transaction, and 605A is business (i.e., vendor).).
As to claim 34, system claim 34 and method claim 23 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 34 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 24 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 23, wherein text indicating the description of the received transaction and text indicating the vendor associated with the received transaction are received from the user over the communications network (“The page number 603 A can be extracted, as well as natural language text 605A indicating the entity (i.e., “ABC Plumbing & Heating”) responsible for creating the invoice or billing the invoice (e.g., a payer), and the natural language text 607A indicating the entity (i.e., “John Doe”) responsible for receiving and paying the bill associated with the invoice (e.g., a payee),” Kelly et al., col. 19, lines 48-54. See also figure 6A.).
As to claim 35, system claim 35 and method claim 24 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 35 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 25 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 22, wherein the first field of the received transaction includes a description of the transaction, and the second field of the received transaction includes a memo associated with the transaction (“The page number 603 A can be extracted, as well as natural language text 605A indicating the entity (i.e., “ABC Plumbing & Heating”) responsible for creating the invoice or billing the invoice (e.g., a payer), and the natural language text 607A indicating the entity (i.e., “John Doe”) responsible for receiving and paying the bill associated with the invoice (e.g., a payee). Other natural language characters 609A are also extracted, which are indicative of the due date, date, and the like of the invoice. The item line information or “description” 615A is also extracted. Further, the structure or format 613A (i.e., the positioning, thickness, color, and/or the like of the letterhead lines and mark) is extracted, as well as characteristics of the mark 611A itself. Further, the additional natural language note section 617A is also extracted. As described herein (e.g., with respect to FIG. 2), each of these characteristics or values can be represented as an embedded feature vector in vector space to represent the specific page 600 and used in downstream processes for sequence characteristic learning (e.g., via an LSTM), as described here,” Kelly et al., col. 19, lines 48-67. Here, 615A is a description, and 617A is a memo.).
As to claim 36, system claim 36 and method claim 25 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 36 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 26 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 22, wherein the first field of the received transaction includes a concatenation of a description of the received transaction and a memo associated with the received transaction (“In some embodiments, the document sequence learning model(s) 210 further concatenates or aggregates (e.g., via a dot product) both feature vectors 202A and 204A into another separate concatenated feature vector so that only the concatenated feature vector is compared to the feature vector 206A for distance determination between the concatenated feature vector and the feature vector 206A,” Kelly et al., p. 12, lines 55-61.).
As to claim 37, system claim 37 and method claim 26 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 36 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 27 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 22, wherein the previous transactions are associated with the user (“For instance, a customer may upload a series of different invoices needing to be processed. That customer's prior invoices, however, may have already been used for training in order to learn sequential characteristic and other patterns for those invoice pages. Accordingly, the prediction for that customer will be highly accurate,” Kelly et al., col. 4, lines 13-19. The invoices are transactions associated with the user.).
As to claim 38, system claim 38 and method claim 27 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 38 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 28 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 27, further comprising:
receiving, from the user over the communications network, a plurality of bank statements
associated with the previous transactions (The examiner notes that the choice of document type, such as (“bank statement”) to be received over the communications network, is a matter of design choice.).
As to claim 39, system claim 39 and method claim 28 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 39 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 29 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 22, wherein the previous transactions are associated with a different user (“Some embodiments use around 150,000 multi-page documents (i.e., the user-identified documents 315) uploaded to the inbox to train, validate and then test. In various embodiments, each document is separated or used as a whole by customers to generate invoices or bills, which is recorded in a data store as each page of a document with the associated page number of a bill, for example. As described above, in some embodiments, the records of the users' action in production are used to get the ground truth, where 1 represents the new start of a separation or document and 0 represents the continuation of the current separation or document. Various embodiments divide the dataset into 3 subsets for training, validation, and testing with 80%, 10%, and 10%, respectively,” Kelly et al., col. 14, lines 34-47.).
Regarding claim 30 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 22, wherein each of the first and second NLP models is a pretrained Bidirectional Encoder Representations from Transformers (BERT) model (Kelly et al., col. 12, lines 30-38.).
As to claim 40, system claim 40 and method claim 30 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 40 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 31 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 22, wherein the first NLP model is configured to generate the first embedding as a first vector of tokens based on text extracted from the first field of the received transaction, and the second NLP model is configured to generate the second embedding as a second vector of tokens based on text extracted from the second field of the received transaction (“As illustrated in FIG. 2, different characteristics of each page, such as natural language characters (e.g., sentences, page numbers) and symbols from each pace are extracted, tokenized, and converted into corresponding feature vectors that are embedded in the feature space via the document processing model(s) 208. For example, regarding the page 202, the symbol 202-1, the natural language text 202-2, and the page number 202-3 (and/or the X, Y coordinate positioning of this information (e.g., via a bounding box)) are encoded and aggregated (e.g., via a dot product), via the document processing model(s) 208, to the corresponding embedded feature vector 206A,” Kelly et al., col. 11, lines 21-32. And, “However, certain models, such as NLP-based models (e.g., BERT) can predict that the text features 204-2 is a continuation of the text 202-2 and/or that “page 2” is continuation of “page 1.” In this way, the document sequence learning model(s) 210 can represent or use NLP-based models,” Kelly et al., col. 12, lines 33-38. This second excerpt explains that multiple NLP models may be applied for extracting text.).
As to claim 41, system claim 41 and method claim 31 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 41 is similarly rejected under the same rationale as applied above with respect to method claim.
Regarding claim 32 (New), Jahjah et al., as modified by Kelly et al., discloses the method of claim 22, wherein:
the first NLP model is configured to selectively assign the received transaction to the
cluster of the previous transactions based on similarities between text extracted from the first
field of the received transaction and text extracted from the first fields of the previous
transactions (Kelly et al., fig. 5, shows a circle around each cluster which indicates which data points belong to that cluster. The circle provides a radius/threshold from the center of the cluster. See also Kelly et al., col. 12, lines 26-54.); and
the second NLP model is configured to selectively identify the received transaction as miscategorized based on dissimilarities between text extracted from the second field of the
received transaction and text extracted from the second fields of the previous transactions (Jahjah et al., para [0084]. See also, Jahjah et al., para [0011].).
As to claim 42, system claim 42 and method claim 32 are related as method and system of using same, with each claimed element’s function corresponding to the method step. Accordingly claim 42 is similarly rejected under the same rationale as applied above with respect to method claim.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANNE L THOMAS-HOMESCU whose telephone number is (571)272-0899. The examiner can normally be reached on Mon-Fri 8-6.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Bhavesh M Mehta can be reached on 5712727453. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see https://ppair-my.uspto.gov/pair/PrivatePair. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/ANNE L THOMAS-HOMESCU/Primary Examiner, Art Unit 2656