DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Priority
Receipt is acknowledged that application claims priority to foreign application with application number IN202421056497 dated 07/25/2024. Copies of certified papers required by 37 CFR 1.55 have been received. Priority is acknowledged under 35 USC 119(e) and 37 CFR 1.78.
Response to Amendments
The amendment filed 08/26/2026 has been entered. Applicant’s amendments to the drawings and specification have overcome each and every objection previously set forth in the Non-Final Office Action mailed 05/28/2026. Applicant’s amendments to the claims have overcome the 35 U.S.C. 112(b) rejections previously set forth in the Non-Final Office Action mailed 05/28/2026. Claims 1-20 remain pending in the application. Outstanding claim objections remain and are copied below.
The Examiner notes that (1) the Office Action date and claim summary at the beginning of the Remarks filed 08/26/2026 is erroneous and (2) the recitation of claim 1 on pg. 20-21 of the Remarks does not exactly match claim 1, as filed. Regardless, the Examiner is only considering the claims as filed in the Claims document filed 08/26/2026. Therefore, it is assumed that no claims have been cancelled and claims 2, 4, 9, 11, and 15-17 have been amended.
Response to Arguments
Regarding claim 1, the Applicant argues the following on pg. 21 of the Remarks (expanded on through pg. 22):
PNG
media_image1.png
180
737
media_image1.png
Greyscale
The Examiner has fully considered Applicant’s arguments with respect to independent claim 1, but they are not persuasive. Specifically, the Applicant distinguishes the relied upon prior art, Tata, with claim 1 using the following sequences:
PNG
media_image2.png
375
749
media_image2.png
Greyscale
The provided general processing sequence for Tata disregards the mapping of elements from Tata's disclosure to the broadest reasonable interpretation of the claims. It merely simplifies Tata’s nuanced disclosure with various non-limiting embodiments and compares the vocabulary used by Tata with that of the claims to argue that Tata's disclosure cannot read on the current claim set. Applicant summarizes the claimed ordered relationship as:
PNG
media_image3.png
286
739
media_image3.png
Greyscale
The Examiner submits that Tata’s disclosure teaches this ordered sequence. The mapping is set forth more fully below, but first, here is a general outline of how the steps summarized by the Applicant correspond to portions of Tata's disclosure:
Model selects context window based on parameters: Para 44: “neighborhood zone” which is based on distances relative to the candidate text portion
Elements within the context window: Para 45 describes neighbors of the candidate text portion
Feature vectors within the window: Para 109-110 describes generating feature vector data for neighbor portions
Separate model trained on feature vectors: Para 112 describes feature vectors input to a scoring model
Identify content of interest: Para 103 describes model output, further detailed in para 113-115
In summary, the steps of claim 1 are disclosed by Tata's method and Tata further specifies that elements of the method can be rearranged (see para 107). The mapping with respect to the two-model architecture is clarified below with respect to Tata's FIG. 7, which is an example according to the embodiments of the disclosure.
The Office Action (the Non-Final mailed 05/28/2026 and the rejection below) maps Tata's “scorer system,” described in para 45-46, to the context window identification model because para 45-46 describes software that governs the selection of the “neighborhood zone” (which is mapped to the context window). Para 45 describes that portions overlapping with this zone are considered neighbors and para 109-110 describes feature vector data describing the neighbors. These steps correlate to step 704-708 in FIG. 7. Para 108-109 describes that these steps may be performed by the document analysis system. The software governing these steps performed by the document analysis system can be considered a “model” (model 1).
As set forth in para 43 of Tata, the prediction scorer may be an isolated model that is trained/accessed separately from the document analysis system. The Office Action maps Tata's machine learning model to the field element identification model. This is the model referenced throughout the disclosure that performs the prediction step of associating text portions with fields. The prediction steps correlate to step 710-712 in FIG. 7. Para 112 describes that these steps can be performed by a machine-learned scoring model (model 2). Accordingly, the two-model architecture is established.
Tata explains multiple times throughout the disclosure that not all steps need be performed by one model or solely by the document analysis system. Para 107 further supports this, stating the following: “Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, and/or modified in various ways without deviating from the scope of the present disclosure.” Accordingly, even though the two-model architecture is established above, the method steps could also reasonably be split between two models.
To require a narrower ordered sequence, the Examiner suggests further detailing each step. For example, phrases such as “by applying the field element identification model” do not require that the position data is directly output by the model, but only that the model is part of the steps that lead the system to arrive at that position data. Further detailing the specific inputs, outputs, structure, training steps, and/or the interrelationship order between the models (as interpreted by the applicant) may help to establish the interpretation set forth in the Remarks.
Additionally, the Applicant argues that the mapping of Tata’s neighborhood zone to the claimed context window is inadequate because of the following (pg. 23 of the Remarks):
PNG
media_image4.png
329
741
media_image4.png
Greyscale
Examiner respectfully disagrees. The Applicant's use of a "window" surrounding a "field element" based on "parameters" is very broad in the context of identifying elements in a document. The neighborhood zone of Tata is a window surrounding a candidate text portion based on distances relative to the candidate text portion. Thus, the Applicant's current description of the selecting the context window in claim 1 requires further limiting details to overcome Tata.
Next, the Applicant argues on pg. 23-26 that the combination of Tata and Santos does not disclose the ordered relationship described earlier and utilizes isolated features rather than addressing the claimed combination as a whole. As described above, Tata teaches the ordered two-model architecture described by the Applicant. Accordingly, Santos modifies aspects within that method/architecture. In view of the foregoing, the rejection of claim 1 using Tata in view of Santos is maintained.
Regarding corresponding independent claims 8 and 15, and all dependent claims (see pg. 26-27 of the Remarks), the rejections of claims 2-20 are maintained based on the reasons set forth for claim 1 above.
Regarding the limitations specific to claims 2 and 9, Applicant argues on pg. 28-30 that the elements taught by Buhler, Ayyadevara, and Burch do not teach the ordered relationship set forth in claim 1. The Examiner disagrees for the reasons set forth above with regard to claim 1.
In response to Applicant's arguments directed towards the context window identification model and training, these arguments attack the references individually because Tata is relied on to teach the model and training. One cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). In each claim rejection, the statements describing the motivation to combine account for how the introduced teachings would apply to elements taught by Tata. The arguments provided ignore this information and fail to explain why the combination would not be feasible to one of ordinary skill in the art.
In response to Applicant's arguments regarding different purposes and intended uses in different references, each rejection describes the similarities between the references and the motivation to combine. If the prior art structure is capable of performing the intended use set forth by the claim, then it meets the claim. All references are directed towards extracting and/or organizing information from documents, and thus, a person or ordinary skill in the art would look to them based on the motivation to combine set forth in the rejections.
Regarding the limitations specific to claims 3, 10, and 16, Applicant argues the following on pg. 31:
PNG
media_image5.png
164
755
media_image5.png
Greyscale
The Examiner respectfully disagrees and sets forth that the details limiting which type of information may be included in the “plurality of parameters” is not claimed. Therefore, it is noted that the features upon which applicant relies are not recited in the rejected claims. Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993).
Additionally, the claim does not specify which parameters are used in the models. Claim 1 states that selecting the context window is “based on a plurality of parameters.” Therefore, the selection could use one type of parameter out of the plurality of parameters and still be “based on” the plurality of parameters. In view of the foregoing, the rejections are maintained.
Regarding the limitations specific to claims 7, 14, and 20, Applicant argues the following on pg. 32-33:
PNG
media_image6.png
295
749
media_image6.png
Greyscale
PNG
media_image7.png
479
756
media_image7.png
Greyscale
Examiner respectfully disagrees. Applicant argues that the steps taught by Kaza are “materially different” from the claimed elements; however, the elements that the Applicant uses to define the claim limitations are not required by the claim, and thus, cannot be materially different because they are not defined in the claim. For example, the “closed-loop adjustment” referenced by the Applicant is not recited in the claims. There is nothing recited in the claims that excludes user correction/retraining steps as being an adjustment loop, based on the plain language meaning of the claim limitations.
Applicant’s arguments with regard to the “weight factor” disregard the description of the combination of elements from Tata and Kaza provided in the claim rejection. One cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986).
Regarding the limitations specific to claims 4, 11, and 17, Applicant argues the following on pg. 33-34:
PNG
media_image8.png
337
741
media_image8.png
Greyscale
Examiner respectfully disagrees. The claim does not further define “generic and domain specific patterns” or “unique replacement element.” Tata teaches replacing text portion numbers, a generic pattern that may be identified in a document, with special representational segments, a replacement element. Tata teaches that the neighbor text portion encoding is part of the training process (see para 98). Therefore, the rejection of claims 4, 11, and 17 is maintained.
Claim Objections
Claims 2, 4, 9, 11, and 17 are objected to because of the following informalities:
In claims 2, 4, 9, 11, and 17: “and a combination thereof” should read “or a combination thereof”.
Appropriate correction is required.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are:
Input module in claim 1.
Training module in claims 1, 2, 4, 5.
Prediction module in claim 1.
Validation module in claims 6, 7.
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
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.
Claims 1, 5-6, 8, 12-13, 15, and 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Tata et al. (U.S. Patent No. 2021/0374395 A1), hereinafter Tata, in view of BASTOS DOS SANTOS et al. (WO Patent No. 2010/129330 A1), hereinafter Santos.
Regarding claim 1, Tata teaches a system to identify a content of interest from a corpus of documents (Tata, para 22: “the document analysis system can determine one or more field types that are expected to be found in the document. The document analysis system can, for each field type, determine one or more candidate text portions from the plurality of text portions extracted from the document”), the system comprising:
an input module (Tata, computer program code of the system that executes the following input tasks, see para 58-62) configured to: obtain the corpus of documents, wherein each document contains the content of interest (Tata, received documents in para 35 containing the text portions in para 22 above); and receive a plurality of training documents (Tata, para 106: “training data”), wherein each training document includes content tagged into a plurality of field elements, wherein one field element is the content of interest (Tata, field types in para 67; particular field type in para 68 is the content of interest, referred to as the candidate text portion during calculations);
a training module coupled to the input module (Tata, computer program code of the system that executes the following training tasks, see para 58-62) and configured to: select a context window surrounding a field element in a training document based on a plurality of parameters defined for the field element (Tata, para 45: “neighborhood zone of a candidate text portion”; based on distance to bounding boxes, position, and height relative to the candidate text portion, or field element, described further in para 88-89), and by applying a context window identification model to the training document (Tata, the score system, which defines the neighborhood zone, para 45-46);
estimate for each data element present within the context window of the training document a distance of the each data element from the field element (Tata, para 46: “The scorer system can calculate the relative position of a neighbor text portion as the difference between its normalized two-dimensional coordinates and those of the candidate text portion”), wherein the distance and a position of the each data element with respect to the field element, is used to create a feature vector (Tata, relative location and location of neighbor text portions; para 109-110: “generate a respective input feature vector for each candidate text portion for the field type, wherein the respective input feature vector for each candidate text portion comprises data describing a respective position of one or more neighbor text portions that are proximate to the candidate text portion”; location/position of text portions, see para 66 and 79, are used to identify each neighboring text portion); and
provide a set of feature vectors developed for the plurality of field elements across the plurality of training documents as an input, to train a field element identification model (Tata, ML model learns an embedding based on input feature vector data, see para 55 and para 18: “the proposed systems and methods can include or leverage a machine learning model (e.g., neural network) that learns a dense representation for an extraction candidate based on the tokens in its neighborhood and their relative location”); and
a prediction module coupled to the training module (Tata, computer program code of the system that executes the following prediction tasks, see para 58-62) and configured to: apply the context window identification model to the each document to identify one or more candidate context windows that contain the content of interest (Tata, score system/model is applied to each document, see description with reference to FIG. 2-3 in para 85-90); and
identify a beginning position of the content of interest within a candidate context window by applying the field element identification model on the one or more candidate context windows, wherein the beginning position is used to retrieve the content of interest from the each document (Tata, para 103: “model 602 provides output data 610 that describes which text portions are associated with each field in the target schema”; all candidate text portions have an associated position, a particular location in para 37, thus identifying a particular text portion as the field of interest also identifies the beginning position to retrieve the content of the text portion).
Tata teaches weighting the neighbor encodings (See para 27), but fails to explicitly teach wherein the distance is a weighted distance. However, Santos teaches a similar method (Santos, abstract: “Systems and methods analyze the physical structure of text rows in a document image 21800, including the positions of one or more alignments of one or more character blocks 132 in one or more text rows of the document image 21800”). In the method of Santos, a weighted distance (Santos, para 547: “the forms processing system 104A determines the standard row distance for the left alignments and determines the weighted row distance for the right alignments”) is utilized to apply different weights to different attributes describing the positions of field elements in a document (Santos, left versus right alignment, para 550: “The row distance of row 1 from the master row is d1=cd(r1, MR)= 10, which includes a standard row distance of 6 for the left alignments and a weighted row distance of 4 for the right alignments”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the weighted distance of Santos with the system of Tata in order to generate a more accurate score by combining multiple attributes (For example, x and y distance values in FIG. 5 of Tata) using weight values that reflect the importance of each attribute (Santos, para 547: “more weight is placed on the left alignments than the right alignments. This may be used, for example, where the left alignments are more important or may provide a better determination of the total classification of text rows into classes”).
Regarding claim 5 (dependent on claim 1), Tata in view of Santos teaches wherein the weighted distance of the each data element from the content of interest is computed by determining a distance and direction along a horizontal and a vertical axis of the each data element from the content of interest and applying a weight factor associated with each direction (Tata teaches relative horizontal and vertical axis positions – see FIG. 5 and para 46: “The scorer system can calculate the relative position of a neighbor text portion as the difference between its normalized two-dimensional coordinates and those of the candidate text portion. The relative positions for the padding neighbors can be set to (1.0, 1.0)”, (1.0,1.0) is an example denoting a distance along each axis, which also provides a direction; Santos teaches applying weights to attributes describing fields in a document, see claim 1 rejection).
Regarding claim 6 (dependent on claim 5), Tata in view of Santos teaches further comprising a validation module coupled to the training module (Tata, computer program code of the system that executes the following validation tasks, see para 58-62), wherein the validation module is configured to: validate the retrieved content of interest based on the plurality of parameters defined for the field element (Tata, para 99: “The score model 216 can be trained using binary cross-entropy between this prediction and the target label as the loss function”; para 106: “a loss function can be backpropagated through the model(s)”).
While Tata teaches retrieving a content of interest, Tata fails to explicitly teach provide the content of interest on a user interface. However, Santos teaches providing the output of the system on a user interface (Santos, para 196: “the extracted data is generated for display to one or more displays, such as to a user interface 114. The user interface 114 may be generated by the forms processing system 104 or an output system. The user interface 114 displays the extracted data and/or other data, including an image of the analyzed document, document model data, document model images, and/or other documents, images, and/or other data”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the user interface of Santos with the system and content of interest of Tata in view of Santos in order to convey the results of the system to a user (Santos, para 195: “The user interface 114 displays the data and enables a user to enter data and/or instructions”).
Regarding claim 8, Tata in view of Santos teaches a computer-implemented method (Tata, para 58-62). All claim limitations are met and rendered obvious by Tata in view of Santos because the method steps of claim 8 are the same as those in claim 1.
Regarding claim 12 (dependent on claim 8), all claim limitations are met and rendered obvious by Tata in view of Santos because the method steps of claim 12 are the same as those in claim 5.
Regarding claim 13 (dependent on claim 12), Tata in view of Santos teaches further comprises: validating the identified content of interest based on the plurality of parameters defined for the field element (Tata, para 99: “The score model 216 can be trained using binary cross-entropy between this prediction and the target label as the loss function”; para 106: “a loss function can be backpropagated through the model(s)”).
Regarding claim 15, Tata teaches a computer-implemented method (Tata, para 58-62) for identifying a content of interest in a document (Tata, para 22: “the document analysis system can determine one or more field types that are expected to be found in the document. The document analysis system can, for each field type, determine one or more candidate text portions from the plurality of text portions extracted from the document”), the method comprising:
obtaining the document containing the content of interest (Tata, received documents in para 35 containing the text portions in para 22 above);
receiving a plurality of training documents (Tata, para 106: “training data”), wherein each training document includes one or more textual and image content (Tata, each document includes image content, para 21; textual content described in next citation) that comprises a plurality of field elements, wherein one field element corresponds to the content of interest (Tata, textual field types in para 67; particular field type in para 68 is the content of interest, referred to as the candidate text portion during calculations);
defining a plurality of parameters for the field element (Tata, distance to bounding boxes, position, and height relative to the candidate text portion, or field element, described further in para 88-89);
estimating for each data element present within the training document a distance of the each data element from the field element (Tata, para 46 : “The scorer system can calculate the relative position of a neighbor text portion as the difference between its normalized two-dimensional coordinates and those of the candidate text portion”; see FIG. 5 where positions and relative distances are estimated for each data element – invoice date and number), wherein the distance and a position of the each data element with respect to the field element is used to create a feature vector (Tata, relative location and location of neighbor text portions; para 109-110: “generate a respective input feature vector for each candidate text portion for the field type, wherein the respective input feature vector for each candidate text portion comprises data describing a respective position of one or more neighbor text portions that are proximate to the candidate text portion”; location/position of text portions, see para 66 and 79, are used to identify the neighboring text portion);
providing a set of feature vectors developed for the plurality of field elements across the plurality of training documents as an input, in training a field element identification model (Tata, ML model learns an embedding based on input feature vector data, see para 55 and para 18: “the proposed systems and methods can include or leverage a machine learning model (e.g., neural network) that learns a dense representation for an extraction candidate based on the tokens in its neighborhood and their relative location”); and
identifying a beginning position of the content of interest by applying the field element identification model on the document (Tata, para 103: “model 602 provides output data 610 that describes which text portions are associated with each field in the target schema”; all candidate text portions have an associated position, particular location in para 37, thus identifying a particular text portion as the field of interest also identifies the beginning position to retrieve the content of the text portion).
Tata teaches weighting the neighbor encodings (See para 27), but fails to explicitly teach wherein the distance is a weighted distance. However, Santos teaches a similar method (Santos, abstract: “Systems and methods analyze the physical structure of text rows in a document image 21800, including the positions of one or more alignments of one or more character blocks 132 in one or more text rows of the document image 21800”). In the method of Santos, a weighted distance (Santos, para 547: “the forms processing system 104A determines the standard row distance for the left alignments and determines the weighted row distance for the right alignments”) is utilized to apply different weights to different attributes describing the positions of field elements in a document (Santos, left versus right alignment, para 550: “The row distance of row 1 from the master row is d1=cd(r1, MR)= 10, which includes a standard row distance of 6 for the left alignments and a weighted row distance of 4 for the right alignments”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the weighted distance of Santos with the system of Tata in order to generate a more accurate score by combining multiple attributes (For example, x and y distance values in FIG. 5 of Tata) using weight values that reflect the importance of each attribute (Santos, para 547: “more weight is placed on the left alignments than the right alignments. This may be used, for example, where the left alignments are more important or may provide a better determination of the total classification of text rows into classes”).
Regarding claim 18 (dependent on claim 15), Tata in view of Santos teaches wherein the weighted distance of each data element from the content of interest is computed by determining a distance and direction along a horizontal and vertical axis of the each data element from the content of interest and applying a weight factor that could be a linear, polynomial or exponential, function associated with each direction (Tata teaches relative horizontal and vertical axis positions – see FIG. 5 and para 46: “The scorer system can calculate the relative position of a neighbor text portion as the difference between its normalized two-dimensional coordinates and those of the candidate text portion. The relative positions for the padding neighbors can be set to (1.0, 1.0)”, (1.0,1.0) is an example denoting a distance and direction; Santos teaches applying weights to attributes describing fields in a document, see claim 1 rejection; Weight factors are linear, for example: wd in para 543, standard distance would be weighted by 1).
Regarding claim 19 (dependent on claim 18), all claim limitations are met and rendered obvious by Tata in view of Santos because the method steps of claim 19 are the same as those in claim 13.
Claims 2 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Tata in view of Santos, in further view of Buhler et al. (U.S. Patent No. 10,769,503 B1), hereinafter Buhler, Ayyadevara et al. (U.S. Patent No. 2021/0182548 A1), hereinafter Ayyadevara, and Burch et al. (U.S. Patent No. 10,127,444 B1), hereinafter Burch.
Regarding claim 2 (dependent on claim 1), Tata in view of Santos teaches wherein the training module is further configured to: scan through each page of the training document (Tata, documents are ingested, see para 34-35, and then further analyzed – steps can be repeated for each page/image; para 103: “The set of input data can include images of one or more documents”) for determining one or more zones in which the each data element occurs within the training document (Tata, text portions, or data elements, occur in neighboring zones, see claim 1 rejection and para 46 and 109), wherein each page of the training document is sectioned into a plurality of zones (Tata, document analysis steps are applied to each document image; see context window zone sectioning in claim 1, there are a plurality of neighborhood zones – one around each candidate text portion analyzed);
identify one or more generic and domain specific patterns in the training document; and replace each of the one or more generic and domain specific patterns with a unique replacement element (Tata, special representational segment for all numbers, para 89: “the score model 216 can encode the neighbor text portions using a vocabulary. The vocabulary can include a special representational segment or token for out-of-vocabulary words and a special representational segment or token for all numbers”); and
provide the feature matrix as an input to train the context window identification model (Tata, scoring system/model is trained as part of the document analysis system, see para 43 and para 87: “The score model 216 can be included as a stage in a larger machine-learned model or an independent machine-learned model that is accessed by the document analysis system (e.g., document analysis system 110 in FIG. 1)”), wherein the context window identification model is used to select the context window for the field element (Tata, described in claim 1 rejection).
Tata in view of Santos fails to explicitly teach: determining a frequency of occurrence of each data element; eliminate one or more predefined data elements from the training document; wherein a predefined data element is one of a pronoun, a preposition, a conjunction, a data element identified as least relevant in retrieval of the content of interest or a combination thereof; eliminate one or more data elements having a frequency of occurrence lesser than a threshold value from the training document; and develop a feature matrix comprising a frequency of occurrence of each remaining data element in each zone of the training document.
Buhler similarly teaches a document search method (Buhler, abstract), including eliminate one or more predefined data elements from the training document, wherein a predefined data element is one of a pronoun, a preposition, a conjunction, a data element identified as least relevant in retrieval of the content of interest or a combination thereof (Buhler, col 28, ln 41-43: “Some implementations begin by eliminating common words (sometimes called “stop words”), including prepositions and basic identifiers”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined eliminating certain words, as taught by Buhler, with the system of Tata in view of Santos in order to improve the model’s performance by allowing training to focus on higher value words with semantic meaning that may apply to the field of interest (Buhler, col 28, ln 31-52).
Ayyadevara similarly teaches a document search method (Ayyadevara, abstract). Ayyadevara’s system similarly considers data elements in a neighboring zone of a candidate text portion (Ayyadevara, para 108: “identify FOI neighbors (e.g., neighboring fields) relative to particular FOIs (e.g., FOIs corresponding to a particular ROI) based at least in part on the determined distances between FOIs”). Ayyadevara further teaches determining a frequency of occurrence of each data element (Ayyadevara, para 110: “FIG. 18 provides additional data indicative of the identification of horizontal and vertical neighbors of an FOI, based on an analysis of a plurality of document images. As indicated, “Strength” occurs 30 times (likely corresponding to 30 analyzed document images), and has a small average vertical distance (i.e., 5 pixels)…”; see FIG. 18 attached below) and develop a feature matrix comprising a frequency of occurrence of each data element in each zone of the training document (Ayyadevara, see table in FIG. 18 – neighborhood is analyzed for each examined FOI, thus there is a feature matrix for each zone). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the frequency analysis taught by Ayyadevara with the system of Tata in view of Santos and Buhler in order to improve the performance of the model by providing additional data indicative of identifying neighbors for a field of interest element (Ayyadevara, para 110 citation above).
PNG
media_image9.png
273
764
media_image9.png
Greyscale
Lastly, Burch similarly teaches a document processing method of codifying document information to be used for identifying and extracting information (Burch, abstract). Burch’s method includes eliminate one or more data elements having a frequency of occurrence lesser than a threshold value from the training document (Burch, col 17, ln 6-14: “At 708, the set of canonical feature codifications generated at 704 is filtered to remove codifications with a data type that does not meet a threshold frequency. Various threshold frequencies may be applied, but in in certain embodiments a value of 10% is appropriate. For example, any canonical feature codification that occurs less than 10% of the time in the comparison set of reference codifications is deleted/not considered in further processing of the canonical feature”). Burch’s method compares field types, represented as canonical features, to a reference to learn the structure of the field types that make up the document. It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the threshold of Burch with the system of Tata in view of Santos, Buhler, and Ayyadevara in order to improve the model’s learning of field type spatial relationships by removing low frequency fields from the training data (Burch, col 16, ln 27-35: “If a reference codification of the canonical feature matches a working document text rectangle, that fact is recorded, and over process 700 multiple canonical features (either the same or different) may be recorded against a given text rectangle. Once all canonical features and all reference codifications of those features have been processed, the text rectangles of the working document are codified based on the most frequently assigned canonical features”). After removing certain elements, as taught by Buhler and Burch, the combination of Tata in view of Santos, Buhler, Ayyadevara, and Burch teaches: develop a feature matrix comprising a frequency of occurrence (Taught in combination with Ayyadevara) of each remaining data element in each zone of the training document.
Regarding claim 9 (dependent on claim 8), all claim limitations are met and rendered obvious by Tata in view of Santos, Buhler, Ayyadevara, and Burch because the method steps of claim 9 are the same as those in claim 2.
Claims 3, 7, 10, 14, 16, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Tata in view of Santos, in further view of Kaza et al. (U.S. Patent No. 2022/0035990 A1), hereinafter Kaza.
Regarding claim 3 (dependent on claim 1), Tata in view of Santos teaches wherein the plurality of parameters comprises text alignment of the field element (Tata, zone is determined based on overlap, a type of alignment, of the candidate text portion with other text portion bounding boxes, see para 45), text spacing within the field element (Tata, text spacing between the candidate text portion and other text portions dictates overlap and neighbor padding, see para 45), location parameters (Tata, zone is determined based on position of the candidate text portion, see para 46) and context window parameters defined for the each field element (Tata, certain rules dictate what is considered overlap and padding, para 45: “overlap by more than half with the neighborhood zone of a candidate text portion can be considered to be a neighbor of the candidate text portion… the list can be padded to ensure that there are 20 neighbor text portions, with the padded candidate text portions being represented as a pad token”). Tata in view of Santos fails to explicitly teach wherein the plurality of parameters comprises parameters defined for a data type associated with each field element and fonts of the field element.
However, Kaza similarly teaches a document search method (Kaza, abstract). Kaza identifies specific areas in a document using a plurality of parameters comprising parameters defined for a data type associated with each field element (Kaza, visual indicators for identifying fields, para 27: “The tagged field 210 is associated with a location within the document corresponding to a field and identifies a field type of the field. A tagged field 210 can include visual indicators, such as…”) and fonts of the field element (Kaza, para 27: “a change of font size”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the plurality of parameters taught by Kaza with the plurality of parameters taught by Tata in view of Santos in order to further aid the model in identifying a specific type of field area in a document (Kaza, para 31: “The identification of field locations for tagging by the machine learned model 160 may be performed more accurately and more efficiently than conventional document tagging methods described above”).
Regarding claim 7 (dependent on claim 6), Tata in view of Santos teaches wherein the validation module is further configured to: determine a difference between the content of interest and the retrieved content of interest, when the retrieved content of interest fails to validate (Tata, para 99: “The score model 216 can be trained using binary cross-entropy between this prediction and the target label as the loss function”); but fails to explicitly teach: adjusts the plurality of parameters, and the weight factor associated with the each direction based on the difference determined between the content of interest and the retrieved content of interest; and retrain the context window identification model and the field element identification model on the document with the adjusted plurality of parameters and the weight factor associated with the each direction.
However, Kaza similarly teaches a document search method (Kaza, abstract), including identifying specific areas in a document using a plurality of parameters associated with the field element (Kaza, visual indicators for identifying fields, para 27: “The tagged field 210 is associated with a location within the document corresponding to a field and identifies a field type of the field. A tagged field 210 can include visual indicators, such as…”). Kaza teaches adjusting the plurality of parameters based on the difference determined between the content of interest and the retrieved content of interest; and retrain the trained model on the document with the adjusted plurality of parameters (Kaza, embodiments where the automatic tag is incorrect, para 37: “the user may delete one or more of the tags 260 (for instance, in embodiments where the automatic tag is incorrect, or in embodiments where the user does not want the tag to be present) and/or update one or more of the tags 260 (for instance, in embodiments where the user wants to move a location of the tag, change a field type of the tag, and the like). The auto-tagging engine 140 can subsequently re-train the machine learned model 160 based on the user feedback”; adjustments are made based on a field type tag being incorrect, and are thus based on a difference between what was expected and what was retrieved; changing the field type and location adjusts the plurality of parameters, see next paragraph).
Kaza’s plurality of parameters (Kaza, para 27: “A tagged field 210 can include visual indicators, such as a box surrounding the field, a circle surrounding the field, a highlight applied to the field, a text box located adjacent to the field, a change of font size, color, or emphasis of the field, or some combination thereof”) are utilized to identify a specific area in the document, similar to the weight factor in the field element identification model, which can be considered another parameter utilized to identify a field element in a document. Kaza teaches a method of adjusting a group of parameters (The location and field type causes adjustments to the parameters listed in para 27) in response to a failed validation (Incorrect result). Therefore, in combination with Kaza and the weight factors and models of Tata in view of Santos, Tata in view of Santos and Kaza also teaches adjusting the plurality of parameters, and the weight factor associated with the each direction based on the difference determined between the content of interest and the retrieved content of interest; and retrain the context window identification model and the field element identification model on the document with the adjusted plurality of parameters and the weight factor associated with the each direction (emphasis added).
It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the adjustments and retraining, taught by Kaza, with the system of Tata in view of Santos in order to improve each model’s performance by correcting model input parameters when there are incorrect results (See para 37 citations from Kaza above).
Regarding claim 10 (dependent on claim 8), all claim limitations are met and rendered obvious by Tata in view of Santos and Kaza because the method steps of claim 10 are the same as those in claim 3.
Regarding claim 14 (dependent on claim 13), all claim limitations are met and rendered obvious by Tata in view of Santos and Kaza because the method steps of claim 14 are the same as those in claim 7.
Regarding claim 16 (dependent on claim 15), all claim limitations are met and rendered obvious by Tata in view of Santos and Kaza because the limitations of claim 16 are all included in claim 3.
Regarding claim 20 (dependent on claim 19), Tata in view of Santos teaches further comprising: determining a difference between the content of interest and the identified content of interest, when the identified content of interest fails to validate (Tata, para 99: “The score model 216 can be trained using binary cross-entropy between this prediction and the target label as the loss function”); and one or more weights associated to a direction along an axis, wherein the weights are used for estimating a weighted distance of a data element around the content of interest (Taught by the weight factors in claim 18); but fails to explicitly teach: redefining the plurality of parameters, and one or more weights associated to a direction along an axis; and retraining the context window identification model and the field element identification model on the document with the redefined plurality of parameters and the one or more weights.
However, Kaza similarly teaches a document search method (Kaza, abstract), including identifying specific areas in a document using a plurality of parameters associated with the field element (Kaza, visual indicators for identifying fields, para 27: “The tagged field 210 is associated with a location within the document corresponding to a field and identifies a field type of the field. A tagged field 210 can include visual indicators, such as…”). Kaza teaches redefining the plurality of parameters; and retraining the trained model on the document with the redefined plurality of parameters (Kaza, embodiments where the automatic tag is incorrect, para 37: “the user may delete one or more of the tags 260 (for instance, in embodiments where the automatic tag is incorrect, or in embodiments where the user does not want the tag to be present) and/or update one or more of the tags 260 (for instance, in embodiments where the user wants to move a location of the tag, change a field type of the tag, and the like). The auto-tagging engine 140 can subsequently re-train the machine learned model 160 based on the user feedback”; adjustments are made based on a field type tag being incorrect, and are thus based on a difference between what was expected and what was retrieved; changing the field type and location adjusts the plurality of parameters, see next paragraph).
Kaza’s plurality of parameters (Kaza, para 27: “A tagged field 210 can include visual indicators, such as a box surrounding the field, a circle surrounding the field, a highlight applied to the field, a text box located adjacent to the field, a change of font size, color, or emphasis of the field, or some combination thereof”) are utilized to identify a specific area in the document, similar to the weight factor in the field element identification model, which can be considered another parameter utilized to identify a field element in a document. Kaza teaches a method of adjusting a group of parameters (The location and field type causes adjustments to the parameters listed in para 27) in response to a failed validation (Incorrect result). Therefore, in combination with Kaza and the weight factors and models of Tata in view of Santos, Tata in view of Santos and Kaza also teaches redefining the plurality of parameters, and one or more weights associated to a direction along an axis; and retraining the context window identification model and the field element identification model on the document with the redefined plurality of parameters and the one or more weights (emphasis added).
It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the adjustments and retraining, taught by Kaza, with the model of Tata in view of Santos in order to improve each model’s performance by correcting model input parameters when there are incorrect results (See para 37 citations from Kaza above).
Claims 4, 11, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Tata in view of Santos, in further view of Buhler.
Regarding claim 4 (dependent on claim 1), Tata in view of Santos teaches wherein the training module is further configured to: identify one or more generic and domain specific patterns in the training document; replace each of the one or more generic and domain specific patterns with a unique replacement element (Tata, special representational segment for all numbers, para 89: “the score model 216 can encode the neighbor text portions using a vocabulary. The vocabulary can include a special representational segment or token for out-of-vocabulary words and a special representational segment or token for all numbers”), but fails to explicitly teach eliminate a predefined element from the training document, wherein a predefined element is one or more of a pronoun, a preposition, a conjunction, a data element identified as least relevant in retrieval of the content of interest or a combination thereof.
However, Buhler similarly teaches a document search method (Buhler, abstract), including eliminate a predefined element from the training document, wherein a predefined element is one or more of a pronoun, a preposition, a conjunction, a data element identified as least relevant in retrieval of the content of interest or a combination thereof (Buhler, col 28, ln 41-43: “Some implementations begin by eliminating common words (sometimes called “stop words”), including prepositions and basic identifiers”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined eliminating certain words, as taught by Buhler, with the system of Tata in view of Santos in order to improve the model’s performance by allowing training to focus on higher value words with semantic meaning that may apply to the field of interest (Buhler, col 28, ln 31-52).
Regarding claim 11 (dependent on claim 8), all claim limitations are met and rendered obvious by Tata in view of Santos and Buhler because the method steps of claim 11 are the same as those in claim 4.
Regarding claim 17 (dependent on claim 15), all claim limitations are met and rendered obvious by Tata in view of Santos and Buhler because the method steps of claim 17 are the same as those in claim 4.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EMMA E DRYDEN whose telephone number is (571)272-1179. The examiner can normally be reached M-F 8-4 EST.
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, ANDREW BEE can be reached at (571) 270-5183. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/EMMA E DRYDEN/Examiner, Art Unit 2677
/ANDREW W BEE/Supervisory Patent Examiner, Art Unit 2677