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 .
Response to Amendment
The office action is responsive to the amendment filed on 04/08/2026. As directed by applicant remark claims 1, 4, 7-8, 10, 14, 16, and 18-20 are amended, claims 5 and 9 are canceled, and claim 22-23 is new. Claims 1-4, 6-8, and 10-23 are pending for examination.
Response to Arguments
Regarding the 35 U.S.C § 103 Rejection:
Applicant’s arguments with respect to claims 1-4, 6-8, and 10-23 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.
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 1-4, 6-8, 11-12, 14-20, and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Erdemir et al. US 2021/0240932 A1 (hereinafter Erdemir) in view of Florencio et al. US 2021/0133438 A1 (hereinafter Florencio) in further view of De Peuter US 2023/0123711 A1 (hereinafter Peuter) in further view of Liang et al. US 2021/0286821 A1 (hereinafter Liang) and in further view of Gao et al. US 2022/0215195 A1 (hereinafter Gao).
Regarding claim 1:
Erdemir teaches the following:
A computer-implemented method comprising: (Erdemir [0003] teaches a method).
extracting, by one or more processors, using one or more optical character recognition operations, and based on a pre-defined list of entities, a group of entity data tokens comprising a plurality of label data tokens and a plurality of value data tokens from image data within an input data record; ( Erdemir [0046] teaches optical character recognition (OCR) techniques may be used to identify words or contiguous sets of characters within a received electronic document (i.e., input data record) and Erdemir [0026] teaches extracting and grouping content from an electronic document and teaches the content can “include any textual, numerical, symbolic characters that may correspond to, for example, labels and values”. To add, Erdemir [0038] teaches a grouping engine that groups phrases together into a single group, such grouping is based on “listing of known labels” (i.e., pre-defined list of entities) as described in paragraph [0048]. Furthermore, [0037] teaches the phrases corresponds to a set of tokens that is intended to describe a single element (i.e. group of entity data tokens) that include label and value data tokens. Moreover, Erdemir [0033] & [0091] teaches the document being analyzed could had been received in paper format and therefore would have been scanned or photographed to result in the electronic image document that is used for analysis (see Fig. 5 element 530). Lastly, Erdemir Fig. 8 element 804 teaches a processor to perform the process steps).
generating, by the one or more processors, a spatial coordinate set, corresponding to a value data token from the plurality of value data tokens, within a spatial coordinate scheme that describes a set of dimensions and bounds of the input data record, wherein the spatial coordinate set describes a spatial position of the value data token within the spatial coordinate scheme
inputting, by the one or more processors, the plurality of label data tokens and the plurality of value data tokens to a vector classification machine learning model to generate a label-value pair comprising a label data token from the plurality of label data tokens and the value data token from the plurality of value data tokens based on the spatial coordinate set, by: (Erdemir [0042] teaches using an algorithm (vector classification machine learning model) to determine one or more parameters ( value data token). Specifically, Erdemir [0065] teaches generating label-value pairs and [0098] teaches the label-value pairs is a label (label data token) and a value (value data token) with the indication that they are associated with each other).
generating a coordinate vector, for the label data token, positioned in relation to the value data token based at least in part on the spatial coordinate set, and
PNG
media_image1.png
370
774
media_image1.png
Greyscale
(Erdemir Fig. 5 and [0097] teaches values are typically associated with label, and [0098] teaches a label value pair is a label and a value with the indication that they are associated with each other, this association can be based on the horizontally or vertical relationship. Such that [0099] and Fig. 6 represent an example of two tokens with associated bounding box, for which a variety of measurements can be made to determine how close together the tokens are. Moreover, according to Erdemir, a measurement that can be used to generate coordinate vectors is the Cartesian distance also known as a Euclidean distance. Specifically, [0102] teaches the “Cartesian distance 650 can be calculated using the horizontal separation 630 and the vertical separation 640... For purposes of calculating the Cartesian distance, the closest conners of the respective bounding boxes are generally used”. Further, Fig. 6 above, teaches applying “Cartesian Distance” element 650 between two tokens element 610 and 620).
inputting the coordinate vector, the label data token, and the value data token to the vector classification machine learning model to generate the label-value pair, wherein (i) the vector classification machine learning model is trained using a plurality of ground- truth label-value pairs corresponding to a plurality of primary historical data records of a historical dataset based at least in part using a label-value pair regular expression and (ii) the label-value pair is generated based at least in part on an output of the vector classification machine learning model (Erdemir [0042] teaches using an algorithm (vector classification machine learning model) to determine one or more parameters ( value data token). Specifically, [0050] teaches “a value (value data token) is selected based on training using an appropriate sample set. The training dictates the standard position (coordinate vectors) of the numerical value in relation to the currency symbol or currency abbreviation”. Further, Erdemir [0065] teaches generating label-value pairs and [0098] teaches the label-value pairs is a label (label data token) and a value (value data token) with the indication that they are associated with each other. Moreover, Erdemir [0065] teaches generating label-value pair based on an algorithm (i.e., vector classification machine learning model) that return results (i.e., output) that matches the best know characteristics (see [0042])).
Erdemir does not teach or suggest wherein the spatial coordinate set describes a spatial position of the value data token within the spatial coordinate scheme including a center point of a rendered form of the value data token; the vector classification machine learning model is trained using a plurality of ground-truth label-value pairs corresponding to a plurality of primary historical data records of a historical dataset based at least in part on an automatic annotation of the historical dataset using a label-value pair regular expression; generating the label-value pair specifically based at least in part on a pairing score assigned to the value data token based at least in part on an output of the vector classification machine learning model; storing, by the one or more processors, the label-value pair as a representation of the input data record to reduce an amount of storage resources used to store the input data record, wherein the representation uses less storage resources than the image data and inputting, by one or more processors, the label-value pair to a record classification machine learning model to receive a classification for the input data record.
However, Florencio teaches the following:
...wherein the vector classification machine learning model is trained using a plurality of ground-truth label-value pairs corresponding to a plurality of primary historical data records of a historical dataset based at least in part on an automatic annotation of the historical dataset using a label-value pair regular expression, and ( Florencio [0038] teaches a training model is generated (trained) “based on a ground truth that includes a plurality of key-value pairs corresponding to the plurality of forms”. Further, Florencio [0103] teaches the system automatically identifies a set of forms ( historical dataset of data) to use for identifying ground truth of key value pairs and [0080] teaches a “feature generation block 456 may receive the modified forms and generate feature rules based on the type of form and also generate features and labels from the form”. Further, [0083] teaches “feature generation block 456 uses a matching regular expression and includes the matching regular expression as a feature”).
Florencio is also in the same field of endeavor as Erdemir (data processing/analysis). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of model generation as being disclosed and taught by Florencio the system taught by Erdemir to yield the predictable results of “improve techniques and systems for processing forms and the data contained within the forms, as well as for interfaces the improve a user's control and ability for facilitating the training and tuning of these systems” ( Florencio [0008]).
While Florencio teaches labelling forms ( historical dataset of data) to use for identifying ground truth, and key-value pairs being stored as metadata in paragraph [0069], Florencio does not suggest wherein the spatial coordinate set describes a spatial position of the value data token within the spatial coordinate scheme including a center point of a rendered form of the value data token; (ii) the label-value pair is generated based at least in part on a pairing score assigned to the value data token based at least in part on an output of the vector classification machine learning model; storing, by the one or more processors, the label-value pair as a representation of the input data record to reduce an amount of storage resources used to store the input data record, wherein the representation uses less storage resources than the image data; an inputting, by one or more processors, the label-value pair to a record classification machine learning model to receive a classification for the input data record.
Nevertheless, Peuter teaches the following:
storing, by the one or more processors, the label-value pair as a representation of the input data record to reduce an amount of storage resources used to store the input data record, wherein the representation uses less storage resources than the image data and (Peuter [0080] teaches tokens include a plurality of tokens that each has positional coordinates and can form key value pairs within a document. In addition, [0030] & [0041] teaches such key value pairs can be stores in a database for later retrieval. Someone skilled in the relevant art will recognize, “label-value pair” is a fundamental data representation which contains a name or “attribute” for the data and an associated “value” for the data, thus the label-value pair inheritably represent input data record. In addition, Peuter [0027] teaches how using positional coordinate (i.e., tokens/ key-value pairs) the embodiments described therein can save both “time and power resources”. As would be familiar to one skilled in the relevant art, saving “time and power resources” in machine learning oftentimes directly reduces storage need by extracting specific relevant data (e.g., positional coordinate) and therefore only focusing on the data that provides optimal optimization).
inputting, by one or more processors, the label-value pair to a record classification machine learning model to receive a classification for the input data record (Peuter [0080] teaches tokens include a plurality of tokens that each has positional coordinates and can form key value pairs within a document. In addition, Peuter [0035] & Fig. 1 teaches inputting the tokens into a classifying component – element 110 that include machine learning models ( i.e., record classification machine learning model) – element 112 in order to assign “a classification to each of the one or more tokens, such as a meaning or topic related to a token”).
Peuter is also in the same field of endeavor as Erdemir and Florencio (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality storing key-value pairs and inputting tokens that form key value pairs into a classification model, as being disclosed and taught by Peuter, in the system taught by Erdemir and Florencio to yield the predictable results of providing “an automated approach for classifying tokens and generating key-value pairs of tokens in order to determine proper relationships between data within a document” (Peuter [0021]).
Peuter does not teach or suggest wherein the spatial coordinate set describes a spatial position of the value data token within the spatial coordinate scheme including a center point of a rendered form of the value data token; an automatic annotation of the historical dataset being performed and (ii) the label-value pair is generated based at least in part on a pairing score assigned to the value data token based at least in part on an output of the vector classification machine learning model;
Nevertheless, Liang teaches the following:
...an automatic annotation of the historical dataset... ( Liang, Abstract teaches “automatically generating ground truth from electronic health records comprising unstructured clinical notes and structured data comprising entries each having respective values for fields”).
Liang is also in the same field of endeavor as Erdemir, Florencio, and Peuter (data processing). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of automatic generation of ground truth from electronic health records ( historical dataset of data records), as being disclosed and taught by Liang, in the system taught by Erdemir, Florencio, and Peuter to yield the predictable results of “improving performance of a natural language processing task by automating generation of ground truth from electronic health records” (Liang [0006]) and to address the “problematic lack of annotation for information extraction on clinical text” since current process for ground truth generation on clinical text is a manual process that is subject to human error (Liang [0014]).
Neither Erdemir, Florencio, Peuter or Liang suggest wherein the spatial coordinate set describes a spatial position of the value data token within the spatial coordinate scheme including a center point of a rendered form of the value data token; and (ii) the label-value pair is generated based at least in part on a pairing score assigned to the value data token based at least in part on an output of the vector classification machine learning model;
Nonetheless Gao teaches the following:
...wherein the spatial coordinate set describes a spatial position of the value data token within the spatial coordinate scheme including a center point of a rendered form of the value data token; (Gao [0037] teaches generating a bounding box “enclosing a phrase and determine the spatial coordinates of the bounding box as the location of the phrase. The spatial coordinates may be defined as {xmin, ymin, xmax, ymax}, where xmin is the leftmost horizontal coordinate, ymin is the lowest vertical coordinate, xmax is the rightmost horizontal coordinate, and ymax is the uppermost vertical coordinate of the bounding box”. In addition, [0042] teaches the location is the center of the bounding box).
generate a label-value pair comprising a label data token and the value data token based on the spatial coordinate set ( Gao [0038] teaches a key-value identifier module extracts information in the form of key-value pairs (i.e., label-value pair). In particular, [0040] teaches the “ key-value identifier module 425 identifies a set of neighbors for each candidate value from the set of phrases that are spatially close to the candidate value on the document... neighbor phrases for a candidate value are those that have a spatial location within a predetermined distance from the location of the candidate value on the document”)
...ii) the label-value pair is generated based at least in part on a pairing score assigned to the value data token neighbor score and sets the selected candidate value as the value for the field and the selected neighbor as the key for the field” (emphasis added). Thus, “neighbor score” as disclosed by Gao can be view as the “pairing score” as it used to generate the key-value pair (i.e., label-value pair)).
Gao is also in the same field of endeavor as Erdemir, Florencio, Peuter and Liang (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of the label-value pair is generated based at least in part on a pairing score assigned to the value data token, as being disclosed and taught by Gao, in the system taught by Erdemir, Florencio, Peuter and Liang to yield the predictable results of “automatically extract information from key-value pairs on form documents without a human operator so that processing can be done more efficiently” (Gao [0017])
Regarding claim 2:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1. Florencio specifically teaches wherein the label data token is identified using a label-detecting regular expression to parse the input data record (Florencio [0076] teaches “identifying regular expressions to use for the different groupings of elements in the form” such that label data tokens can be identify in the form ( input data record)).
Regarding claim 3:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1.
Erdemir teaches generating a plurality of ground-truth coordinate vectors based at least in part on... ( Erdemir [0102] and Fig. 6 teaches the Cartesian distance can be used to generate a plurality of coordinate vectors based on the coordinated of the tokens).
Erdemir does not teach wherein the vector classification machine learning model is generated by: accessing the historical dataset, wherein the historical dataset comprises a plurality of historical primary data records corresponding to a plurality of historical secondary data records; extracting the plurality of ground-truth label-value pairs from the plurality of historical secondary data records; generating a plurality of ground-truth coordinate vectors based at least in part on identifying the plurality of ground-truth label-value pairs in the plurality of historical primary data records; and training the vector classification machine learning model using the plurality of ground- truth coordinate vectors.
Nevertheless, Florencio teaches the following:
wherein the vector classification machine learning model by: ( Florencio Fig. 6 teaches a method for generating training model (vector classification machine learning model) element 685).
generating a plurality of ground-truth coordinate vectors based at least in part on identifying the plurality of ground-truth label-value pairs in the plurality of historical primary data records; and ( Florencio [0069] teaches “identify relative positioning of the key-value data (label-value pairs ) within the forms (historical primary data records)” such that relative positioning information “can be stored as metadata with the form or in a separate data structure, which can be used by the models as supplemental ground truth for identifying location of similar key-value data during subsequent processing with the model on other forms”).
training the vector classification machine learning model using the plurality of ground- truth coordinate vectors. ( Florencio [0013] teaches “the key-value pairing data is used as ground truth for training a model to independently identify the key-value pairing(s)”).
Though Florencio teaches having access to the historical dataset of form in para. [0103], Florencio does not suggest accessing the historical dataset, wherein the historical dataset comprises a plurality of historical primary data records corresponding to a plurality of historical secondary data records; extracting the plurality of ground-truth label-value pairs from the plurality of historical secondary data records; and generating a plurality of ground-truth...
However, Liang teaches the following:
accessing the historical dataset, wherein the historical dataset comprises a plurality of historical primary data records corresponding to a plurality of historical secondary data records; ( Liang [0015] and Fig. 1 teaches an “Electronic health records (HER)” element 105 ( historical dataset) that comprises unstructured notes (historical primary data) element 125” and “structured data” element 115 ( historical secondary data record).
extracting the plurality of ground-truth label-value pairs from the plurality of historical secondary data records; ( Liang [0015] teaches the HER includes structured data (historical secondary data record) and [0036] teaches “auto-generating silver standard ground truth for information extraction from clinical text using structured EHR data”).
Regarding claim 4:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 3. Liang teaches specifically teaches wherein a historical secondary data record of the plurality of historical secondary data records semantically describes data in a corresponding historical primary data record of the plurality of historical primary data records, and... ( Liang [0020] and FIG. 1, teaches linking the structured EHR entries element 115 to unstructured clinical notes element 125. Such that, it results in a “an expanded representation of the patient EHR where each structured entry contains both native data elements and derived insights and also is linked to unstructured clinical notes generated from the same encounter along with the associated note metadata (e.g., note type, note data, note author, etc.)” ( Liang [0022])) and Florencio teaches wherein the plurality of ground-truth label-value pairs are extracted using the label- value pair regular expression to parse the historical secondary data record ( Florencio [0162] teaches identifying forms to use for harvesting (extracting) key-value pairing ground truth from (historical secondary data record) and Florencio [0076] teaches identifying regular expression ( this can be a label- value pair regular expressions) to use for the different grouping of elements in the form).
Regarding claim 6:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1. Erdemir specifically teaches wherein the spatial coordinate set is determined in accordance with bounding boxes generated for the label data token and the value data token via the one or more optical character recognition techniques ( Erdemir [0027] teaches “group content in label value pairs based on document layout analysis” and [0028] teaches “a bounding box is determined for each group such that the box includes all members of the group. The (x and y) coordinates of points on the perimeter of the bounding box may be used for determining the position of the corresponding group, or the distance of the group to other groups”. Further, [0045] teaches optical character recognition (OCR) techniques “may be used to identify words or contiguous sets of characters within the document”).
Regarding claim 7:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1. Erdemir specifically teaches wherein the coordinate vector comprise an angular orientation and a radial distance configured to describe a relative positioning of the label data token and the value data token within the spatial coordinate scheme ( Erdemir [0102] teaches generating coordinate vectors using Cartesian distance. As would be familiar to one skilled in the art, the horizontal separation and the vertical separation can be view as the side of a triangle for which the distance is the length of the hypotenuse. Therefore, this suggest the Cartesian distance used to generate the coordinate vector comprises both an angular orientation and a radial distance).
Regarding claim 8:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 7. Erdemir specifically teaches [coordinate vectors generated] using Cartesian distance in para [0102], Florencio teaches the machine learning models that can be used include naïve bayes classifier in para. [0104]. As would be familiar to one skilled in the art, naïve bayes classifier comprises joint probability distribution
P
X
,
Y
[wherein the vector classification machine learning model comprises a joint probability distribution with respect to at least the angle and the distance] and Liang Abstract teaches “automatically generating ground truth from electronic health records comprising unstructured clinical notes and structured data comprising entries each having respective values for fields” which suggest ground truth coordinate vector could also be generated automatically [a plurality of ground-truth coordinate vectors generated from the automatic annotation].
Regarding claim 11:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1. Specifically Erdemir teaches further comprising: initiating one or more post-extraction actions based at least in part on the label-value pair (Erdemir [0026] teaches extracting and grouping content from electronic document (input data record) and teaches the content can presented. In addition, [0061] teaches the label-value pairs can be marked in the electronic document itself and [0065] emphasize, the “groups of phrases such as label-value pairs, that are generated in accordance with the operations described... may be linearly read out, streamed, stored, presented or otherwise used”).
Regarding claim 12:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 11. Specifically Erdemir teaches wherein the one or more post-extraction actions further comprises at least one of: providing the label-value pair to a disease diagnosis model or generating a summarization data object for the input data record that comprises the label-value pair ( Examiner will like to emphasize, the claim as presented recites the one or more post-extraction actions further comprises at least one of: providing the label-value pair to a disease diagnosis model or generating a summarization data object for the input data record that comprises the label-value pair (emphasis added), for which Erdemir [0026] teaches extracting and grouping content from electronic document (input data record) and teaches the content can presented. In addition, [0061] teaches the label-value pairs can be marked in the electronic document itself and [0065] emphasize, the “groups of phrases such as label-value pairs, that are generated in accordance with the operations described... may be linearly read out, streamed, stored, presented or otherwise used”. Furthermore, Erdemir FIG. 5 teaches generating a summarization data object (label value pair element 520) for the input data record (invoice element 500)).
Regarding claim 14: it is rejected under the same rationality as claim 1. Claim 14 only recites the additional elements of An apparatus comprising a processor and at least one memory comprising computer program code, the at least one memory and the computer program code configured to, with the processor, cause the apparatus to.. for which Liang [0008] teaches “embodiments of the invention or elements thereof can be implemented in the form of an apparatus including a memory, and at least one processor that is coupled to the memory and operative to perform exemplary method steps”.
Regarding claim 15: it is rejected under the same rationality as claim 3. Claim 15 only recites the additional elements of The apparatus, for which Liang [0008] teaches an apparatus.
Regarding claim 16: it is rejected under the same rationality as claim 4. Claim 16 only recites the additional elements of The apparatus, for which Liang [0008] teaches an apparatus.
Regarding claim 17: it is rejected under the same rationality as claim 5. Claim 17 only recites the additional elements of The apparatus, for which Liang [0008] teaches an apparatus.
Regarding claim 18: it is rejected under the same rationality as claim 7. Claim 18 only recites the additional elements of The apparatus, for which Liang [0008] teaches an apparatus.
Regarding claim 19: it is rejected under the same rationality as claim 8. Claim 19 only recites the additional elements of The apparatus, for which Liang [0008] teaches an apparatus.
Regarding claim 20: it is rejected under the same rationality as claim 1. Claim 20 only recites the additional elements of A computer program product comprising at least one computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions including executable portions configured to cause at least one processor to... for which Liang [0008] teaches “a computer program product including a computer readable storage medium with computer usable program code for performing the method steps indicated”.
Regarding claim 23:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1. Erdemir specifically teaches wherein generating the label-value pair further comprises: generating a plurality of coordinate vectors corresponding to the plurality of value data tokens, for the label data token, ( Erdemir [0102] teaches generating coordinate vectors using cartesian distance. Such that, the “Cartesian distance (also known as a Euclidean distance) 650 can be calculated using the horizontal separation 630 and the vertical separation 640... For purposes of calculating the Cartesian distance, the closest conners of the respective bounding boxes are generally used”. Further, Fig. 6 above, teaches applying “Cartesian Distance” element 650 between two tokens element 610 and 620).
And Gao specifically teaches:
generating a plurality of pairing scores corresponding to the plurality of value data tokens based on the plurality of coordinate vectors, and ( Paragraph [0085] and Fig. 7 bottom left of the instant application discloses” the coordinate vectors 708 may each be defined by a distance (e.g., R) and an angle (e.g., 0 with respect to a horizontal axis)” for which Gao Abstract teaches determines neighbor scores (i.e., pairing scores), where a neighbor score for a candidate value (i.e., value data tokens ) is determined based on a spatial relationship (i.e., coordinate vectors). Gao [0041] specifically teaches the spatial relationship is defined by the distance and angle).
generating the label-value pair by selecting the value data token to pair with the label data token based at least in part on a pairing score of the plurality of pairing scores assigned to the value data token satisfying a threshold(Gao [0030] teaches selecting “a candidate value and a respective neighbor based on the neighbor score and sets the selected candidate value as the value for the field and the selected neighbor as the key for the field” (emphasis added). Thus, “neighbor score” as disclosed by Gao can be view as the “pairing score” as it used to generate the key-value pair (i.e., label-value pair). In addition, Gao [0047] teaches the key-value pair (label-value pair) generated based on the selection of a candidate value (i.e., value data token) and respective neighbor (i.e., label data toke) with a neighbor score above a threshold (i.e., satisfying a threshold)).
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Erdemir, Florencio, Peuter, Liang and Gao in further view of Sathyanarayana et al. US 8,676,731 B1 (hereinafter Sathyanarayana) as being disclosed in the IDS filled on 06/16/2022.
Regarding claim 10:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1. Neither Erdemir, Florencio, Peuter, Liang or Gao teach wherein the pairing score assigned to the value data token is further based at least in part on an entity-specific value distribution associated with the label data token.
Nonetheless, Sathyanarayana teaches the following:
wherein the pairing score assigned to the value data token is further based at least in part on an entity-specific value distribution associated with the label data token ( Sathyanarayana (col. 1:67 & col. 2:1-2) teaches the confidence model aggregated the confidence component to computes a final confidence (i.e., pairing score) for the extracted data value (i.e., value data token). Further, according to the applicant specification para. [0086] each classifier machine learning model 910 is configured to output a pairing score (e.g., between 0 and 1), for which Sathyanarayana (col. 6:38-41) teaches “the confidence components are aggregated using a pre-defined non-linear mechanism to derive a real score in the range of [0-1] to yield a final confidence score”). In addition, Sathyanarayana (col. 5:30-33) teaches “deriving a statistical likelihood (distribution) of the transformation recognizing the input and generating the output representative of the same data item” and (col. 10:20-23) teaches “the evaluated final confidence is used to categorize the data to one of the predefined quality groups based on a frequency distribution study of the final confidence values for the generated data 130”).
Sathyanarayana is also in the same field of endeavor as Erdemir, Florencio, Peuter, Liang and Gao (data processing). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of “pairing scores” as being disclosed and taught by Sathyanarayana, in the system taught by Erdemir, Florencio, Peuter, Liang and Gao to yield the predictable results of automatically capture data from documents, doing so reducing “the effort of verification in an intelligent manner for further reduction in human efforts for data capture” (Sathyanarayana col. 1:54-56)
Claim 21 is rejected under 35 U.S.C. 103 as being unpatentable over Erdemir, Florencio, Peuter, Liang and Gao in further view of Krishnan et al. US 2006/0184475 A1 (hereinafter Krishnan).
Regarding claim 21:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1. Peuter specifically teaches wherein (i) the vector classification machine learning model comprises an ensemble classifier that comprises a plurality of classifier machine learning models, ( Peuter Fig. 1 – element 126 teaches the machine learning model (i.e., vector classification machine learning model) and [0026] teaches one or more machine learning model is trained and used to generate key-value pairs between the tokens within a document. Thus, suggesting the machine learning model element 126 in Fig. 1 can comprises one or more machine learning models used to generate key-value pairs).
and (ii) a classifier machine learning model of the plurality of classifier machine learning models is trained using a of a plurality of ground truth coordinate vectors and a negative sample of a plurality of incorrect label-value pairs (Peuter [0037-0038] teaches relationships (i.e., coordinate vectors) between tokens similarly includes how the tokens of the key-value pair are related, such as ... if the key and value tokens both share a same positional coordinate. Further, Peuter [0095] teaches the machine learning model (i.e., classifier machine learning model) is trained using a negative sample of a plurality of incorrect label-value pairs as the machine learning model is trained to learn to differentiate between correct and incorrect key value pairs and thus suggesting the machine learning model is trained with negative sample in order to be able to differentiate. Moreover, [0095] also teaches the machine learning model is trained using the relationship that include relationship between the tokens of historical documents, thus implying the machine learning model is also trained using historical relationship (i.e., ground truth coordinate vectors)).
While Peuter paragraph [0026] teaches the machine learning models therein can generate key value pairs, neither Erdemir, Florencio, Peuter, Liang or Gao explicitly disclose a classifier machine learning model of the plurality of classifier machine learning models is trained using a subset of a plurality of ground truth coordinate vectors and a negative sample subset of a plurality of incorrect label-value pairs.
Nonetheless, Krishnan teaches the following:
wherein (i) the vector classification machine learning model comprises an ensemble classifier that comprises a plurality of classifier machine learning models, (Krishnan [0020] teaches “a number of classifiers are constructed where each classifier is trained on a distinct set of features” further, [0030]-[0031] teaches the plurality of classifiers can each be a different or same type of classifier).
and (ii) a classifier machine learning model of the plurality of classifier machine learning models is trained using a subset of a plurality of ground truth coordinate vectors and a negative sample subset of a plurality of incorrect label-value pairs (Krishnan [0043] teaches a sub-set of features (i.e., subset of a plurality of ground truth coordinate vectors and a negative sample subset of a plurality of incorrect label-value pairs) are extracted from previously collected training data (such training data includes ground truths, see para. [0012]) and teaches how a classifier is trained using the selected sub-set of training data).
Krishnan is also in the same field of endeavor as Erdemir, Florencio, Peuter, Liang and Gao (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of multiple classifiers and a classifier trained using ground truth and negative training samples, as being disclosed and taught by Krishnan, in the system taught by Erdemir, Florencio, Peuter, Liang and Gao to yield the predictable results of providing a better classifier using less than all the features ( Krishnan [0021]).
Claim 22 is rejected under 35 U.S.C. 103 as being unpatentable Erdemir, Florencio, Peuter, Liang and Gao in further view of Majumdar et al. US 2017/0039469 A1 (hereinafter) in further view of Kuo US 2011/0092222 A1 (hereinafter Kuo).
Regarding claim 22:
Erdemir, Florencio, Peuter, Liang and Gao teach The computer-implemented method of claim 1.
Neither Erdemir, Florencio, Peuter, Liang or Gao specifically teach wherein the vector classification machine learning model is trained using a plurality of negative coordinate vectors comprising one or more randomly-sampled distance aspects and one or more randomly-sampled angle aspects.
Nonetheless, Majumdar teaches:
wherein the vector classification machine learning model is trained using a plurality of negative coordinate vectors
Majumdar is also in the same field of endeavor as Erdemir, Florencio, Peuter, Liang and Gao (machine learning). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of training a classifier with negative coordinate vector, as being disclosed and taught by Majumdar, in the system taught by Erdemir, Florencio, Peuter, Liang and Gao yield the predictable results of “improving systems and methods for detecting unknown classes and initializing classifiers for unknown classes” (Majumdar [0003])
Majumdar does not explicitly teaches the coordinate vector comprising one or more randomly-sampled distance aspects and one or more randomly-sampled angle aspects.
However, Kuo teaches the following:
....a plurality of coordinate vectors comprising one or more randomly-sampled distance aspects and one or more randomly-sampled angle aspects (Kuo [0061] teaches coordinate vectors comprising random angle from x-axis and random distance).
Kuo is also in the same field of endeavor as Erdemir, Florencio, Peuter, Liang , Gao and Kuo (data structure). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the functionality of feature vectors comprising random distance and angle, as being disclosed and taught by Kuo, in the system taught by Erdemir, Florencio, Peuter, Liang , Gao and Kuo to yield the predictable results of improve the positioning precision (Kuo [0050]).
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 GISEL G FACCENDA whose telephone number is (703)756-1919. The examiner can normally be reached Monday - Friday 8:00 am - 4:00 pm.
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, Abdullah Al Kawsar can be reached at (571) 270-3169. 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.
/G.G.F./Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
Prosecution Insights