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 .
DETAILED ACTION
The Action is responsive to the Amendments and Remarks filed on 1/22/2026. Claims 1-19 and 21 are pending claims. Claims 1, 8, and 15 are written in independent form. Claim 20 has been cancelled. Claim 21 is newly added.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-19 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Reisswig (U.S. Pre-Grant Publication No. 2021/0383067) and further in view of He et al. (U.S. Pre-Grant Publication NO. 2018/0357262, hereinafter referred to as He262), Zeng et al. (U.S. Pre-Grant Publication No. 2022/0067275, hereinafter referred to as Zeng), and Vemulpali (U.S. Pre-Grant Publication No. 2021/0168125).
Regarding Claim 1:
Reisswig teaches a computer-implemented method comprising:
Identifying, by one or more processors, a plurality of table tokens of a table data object, wherein a table token of the plurality of table tokens is associated with a first relative token position measure within the table data object;
Reisswig teaches “raw document 510 is provided as input to process 511 for token extraction” where “following token extraction, additional preprocessing operation can be performed” including “named entity recognition on tokens of document 510”, tokens can be encoded using word embedding, to obtain respective vectors for each word token”, and “positional encoding can be derived for each token, indicating a spatial position of that token in the raw document 510” (Para. [0089]).
generating, by the one or more processors and using a token-wise entity type classification machine learning model, a predicted entity type for the table token wherein:
Reisswig teaches “the outputs of the one or more preprocessors 511, 513, 515, 517 can be combined to obtain an input record 520 for level one of the classifier 523” where “each of the input vectors of record 520 can include a word embedding representation of the corresponding token, as well as positional encoding information, and optionally additional metadata” (Para. [0090]). Reisswig further teaches generating predicted entity types by teaching performing entity recognition “to identify predefined document fields” which “can be used to tag or label corresponding vertices of [level one] graph 524” (Para. [0091]).
(i) the predicted entity type is selected from a plurality of defined entity types associated with an inter-related entity type taxonomy for the table data object, and
Reisswig teaches “tokens of a text document are mapped to previously defined fields” (Para. [0084]) where “these tokens can be determined through preprocessing of an input document” (Para. [0085]) and “in other examples the fields 440 can optionally be organized as part of a schema 430” (Para. [0086]). Therefore, Reisswig teaches selecting labels from a plurality of predefined labels associated with a schema for the table data object.
(ii) the inter-related entity type taxonomy defines, for a defined entity type of the plurality of defined entity types, a potentially related defined entity type set that includes at least another defined entity type of the plurality of defined entity types;
Reisswig teaches “In some examples, obtaining the input document record can include preprocessing an original form of the input document by recognizing at least one field in the input document as a named entity and replacing the recognized field(s) in the input document with the named entity.” (Para. [0004]). Reisswig further teaches “a schema of fields for a given class can be defined at or prior to phase 102 and used at phase 104 to recognize tokens of the input document as instances of respective predefined fields” (Para. [0074]) and “In some examples, the schema 430 can be organized hierarchically in a manner similar to the relationships between document fields” (Para. [0087]) thereby teaching the schema defining, for each field type, a potentially related set of types in the hierarchically organized schema.
Generating, by the one or more processors, a related token set of the plurality of table tokens for the table token, wherein:
Reisswig teaches “a level two input record 530 can be constructed using the input record 520 for level one and the level one graph structure” where “the composite tokens can replace their constituent simple tokens” (Para. [0092]). Thereby teaching generating a related token set of the plurality of table tokens for the table token.
(i) a related table token in the related token set is associated with a corresponding defined entity type that is in the potentially related defined entity type set, and
Reisswig teaches, as part of constructing the level two input record 530, “the level one graph 524 can be evaluated by entity recognition module 525 to identify predefined document fields in the graph 524, and these recognized fields can be used to tag or label corresponding vertices of graph 524” (Para. [0091]) thereby teaching that each related table token is labeled/associated with a defined entity type in the potentially-related defined entity type set.
(ii) the related table token in the related token set is selected based at least in part on the first relative token position measure for the table token and a second relative token position measure for the related table token;
Reisswig teaches “Each of the input vectors of record 520 can include a word embedding representation of the corresponding token, as well as positional encoding information, and optionally additional metadata, such as font size, style, or color. For one-dimensional position encoding (i.e. a representation of document 510 as a linear sequence of tokens), the positional encoding information can be added (superposed) on the word embedding vector, or can be concatenated with the word embedding vector” (Para. [0090]) where “the composite tokens can replace their constituent simple tokens” (Para. [0092]). Reisswig further teaches “the grouping of tokens is not restricted to spatial neighborhood since the graph approach can relate any token with any other token of an input record” (Para. [0033]) thereby teaching the use of token positioning but not being restricted solely on the spatial neighborhood.
Generating, by the one or more processors, and based at least in part on the plurality of table tokens and the related token set, a graph-based table representation for the table data object, wherein:
Reisswig teaches “preprocessed document 250 and the trained neural network 230 can be input to block 265, where the neural network 230 can evaluate the document 250 to generate a document graph 270” (Para. [0078]).
(i) a plurality of nodes of the graph-based table representation respectively correspond to the plurality of table tokens,
Reisswig teaches “in some examples, each vertex of the graph structure can represent a single token or a group of multiple related tokens” (Para. [0008])
(ii) a plurality of links of the graph-based table representation respectively correspond to a plurality of cross-token relationships defined by the related token set,
Reisswig teaches “in some examples, each vertex of the graph structure can represent a single token or a group of multiple related tokens” where “each edge of the graph structure can represent a relation between a pair of the vertices joined by the edge” (Para. [0008]). Therefore, the links/edges of the graph-based table representation correspond to relationships between related token(s) at each node.
Generating, by the one or more processors, a semantic table representation;
Reisswig teaches generating a semantic table representation by teaching “one or more vertices of the structure graph can be mapped to columns of a database” (Para. [0134]) and “respective values of the mapped vertices can be stored in the database. For example, the value of Provider Name 741 can be “DEF Laboratory” as extracted from document 710, and this same value can be stored at the corresponding column 912 of record 915 in table 910. For a key-value pair such as Document Date 741, the Value 714 can be stored in table 920, while the Key 715 “Date:” can be omitted, as being implicit in the column “Document Date.” In other examples, the Key of a Key-Value pair can match the column label of a database table.” (Para. [0135]).
Reisswig explicitly teaches all of the elements of the claimed invention as recited above except:
(iii) a first node of the plurality of nodes is a parent of or a child to a second node of the plurality of nodes based on a cross-token proximity score and a threshold criterion, wherein:
(a) the cross-token proximity score comprises a weighted combination of a set of proximity indicators that indicate different positional or organizational distance measures between a first table token of the first node and a second table token of the second node within the table data object,
(b) the threshold criterion defines at least one of a cross-token proximity score threshold or a relative threshold, and
(c) the first node is assigned as a parent to the second node based on the cross-token proximity score achieving the at least one of the cross-token proximity score threshold or the relative threshold;
performing a defined-pattern graph traversal operation on the graph-based table representation; and
Performing, by the one or more processors, a load balancing operation for a post prediction system using the semantic table representation to set a number of allowed computing entities used by the post prediction system.
However, in the related field of endeavor of determining a hierarchical concept tree using a large corpus of table values, He262 teaches:
(iii) a first node of the plurality of nodes is a parent of or a child to a second node of the plurality of nodes based on a cross-token proximity score and a threshold criterion, wherein:
He262 teaches “each entity nodes represents a distinct value identified in a given table cell” (Para. [0020]) where “the candidate cluster module 218 collapses all node pairs whose edge scores are higher than a selected similarity threshold.” (Para. [0043]) and the node pair edge scores are cross-token proximity scores (nodes connecting values extracted from across various database tables and spreadsheet files) by teaching “The table processing server 122 is configured to retrieve one or more values represented by the intersection of the one or more columns and/or one or more rows for each of the database tables 112,118 and for each of the files 114-116. The values extracted from the various database tables 112,118 and spreadsheet files 114-116 are stored as extracted values 120.” (Para. [0029]).
(a) the cross-token proximity score comprising a weighted combination of a set of proximity indicators that indicate different positional or organizational distance measures between a first table token of the first node and a second table token of the second node within the table data object,
He262 teaches “For that iteration, this effectively reduces to computing connected component in a distributed setting, since it can disconnect one or more (e.g., all) nodes whose scores are lower than the threshold, and treat all remaining edges as unweighted.” (Para. [0043]) and “’high quality’ concepts are considered those concepts where the values within the table column and/or spreadsheet column share have some characteristic in common (e.g., the values represent geographic locations, the values correspond to particular regions or identities, the values correspond to names of personnel and/or buildings, etc.)” (Para. [0050]) thereby teaching a weighted edge score for one or more characteristics in common including positional or organizational distance measures between characteristics of the first node and characteristics of the second node representing table data.Reisswig further explicitly teaches using data structures such as token and objects to represent values and characteristics (Paras. [0008] [0052] & [0062])).
(b) the threshold criterion defines at least one of a cross-token proximity score threshold or a relative threshold, and
He262 teaches “After the table corpus processing server 122 has determined the one or more similarity values 226, the table corpus processing server 122 then determines one or more clusters of similar values. Accordingly, in one embodiment, the table corpus processing server 122 includes a candidate cluster module 218 to determine the one or more clusters of similar table values 222 and similar spreadsheet values 224.” (Para. [0041]) where “edge scores are higher than a selected similarity threshold” (Para. [0043])
(c) the first node is assigned as a parent to the second node based on the cross-token proximity score achieving the at least one of the cross-token proximity score threshold or the relative threshold;
He262 teaches “After the table corpus processing server 122 has determined the one or more similarity values 226, the table corpus processing server 122 then determines one or more clusters of similar values. Accordingly, in one embodiment, the table corpus processing server 122 includes a candidate cluster module 218 to determine the one or more clusters of similar table values 222 and similar spreadsheet values 224.” and “as known to one of ordinary skill in the art, MapReduce is a component that parcels out work to various nodes within a cluster (or map), and it organizes and reduces the results from each node within the cluster into a cohesive answer to a given query. The cluster nodes determined by the candidate cluster module 218 may be stored as the candidate clusters 228. The resulting hierarchical tree structure constructed by the candidate cluster module 218 may be stored as the candidate concept tree 230.” (Para. [0041]).
Thus, it would have been obvious to one of ordinary skill in the art, having the teachings of He262 and Reisswig at the time that the claimed invention was effectively filed, to have modified the systems and methods for data-driven structure extraction from text documents, as taught by Reisswig, with the expanded use of including spreadsheet files as a source of table values in the corpus of table values, as taught by He262.
One would have been motivated to make such modification because while Reisswig teaches “The term “record” refers to a collection of data fields having respective values, and is used in different contexts. In the context of a classifier, a record can be an input record (e.g. a sequence of features to be classified) or an output record (e.g. one or more output labels). In some examples, an output record of one level or layer of a classifier can be an input record of a following level or layer of the classifier. In the context of training, a training record can include both the input record and the desired output label(s). In the context of a database, a record can be a row of a database table, with the fields of the record containing respective values for the columns of the database table.” (Para. [0008]), He262 further teaches expanding the type of data sources to include spreadsheet files by teaching “the database retrieval module 212 and the spreadsheet file retrieval module 214 are configured to extract one or more values from the database tables 112,118 and the spreadsheet files 114,116. The extracted values 120 are then stored as the retrieved table values 222 and the retrieved spreadsheet values 224.” (Para. [0037]) and it would have been obvious to a person having ordinary skill in the art that comparing values from more sources, such as spreadsheet files, would increase the dynamic ability of the system and method taught by Reisswig.
Reisswig and He262 explicitly teach all of the elements of the claimed invention as recited above except:
performing a defined-pattern graph traversal operation on the graph-based table representation; and
Performing, by the one or more processors, a load balancing operation for a post prediction system using the semantic table representation to set a number of allowed computing entities used by the post prediction system.
However, in the related field of endeavor of data extraction from unstructured documents, Zeng teaches:
performing a defined-pattern graph traversal operation on the graph-based table representation; and
Zeng teaches “wherein generating the structured text representation of the unstructured document using the text graph comprises performing a depth-first traversal on the text graph, a bread-first traversal on the text graph, a table identification search on the text graph, a box identification search on the text graph, or a level number search on the text graph.” (Para. [0092]) and “In an example, the structured text representation 1200 may be filtered using a depth-first traversal, a breadth-first traversal, a specific table identification, a specific box identification, or a specific level number” (Para. [0061]) thereby teaching performing one or more defined-pattern graph traversal operations.
Thus, it would have been obvious to one of ordinary skill in the art, having the teachings of Zeng, He262, and Reisswig at the time that the claimed invention was effectively filed, to have modified the systems and methods for data-driven structure extraction from text documents, as taught by Reisswig, and the expanded use of including spreadsheet files as a source of table values in the corpus of table values, as taught by He262, with the question & answering of the structured text, as taught by Zeng.
One would have been motivated to make such modification because Reisswig merely teaches storing the structured extracted data in a database 280 and Zeng teaches “The question and answer processes 118 may involve a user asking a question, and the question and answer processes 118 returning a result of the answer from the structured text representation 108 of the unstructured document 110. For example, a question “who is the customer identified in this document?” may return an answer indicated the name of a customer associated with the document.” (Zeng - Para. [0031]) and it would have been obvious to a person having ordinary skill in the art that allowing users to ask questions about the processed raw documents, such as about an invoice (Reisswig - Para. [0080] & Fig. 3A), would improve the use-ability of the stored structured data.
Zeng, He262, and Reisswig explicitly teach all of the elements of the claimed invention as recited above except:
Performing, by the one or more processors, a load balancing operation for a post prediction system using the semantic table representation to set a number of allowed computing entities used by the post prediction system.
However, in the related field of endeavor of connectivity and security of centralized and distributed applications, Vemulpali teaches:
Performing, by the one or more processors, a load balancing operation for a post prediction system using the semantic table representation to set a number of allowed computing entities used by the post prediction system.
Vemulpali teaches a “remote endpoint identity table 3740 with respect to the load balancing of flows for service endpoints” where “the identity IP 3706 value may match to values given in the service semantic table 1510 identity IP/subnet column 1517” (Para. [0281]) and, with respect to the remote endpoint identity table 3740, “The Cap 3790 value is received once capacity is reserved at the provider application endpoint as described below (with reference to FIG. 43A-43B) to limit the number of flows towards the provider application endpoint instance in context of load balancing” (Para.[0287]).
Thus, it would have been obvious to one of ordinary skill in the art, having the teachings of Vemulpali, Zeng, He262, and Reisswig at the time that the claimed invention was effectively filed, to have modified the systems and methods for data-driven structure extraction from text documents, as taught by Reisswig, the expanded use of including spreadsheet files as a source of table values in the corpus of table values, as taught by He262, and the question & answering of the structured text, as taught by Zeng, with the deployment of perimeter services, as taught by Vemulpali.
One would have been motivated to make such modification because Reisswig teaches “the server can implement the disclosed technologies as a service in a SaaS (software as a service) model, which can be called from a remote or client-side application” (Para. [0143]) and Vemulpali teaches “As the enterprises use ‘as a service’ services, their infrastructure and services are distributed across and they have to connect and secure their networks, applications, and improve performance of applications. Therefore, enterprises deploy perimeter services at the edge of branches, datacenters, public clouds, and VPCs. These perimeter services include firewalls, IDS/IPS, WAN optimization, VPNs, load balancers, DPI, tunnel proxies, and application performance gateways” (Para. [0014]).
Regarding Claim 2:
Vemulpali, Zeng, He262, and Reisswig further teach:
Wherein generating the plurality of table tokens comprises performing a named entity recognition operation on a text data object associated with the table data object to generate the plurality of table tokens.
Reisswig teaches “raw document 510 is provided as input to process 511 for token extraction” where “following token extraction, additional preprocessing operation can be performed” including “named entity recognition on tokens of document 510”, tokens can be encoded using word embedding, to obtain respective vectors for each word token”, and “positional encoding can be derived for each token, indicating a spatial position of that token in the raw document 510” (Para. [0089]).
Regarding Claim 3:
Vemulpali, Zeng, He262, and Reisswig further teach:
Wherein the text data object is generated by performing an optical character recognition operation on an image data object associated with the table data object.
Reisswig teaches “the preprocessing can include optical character recognition (OCR), token extraction, word embedding, named entity recognition (NER), positional encoding, or other metadata encoding. Some of these functions can be omitted from block 255, or can be incorporated within the trained neural network 230. The preprocessed document 260 can include an input record in the form of a sequence of vectors for respective tokens of the raw document 250” (Para. [0077]).
Regarding Claim 4:
Vemulpali, Zeng, He262, and Reisswig further teach:
Generating, by the one or more processors and using the token-wise entity type classification machine learning model, the predicted entity type of the table token by processing an input table token sequence of the plurality of table tokens for a table token associated with the input table token sequence.
Reisswig teaches “the outputs of the one or more preprocessors 511, 513, 515, 517 can be combined to obtain an input record 520 for level one of the classifier 523” where “each of the input vectors of record 520 can include a word embedding representation of the corresponding token, as well as positional encoding information, and optionally additional metadata” (Para. [0090]). Reisswig further teaches generating predicted entity types by teaching performing entity recognition “to identify predefined document fields” which “can be used to tag or label corresponding vertices of [level one] graph 524” (Para. [0091]).
Regarding Claim 5:
Vemulpali, Zeng, He262, and Reisswig further teach:
Wherein the input table token sequence is generated based at least in part on an optical character recognition output sequence of a text data object generated by performing an optical character recognition operation on an image data object associated with the table data object.
Reisswig teaches “the preprocessing can include optical character recognition (OCR), token extraction, word embedding, named entity recognition (NER), positional encoding, or other metadata encoding. Some of these functions can be omitted from block 255, or can be incorporated within the trained neural network 230. The preprocessed document 260 can include an input record in the form of a sequence of vectors for respective tokens of the raw document 250” (Para. [0077]).
Regarding Claim 6:
Vemulpali, Zeng, He262, and Reisswig further teach:
Wherein the defined-pattern graph traversal operation comprises a breadth first search graph traversal operation.
Zeng teaches “wherein generating the structured text representation of the unstructured document using the text graph comprises performing a depth-first traversal on the text graph, a bread-first traversal on the text graph, a table identification search on the text graph, a box identification search on the text graph, or a level number search on the text graph.” (Para. [0092]) and “In an example, the structured text representation 1200 may be filtered using a depth-first traversal, a breadth-first traversal, a specific table identification, a specific box identification, or a specific level number” (Para. [0061]).
Regarding Claim 7:
Vemulpali, Zeng, He262, and Reisswig further teach:
Wherein the defined-pattern graph traversal operation comprises a depth first search graph traversal operation.
Zeng teaches “wherein generating the structured text representation of the unstructured document using the text graph comprises performing a depth-first traversal on the text graph, a bread-first traversal on the text graph, a table identification search on the text graph, a box identification search on the text graph, or a level number search on the text graph.” (Para. [0092]) and “In an example, the structured text representation 1200 may be filtered using a depth-first traversal, a breadth-first traversal, a specific table identification, a specific box identification, or a specific level number” (Para. [0061]) thereby teaching performing one or more defined-pattern graph traversal operations.
Regarding Claim 8:
Some of the limitations herein are similar to some or all of the limitations of Claim 1.
Vemulpali, Zeng, He262, and Reisswig further teach a system comprising:
One or more processors (Reisswig - Paras. [0010] & [0161]); and
At least one memory storing processor-executable instructions that, when collectively or independently executed by any of the one or more processors, comprise causing the one or more processors to perform steps (Reisswig - Paras. [0010] & [0161]).
Regarding Claim 9:
All of the limitations herein are similar to some or all of the limitations of Claim 2.
Regarding Claim 10:
All of the limitations herein are similar to some or all of the limitations of Claim 3.
Regarding Claim 11:
All of the limitations herein are similar to some or all of the limitations of Claim 4.
Regarding Claim 12:
All of the limitations herein are similar to some or all of the limitations of Claim 5.
Regarding Claim 13:
All of the limitations herein are similar to some or all of the limitations of Claim 6.
Regarding Claim 14:
All of the limitations herein are similar to some or all of the limitations of Claim 7.
Regarding Claim 15:
Some of the limitations herein are similar to some or all of the limitations of Claim 1.
Vemulpali, Zeng, He262, and Reisswig further teach:
One or more non-transitory computer-readable storage media including instructions that, when executed by one or more processors, cause the one or more processors to perform steps (Reisswig - Paras. [0010] & [0161]).
Regarding Claim 16:
All of the limitations herein are similar to some or all of the limitations of Claim 2.
Regarding Claim 17:
All of the limitations herein are similar to some or all of the limitations of Claim 3.
Regarding Claim 18:
All of the limitations herein are similar to some or all of the limitations of Claim 4.
Regarding Claim 19:
All of the limitations herein are similar to some or all of the limitations of Claim 5.
Regarding Claim 21:
Vemulpali, Zeng, He262, and Reisswig further teach:
Wherein the set of proximity indicators comprises a table location indicator, wherein the table location indicator has a value of one if the first node and the second node are in a same boundary-separate table location, wherein the same boundary-separated table location is one of a set of: a cell, a row, a column, or a header.
He262 teaches “the determined similarity value(s) may be stored as the similarity values 226 and the similarity threshold value that the cluster selection module 220 uses to determine whether the concept o and a column in the table c are, in fact, similar, may be stored as the threshold value(s) 238.” (Para. [0052]).
Response to Amendment
Applicant’s Amendments, filed on 1/22/2026, are acknowledged and accepted.
Response to Arguments
On pages 12-13 of the Remarks filed on 1/22/2026, Applicant argues that Allen does not teach “positional and/or organizational proximity of table portions” and does not disclose the rest of the amended claim limitations.Applicant’s argument is convincing that Allen does not teach all of the amended claim limitations, thus necessitating the new grounds of rejection presented above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Bhotika et al. (U.S. Pre-Grant Publication No. 2020/0160050) teaches techniques for layout-agnostic complex document processing. A document processing service can analyze documents that do not adhere to defined layout rules in an automated manner to determine the content and meaning of a variety of types of segments within the documents. The service may chunk a document into multiple chunks, and operate upon the chunks in parallel by identifying segments within each chunk, classifying the segments into segment types, and processing the segments using special-purpose analysis engines adapted for the analysis of particular segment types to generate results that can be aggregated into an overall output for the entire document that captures the meaning and context of the document text.
Craymer, III (U.S. Pre-Grant Publication No. 2020/0264852) teaches process subroutine-structured graph-based intermediate representations during formal language processing implemented by a computing device. The method includes classifying a set of subroutines identified in an intermediate representation of code according to mutually recursive relationships between subroutines in the set of subroutines, recording the mutually recursive relationships, labeling to track the mutually recursive relationships, constructing a set of graph representations, collecting partial positions that distinguish points of action in generated code, labeling nodes of the graph of the intermediate representation, generating a subsequent intermediate representation by serialization of the graph of the intermediate representation through pre-order depth-first traversal, and creating the generated code from the intermediate representation.The reference further teaches “The pre-order depth first serialization process is described here is done via a graph traversal process. The process is utilized in the construction of a token stream, with some logic added for tokenization and to deal with the problem of recursion. The graph to be processed has a single start node and a single end node. Each node in the graph has “before” edges and “after” edges. Graphs for representing formal language are strongly directed and have a regular structure with fork and join points.” (Para. [0116])The reference further teaches “Labeling is carried out as a four-step process, as shown in FIG. 11: convert a collection of partial positions that require labeling to a graph, identify networked communities from that graph, create labels during an initial set of breadth-first traversals over each community, then propagate those labels to cover all relevant contexts.” (Para. [0141]).
Kumar (U.S. Pre-Grant Publication No. 2020/0184267) teaches training a system to extract information from documents comprises feeding digital form of training documents to an OCR module, which identifies multiple logical blocks in the documents and text present in the logical blocks. One or more tags for the whole of the document, the logical blocks and word tokens on the document are received by a tagging module. A text input comprising the text identified in the document and the tags for the whole of the document are received by a machine learning module. A first image of the document with layout of the one or more of the identified blocks superimposed, and the tags of the logical blocks in the document are received by the machine learning module, wherein the received text input, first image and tags for the logical blocks corresponds to a plurality of the training documents.
Li et al. (U.S. Pre-Grant Publication No. 2021/0201182) teaches performing a structured extraction on a text, a device and a storage medium. The method may include: performing a text detection on an entity text image to obtain a position and content of a text line of the entity text image; extracting multivariate information of the text line based on the position and the content of the text line; performing a feature fusion on the multivariate information of the text line to obtain a multimodal fusion feature of the text line; performing category and relationship reasoning based on the multimodal fusion feature of the text line to obtain a category and a relationship probability matrix of the text line; and constructing structured information of the entity text image based on the category and the relationship probability matrix of the text line.The reference further teaches “performing a structured extraction on a text may acquire the entity text image, and perform the text detection on the entity text image using a text detection technology such as an OCR technology, to obtain the position and the content of the text line of the entity text image. Here, the entity text image may be an image obtained by capturing or scanning the entity text, and is commonly an image of various card certificates and bills. The text line may refer to a text region in the entity text image, the shape of which is generally quadrilateral. Therefore, the position of the text line may be generally represented by the coordinates of the four corner points thereof. The content of the text line may be text content.” (Para. [0033]).
Non-Patent Literature Khairani et al., "Named-Entity Recognition and Optical Character Recognition for Detecting Halal Food Ingredients: Indonesian Case Study", IEEE, September 2022 10th International Conference on Cyber and IT Service Management (CITSM) pp. 01-05, doi: 10.1109/CITSM56380.2022.9935966. (Year: 2022) teaches solutions using OCR and NER technology to read and recognize the compositional entities listed on packaged product. The purpose is to guide Muslim consumers identifying ingredients of a consumers products by defining three food category entities in NER: Halal, Haram, Syubhat (doubtful). The proposed system is built using OCR to scan the composition listed on packaged products and processed with the trained NER Model; the evaluation of the model gets an F-Score value of 0.967, and in system testing, by testing 24 packaged products, it produces an OCR accuracy value of 90% and the accuracy of the NER model. for a food reading of 84%.
Lobez Comeras et al. (U.S. Pre-Grant Publication No. 2016/0188570) teaches “generate all possible concept instance candidates along with parent-son relations through patten matching and transformation of the parse tree” (Para. [0049]) where “the instance candidate generation uses also the contextual tokens” (Para. [0056]) and “create predicate-based relation links between tree nodes” (Para. [0062]) and then “assign a score to each node based on certain criteria. In an embodiment, the criteria can include one or more of the following: how many times the node appears in the text; how many times the node appears in predicate-based relations; and how many children the nodes have” (Para. [0063]).
Erler et al. (U.S. Pre-Grant Publication No. 2019/0317963) teaches a method for traversing hierarchical data is provided. The method may include generating, based on a source table stored in a database, an index for traversing a graph corresponding to the source table. The source table may identify a parent node for each node in the graph. The generating of the index may include iterating over the source table to generate an inner node map. The inner node map may include at least one mapping identifying one or more children nodes descending from an inner node in the graph. The graph may be traversed based at least on the index. The index may enable the graph to be traversed depth first starting from a root node of the graph and continuing to a first child node descending from the root node of the graph. Related systems and articles of manufacture, including computer program products, are also provided.
Shang et al. (U.S. Patent No. 11,423,307) teaches employing a graph neural network (GNN) to construct a taxonomy. The GNN is subject to a training cycle and an inference cycle. The training cycle encodes cross-domain terms pairs from a set of noisy cross domain pairs extracted from a corpora, and outputs a preliminary taxonomy. The inference cycle identifies candidate term pairs and selectively subjects the candidate term pairs to selective filtering to produce a system predicted taxonomy from the preliminary taxonomy.
Allen et al. (U.S. Patent No. 9,147,273) teaches “the MANA module 103 generates a multi-layered network based on the extracted group attributes and values. According to one embodiment, the values are normalized edge weights for each attribute, and the attributes and corresponding normalized weights are represented as a finite vector. Any one of various well known network graph generation tools may be employed to create the multi-layered network. The tool generates a link between two nodes where any one of the attributes has a normalized weight greater than 0.” (Col. 18 Lines 4-14) and “For each pair of vertices (i,j) for which K attributes…are not observed in network data, calculate the mean probability…for each k…that are linked by average over the corresponding pijk in each of the sampled dendrograms D. For example, if K=(meetings, phone calls), then k is a particular attribute in K, such as meetings or phone calls.” (Col. 18 Lines 55-60).
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 ROBERT F MAY whose telephone number is (571)272-3195. The examiner can normally be reached Monday-Friday 9:30am to 6pm.
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, Boris Gorney can be reached on 571-270-5626. 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.
/ROBERT F MAY/Examiner, Art Unit 2154 3/20/2026
/BORIS GORNEY/Supervisory Patent Examiner, Art Unit 2154