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 .
Response to Amendment
The response filed 07/17/2025 has been entered. Applicant has amended claims 1, 12, and 20. Claims 1-20 are currently pending in the instant application.
Response to Arguments
Applicant’s arguments, see pages 9-13, filed 07/17/2025 with respect to the rejection(s) of claim(s) 1-20 under 35 U.S.C. 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Duta (US 2020/0117944). Duta teaches the amended limitations that regarding key value pairs. Komarov teaches the amended limitations of the embedding model as seen in the current rejection below
Applicant's arguments filed 07/17/2025 have been fully considered but they are not persuasive. Regarding the arguments concerning the 101 rejection, Examiner respectfully disagrees. The amended claim limitations merely recite a generic embedding model (there is not recitation of training the model specifically for the document set). As claimed, the model is interpreted to be pre-trained before implementation such as an off the shelf embedding model that one could purchase and feed the document collection to the off the shelf model. The claim is merely applying the dataset to a generic model. For the reasons above, Examiner maintains the current rejection.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim 1-20 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1,
At Step 1:
The claim is directed to a "system" and thus directed to a statutory category.
At Step 2A, Prong One:
The claim recites the following limitations directed to an abstract idea:
" determining that a given key-value pair of the plurality of key-value pairs is a correct key-value pair based on the given key-value pair having a value token matching a type that corresponds to a type for a key token of the given key-value pair " as drafted recites a mental process as an evaluation or judgement. One can mentally determine that the key-value pair is a correct key-value pair based on the determining that the key-value pair has the value token matching the type of expected information associated with the key token as seen in [0050-0051]. For example one can determine for a line in the documents relating to a total of money, the matching value is expected to be a numerical value as seen in Figure 2A, if the value is a string of characters then it is not the correct pairing .
At Step 2A, Prong Two:
The claim recites the following additional elements:
“A method for extracting information,… to a component of an application associated with the document " which is a high-level recitation of a generic computer components and represents mere instructions to apply on a computer as in MPEP 2106.05(f), which does not provide integration into a practical application.
“generating, using an embedding model that is trained to generate embedding representations of documents, an embedding representation of the document; providing the embedding representation of the document as input to a classifier machine learning model, wherein the classifier machine learning model is trained to generate an output based on the embedding representation of the document, the output comprising " is generally linking the abstract idea to the particular field of use or technological environment of model as in MPEP 2106.05(h). As reflected in the claim, the step is akin to using the model as a mere tool as under MPEP 2106.05(f).
" a plurality of key-value pairs, wherein each key-value pair of the plurality of key-value pairs is based on the horizontal coordinate of each token of the plurality of tokens and the vertical coordinate of each token of the plurality of tokens, wherein each key-value pair comprises two tokens of the plurality of tokens with a same horizontal coordinate or a same vertical coordinate; and a type for each of the tokens in each key-value pair of the plurality of key- value pairs; " is insignificant extra-solution activity as retrieval/receiving of data (i.e. mere data gathering) such as 'obtaining information' as identified in MPEP 2106.05(g) and does not provide integration into a practical application.
"and providing the classification and the correct key-value pair to a component of an application associated with the document." is insignificant extra-solution activity as it amounts to necessary data gathering and outputting, (i.e., all uses of the recited judicial exception require such data gathering or data output). See Mayo, 566 U.S. at 79, 101 USPQ2d at 1968; OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1092-93 (Fed. Cir. 2015) (presenting offers and gathering statistics amounted to mere data gathering). does not provide integration into a practical application.
Viewing the additional limitations together and the claim as a whole, nothing provides integration into a practical application.
At Step 2B:
The conclusions for the mere implementation using a computer and field of use are carried over and does not provide significantly more
.
With respect to the "receiving" and “providing” identified as insignificant extra-solution activity above when re-evaluated this element is well-understood, routine, and conventional as evidenced by the court cases in MPEP 2106.05(d)(II), "i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); … OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network); buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network);" and thus remains insignificant extra-solution activity that does not provide significantly more.
Looking at the claim as a whole does not change this conclusion and the claim is ineligible.
Regarding claim 2, further recites The method of Claim 1, further comprising: identifying a set of tokens of the plurality of tokens, wherein each token of the set of tokens is of a same type, wherein generating the plurality of key-value pairs is further based on the identifying and the key-value pairs do not comprise tokens of the set of tokens. The claimed “identifying…” further recite the abstract idea in claim 1, one can mentally a set of tokens of the plurality of tokens, wherein each token of the set of tokens is of a same type, wherein generating the plurality of key-value pairs is further based on the identifying and the key-value pairs do not comprise tokens of the set of tokens as discussed in [0020]. The claim does not recite any additional elements at all. Therefore, the abstract idea is not integrated into a practical application nor is there significantly more.
Regarding claim 3, further recites The method of Claim 1, further comprising: determining that the value token comprises a numerical value within a range, wherein determining the correct key-value pair is further based on the value token comprising the numerical value within the range. The claimed “determining…” further recite the abstract idea in claim 1, one can mentally determine that the value token comprises a numerical value within a range, wherein determining the correct key-value pair is further based on the value token comprising the numerical value within the range as discussed in [0020]. The claim does not recite any additional elements at all. Therefore, the abstract idea is not integrated into a practical application nor is there significantly more.
Regarding claim 4, further recites The method of Claim 1, further comprising:determining a first key-value pair of the plurality of key-value pairs and a second key-value pair of the plurality of key-value pairs, wherein a first token is common to the first key-value pair and the second key-value pair; determining that a second token of the first key-value pair comprises a same horizontal coordinate as the first token; determining that a third token of the second key-value pair comprises a same vertical coordinate as the first token; and determining that one of the second token or the third token matches the type of expected information. The claimed “determining…” further recite the abstract idea in claim 1, one can mentally determine that the value token comprises a numerical value within a range, wherein determining the correct key-value pair is further based on the value token comprising the numerical value within the range as discussed in [0020]. The claim does not recite any additional elements at all. Therefore, the abstract idea is not integrated into a practical application nor is there significantly more.
Regarding claim 5, further recites The method of Claim 4, wherein the first token is the key token and the one of the second token or the third token is the value token. The claim limitations further recite is insignificant extra-solution activity as retrieval/receiving of data (i.e. mere data gathering) such as 'obtaining information' as identified in MPEP 2106.05(g) and does not provide integration into a practical application.
Regarding claim 6, further recites The method of Claim 1, further comprising determining a classification for at least one token of the plurality of tokens, wherein the at least one token of the plurality of tokens comprises the key token. is insignificant extra-solution activity as retrieval/receiving of data (i.e. mere data gathering) such as 'obtaining information' as identified in MPEP 2106.05(g) and does not provide integration into a practical application.
Regarding claim 7, further recites The method of Claim 6, wherein determining the classification for the at least one token of the plurality of tokens comprises predicting, by a machine learning model, that the at least one token is associated with the classification. The claimed “determining…” further recite the abstract idea in claim 1, one can mentally determine that the value token comprises a numerical value within a range, wherein determining the correct key-value pair is further based on the value token comprising the numerical value within the range as discussed in [0025]. The claim does not recite any additional elements at all. Therefore, the abstract idea is not integrated into a practical application nor is there significantly more.
Regarding claim 8, further recites The method of Claim 7, wherein predicting, by the machine learning model, that the at least one token is associated with the classification comprises:determining a probability that each token of the plurality of tokens belongs to each respective classification of a plurality of classifications; and matching each token of the at least one token to the classification based on a highest probability. The claimed “determining…” further recite the abstract idea in claim 1, one can mentally determine that the value token comprises a numerical value within a range, wherein determining the correct key-value pair is further based on the value token comprising the numerical value within the range as discussed in [0025]. The claim does not recite any additional elements at all. Therefore, the abstract idea is not integrated into a practical application nor is there significantly more.
Regarding claim 9, further recites The method of Claim 6, further comprising determining whether each token of the plurality of tokens is associated with another token, wherein determining the classification for the at least one token of the plurality of tokens from the document comprises predicting a classification for each token of the plurality of tokens that is not associated with another token. The claimed “determining…” further recite the abstract idea in claim 1, one can mentally determine that the value token comprises a numerical value within a range, wherein determining the correct key-value pair is further based on the value token comprising the numerical value within the range as discussed in [0025]. The claim does not recite any additional elements at all. Therefore, the abstract idea is not integrated into a practical application nor is there significantly more.
Regarding claim 10, further recites The method of Claim 1, wherein the two tokens of each key-value pair comprise at least one token that comprises a text string. is insignificant extra-solution activity as retrieval/receiving of data (i.e. mere data gathering) such as 'obtaining information' as identified in MPEP 2106.05(g) and does not provide integration into a practical application.
Regarding claim 11, further recites The method of Claim 1, further comprising providing the classification and the correct key-value pair to a user. The claim limitations is insignificant extra-solution activity as it amounts to necessary data gathering and outputting, (i.e., all uses of the recited judicial exception require such data gathering or data output). See Mayo, 566 U.S. at 79, 101 USPQ2d at 1968; OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1092-93 (Fed. Cir. 2015) (presenting offers and gathering statistics amounted to mere data gathering). does not provide integration into a practical application.
Claims 12-20 are rejected using similar reasoning seen in the rejection of claims 1-11 due to reciting similar limitations but directed towards a method and non-transitory computer-readable media.
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) 1-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Reisswig (US 2021/0383067) in view of Komarov et al (WO 2020/201247).
Regarding claim 1, Reisswig teaches A method for extracting information, comprising: receiving a document comprising a plurality of tokens, wherein each token in the plurality of tokens has a horizontal coordinate and a vertical coordinate ([0080] FIGS. 3A-3D are diagrams 301-304 illustrating processing of a document 310 according to an example of the disclosed technologies. To avoid clutter, reference designators for common items are not duplicated in diagrams 301-304. Diagram 301 of FIG. 3A illustrates the raw document 310, which can be an invoice for purchase of scientific equipment, in this case two glass slides and one microscope, from a provider “ABC Supplies.” Individual tokens can be extracted for text such as “ABC,” “Supplies,” “Description,” “Glass,” “Slides,” and so forth. Furthermore, certain text tokens (e.g. “Date,” “Price”) can be recognized as named entities using a named entity recognition tool. An auxiliary table can be used to store pairs of (named entity, original document text) to allow retrieval of the original document text at a later stage of document processing); a plurality of key-value pairs, wherein each key-value pair of the plurality of key-value pairs is based on the horizontal coordinate of each token of the plurality of tokens and the vertical coordinate of each token of the plurality of tokens, wherein each key-value pair comprises two tokens of the plurality of tokens with a same horizontal coordinate or a same vertical coordinate; ([0081] Diagram 302 of FIG. 3B shows the document 310 after preprocessing and one level of graph extraction. For purpose of illustration, named entities are prefixed with “F:” (e.g. “F:Date:”) for text fields, or with “N:” (e.g. “N:01/01/2020”) for numeric fields. Various arrows in FIG. 3B illustrate some relationships that can be determined at this stage. Numerous unidirectional arrows identify key-value relationships. These include date token “N:01/01/2020” as a value of key “F:Date,”, “12345” as a value of key “F:Invoice No.:”, “1” and “2” as distinct values for a same key “F:Qty,” “$22” and “$100” as distinct values of key “F:Price,” “Microscope” as a value of key “F:Description,” and “$122” as a value of key “F:Total.); and providing the correct key-value pair to a component of an application associated with the document ([0097] To summarize, the level one input can be obtained by preprocessing input text document 510 to generate the level one input record 520, which can include a sequence of input vectors representing respective tokens of the document 510. At each level 523, 533, 543, a respective input record 520, 530, 540 can be processed with a respective neural network to generate a portion 524, 534, 544 of the document's graph structure. For levels after the first level 523, that is, at subsequent levels 533, 543, the corresponding input record 530, 540 can be derived from the previous level's input record 520, 530 and the previous level's output graph portion 524, 534. Finally, a consolidated graph structure 550 which can include the graph portions 524, 534, 544, can be formed and outputted).
Reisswig does not explicitly teach generating, using an embedding model that is trained to generate embedding representations of documents, an embedding representation of the document; providing the embedding representation of the document as input to a classifier machine learning model, wherein the classifier machine learning model is trained to generate an output based on the embedding representation of the document, the output comprising: determining that a given key-value pair of the plurality of key-value pairs is a correct key-value pair based on the given key-value pair having a value token matching a type that corresponds to a type for a key token of the given key-value pair;
Komarov teaches generating, using an embedding model that is trained to generate embedding representations of documents, an embedding representation of the document; providing the embedding representation of the document as input to a classifier machine learning model, wherein the classifier machine learning model is trained to generate an output based on the embedding representation of the document, the output comprising: (FIG. 5 shows a flow diagram of an embodiment of an exemplary method for pretraining the learning module. In block 300, a plurality of initial data records are provided. These initial data sets are stored in the first data model, for example a document-oriented data model, by the multi-model database management system. In block 302, a plurality of initial tokens are generated from initial field values of the initial data records. In block 304, these initial tokens are each assigned to one or more initial token types, with all initial token assignments being defined as secured facts. Finally, in block 306, that of the searchable index is generated using the plurality of initial tokens. The generated index includes the initial tokens in combination with one or more pointers to one or more of the stored initial data records and / or fields, from whose initial field values the corresponding initial token was generated. Furthermore, the initial tokens in the index each have one or more of the token assignments established as secured facts). determining that a given key-value pair of the plurality of key-value pairs is a correct key-value pair based on the given key-value pair having a value token matching a type that corresponds to a type for a key token of the given key-value pair; ([pg 3 - For example, the learning module detects a dependency between two or more tokens which are repeatedly arranged together in the same form in the data records. If a predefined criterion is met, for example if the number of repetitions exceeds a predefined threshold value, the learning module can introduce a new token type in order to take this dependency into account in the index. For example, the learning module defines one of the tokens as a token type for the other token or tokens of the corresponding arrangement. This additional token assignment is classified as a preliminary assumption. For example, a plurality of records includes the phrase “City XY”. According to embodiments, the learning module introduces the additional token type “city” and assigns the token “XY” to this additional token type “city”.); determining that the key-value pair is a correct key-value pair based on the determining that the key-value pair has the value token matching the type of expected information associated with the key token (pg 3 n addition, the learning module can be configured to assign additional tokens already included in the index to the additional token type “city”. According to execution forms, the learning module checks whether further tokens already included in the index are to be assigned to the additional token type. For this purpose, the learning module analyzes, for example, the index and / or the underlying data records. For example, the learning module analyzes all data records that include the additional token type as a token, eg the “city” token. For example, the learning module assigns all tokens to the additional token type which can be found in the data records in the same or a comparable arrangement relative to the new token type, all of them, for example the “city” token. For example, the phrase “City YZ” can be found in one or more of the data records. In this case, for example, the token “YZ” is also assigned to the token type “City” as a preliminary assumption. According to embodiments, tokens are assigned to the additional token type, which are in the same and / or a comparable arrangement in the data records can be found with a frequency which exceeds a predefined threshold value. According to embodiments, this threshold value for the assignment is less than or equal to the threshold value for the addition of an additional token type).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Reisswig to include determining that a key-value pair of the plurality of key-value pairs has a value token matching a type of expected information with a key token of the key-value pair, wherein the type of expected information is determined based on a classification of the key token; determining that the key-value pair is a correct key-value pair based on the determining that the key-value pair has the value token matching the type of expected information associated with the key token as taught by Komarov. It would be advantageous to improve accuracy of storing key value pairs as taught by Komarov [pg 2].
Reisswig in view of Komarov does not explicitly teach and a type for each of the tokens in each key-value pair of the plurality of key- value pairs;
Duta teaches and a type for each of the tokens in each key-value pair of the plurality of key- value pairs; ([0017] In some examples, a token may include both a key and a value that should be separated. For example, a form may include “Name: John Doe.” The spacing may be such that the entire string, “Name: John Doe”, is considered a token by the tokenizer 150. To account for these types of tokens, separators may be used to split the token into a key and the key's corresponding value. In another example, the tokenizer 150 may take into account separators to separate the initial token into multiple tokens when generating tokens.)
Accordingly, it would have been obvious to one of ordinary skill in the art before the
effective filing date of the claimed invention to have modified the teachings of Reisswig in view of Komarov to include and a type for each of the tokens in each key-value pair of the plurality of key- value pairs; as taught by Duta. It would be advantageous since it allows the correct keys to be paired with the correct values as seen in the cited sections of Duta.
Regarding claim 2, Reisswig in view of Komarov and Duta teaches The method of Claim 1, Reisswig further teaches further comprising: identifying a set of tokens of the plurality of tokens, wherein each token of the set of tokens is of a same type, wherein generating the plurality of key-value pairs is further based on the identifying and the key-value pairs do not comprise tokens of the set of tokens ([0080] FIGS. 3A-3D are diagrams 301-304 illustrating processing of a document 310 according to an example of the disclosed technologies. To avoid clutter, reference designators for common items are not duplicated in diagrams 301-304. Diagram 301 of FIG. 3A illustrates the raw document 310, which can be an invoice for purchase of scientific equipment, in this case two glass slides and one microscope, from a provider “ABC Supplies.” Individual tokens can be extracted for text such as “ABC,” “Supplies,” “Description,” “Glass,” “Slides,” and so forth. Furthermore, certain text tokens (e.g. “Date,” “Price”) can be recognized as named entities using a named entity recognition tool. An auxiliary table can be used to store pairs of (named entity, original document text) to allow retrieval of the original document text at a later stage of document processing).
Regarding claim 3, Reisswig in view of Komarov and Duta teaches The method of Claim 1, Reisswig further teaches further comprising: determining that the value token comprises a numerical value within a range, wherein determining the correct key-value pair is further based on the value token comprising the numerical value within the range ([0083] Turning to diagram 304 of FIG. 3D, the following relationships can be identified at a second level of graph extraction. Provider address 332 can be identified as a property of Provider Name 331, as illustrated with a unidirectional arrow, and Line Item 336 can be identified as a value of key table header 335. Concurrently, original token “2,” composite token “Glass slides,” and original token “$22” can be identified as another line item 341. At a third level of graph extraction, Line Item 341 can also be identified as another value of the key table header 335. Then, Table Header 335 can be grouped with its two line items 341, 336 to form a third level composite token, Table 342. Finally, at a fourth level of graph extraction, the Table Total 338 can be linked as a property of Table 342).
Regarding claim 4, Reisswig in view of Komarov and Duta teaches The method of Claim 1, Reisswig further teaches further comprising: determining a first key-value pair of the plurality of key-value pairs and a second key-value pair of the plurality of key-value pairs, wherein a first token is common to the first key-value pair and the second key-value pair; determining that a second token of the first key-value pair comprises a same horizontal coordinate as the first token; determining that a third token of the second key-value pair comprises a same vertical coordinate as the first token; and determining that one of the second token or the third token matches the type of expected information ([0081] Diagram 302 of FIG. 3B shows the document 310 after preprocessing and one level of graph extraction. For purpose of illustration, named entities are prefixed with “F:” (e.g. “F:Date:”) for text fields, or with “N:” (e.g. “N:01/01/2020”) for numeric fields. Various arrows in FIG. 3B illustrate some relationships that can be determined at this stage. Numerous unidirectional arrows identify key-value relationships. These include date token “N:01/01/2020” as a value of key “F:Date,”, “12345” as a value of key “F:Invoice No.:”, “1” and “2” as distinct values for a same key “F:Qty,” “$22” and “$100” as distinct values of key “F:Price,” “Microscope” as a value of key “F:Description,” and “$122” as a value of key “F:Total.” Additionally, several bidirectional arrows identify peer relationships between respective pairs fields. These include identification of simple tokens “ABC” and “Supplies” as constituents of a larger composite token, and similarly “Glass” and “Slides.” A few peer relationships extend to cyclic or fully connected groups, such as “F:Qty,” “F:Description,” and “F:Price.” Another cycle is formed by “1,” “Microscope,” and “$100,” which are constituents of a single line item record. Finally, the provider address tokens “101,” “Heisenberg,” “Rd,” “Home,” and “Town” can be determined as fully connected, which is indicated by the dash-dot line in diagram 302 for simplicity of illustration. Unidirectional and bidirectional arrows or relationships can correspond to directed and undirected edges of a structure graph).
Regarding claim 5, Reisswig in view of Komarov and Duta teaches The method of Claim 4, Reisswig further teaches wherein the first token is the key token and the one of the second token or the third token is the value token ([0081] Diagram 302 of FIG. 3B shows the document 310 after preprocessing and one level of graph extraction. For purpose of illustration, named entities are prefixed with “F:” (e.g. “F:Date:”) for text fields, or with “N:” (e.g. “N:01/01/2020”) for numeric fields. Various arrows in FIG. 3B illustrate some relationships that can be determined at this stage. Numerous unidirectional arrows identify key-value relationships. These include date token “N:01/01/2020” as a value of key “F:Date,”, “12345” as a value of key “F:Invoice No.:”, “1” and “2” as distinct values for a same key “F:Qty,” “$22” and “$100” as distinct values of key “F:Price,” “Microscope” as a value of key “F:Description,” and “$122” as a value of key “F:Total.” Additionally, several bidirectional arrows identify peer relationships between respective pairs fields. These include identification of simple tokens “ABC” and “Supplies” as constituents of a larger composite token, and similarly “Glass” and “Slides.” A few peer relationships extend to cyclic or fully connected groups, such as “F:Qty,” “F:Description,” and “F:Price.” Another cycle is formed by “1,” “Microscope,” and “$100,” which are constituents of a single line item record. Finally, the provider address tokens “101,” “Heisenberg,” “Rd,” “Home,” and “Town” can be determined as fully connected, which is indicated by the dash-dot line in diagram 302 for simplicity of illustration. Unidirectional and bidirectional arrows or relationships can correspond to directed and undirected edges of a structure graph).
Regarding claim 6, Reisswig in view of Komarov and Duta teaches The method of Claim 1, Reisswig further teaches further comprising determining a classification for at least one token of the plurality of tokens, wherein the at least one token of the plurality of tokens comprises the key token ([0081] Diagram 302 of FIG. 3B shows the document 310 after preprocessing and one level of graph extraction. For purpose of illustration, named entities are prefixed with “F:” (e.g. “F:Date:”) for text fields, or with “N:” (e.g. “N:01/01/2020”) for numeric fields. Various arrows in FIG. 3B illustrate some relationships that can be determined at this stage. Numerous unidirectional arrows identify key-value relationships. These include date token “N:01/01/2020” as a value of key “F:Date,”, “12345” as a value of key “F:Invoice No.:”, “1” and “2” as distinct values for a same key “F:Qty,” “$22” and “$100” as distinct values of key “F:Price,” “Microscope” as a value of key “F:Description,” and “$122” as a value of key “F:Total.” Additionally, several bidirectional arrows identify peer relationships between respective pairs fields. These include identification of simple tokens “ABC” and “Supplies” as constituents of a larger composite token, and similarly “Glass” and “Slides.” A few peer relationships extend to cyclic or fully connected groups, such as “F:Qty,” “F:Description,” and “F:Price.” Another cycle is formed by “1,” “Microscope,” and “$100,” which are constituents of a single line item record. Finally, the provider address tokens “101,” “Heisenberg,” “Rd,” “Home,” and “Town” can be determined as fully connected, which is indicated by the dash-dot line in diagram 302 for simplicity of illustration. Unidirectional and bidirectional arrows or relationships can correspond to directed and undirected edges of a structure graph).
Regarding claim 7, Reisswig in view of Komarov and Duta teaches The method of Claim 6, Reisswig further teaches wherein determining the classification for the at least one token of the plurality of tokens comprises predicting, by a machine learning model, that the at least one token is associated with the classification [0100] Diagram 610 shows integration of transformer 610 in an application, coupled to receive an input record 605 and output an output record 695. The input record 605 can be similar to any of input records 161, 260, 510, 520, 530, or 540 described herein, while output record 695 can be similar to any of output records 169, 270, 510, 524, 534, or 544 described herein. In some examples, input record 610 can include respective features representing a number of tokens of a source text document, or a greater or smaller number of features, as composite tokens are added, or lower level tokens are removed from a previous level's input record. The number of tokens in the input record 605 is indicated as N in diagram 601. In some examples, a transformer can have a data path of uniform width, and outputs 615, 695 can also be organized as N vectors.)
Regarding claim 8, Reisswig in view of Komarov and Duta teaches The method of Claim 7, Reisswig further teaches wherein predicting, by the machine learning model, that the at least one token is associated with the classification comprises: determining a probability that each token of the plurality of tokens belongs to each respective classification of a plurality of classifications; and matching each token of the at least one token to the classification based on a highest probability([0101] An additional output layer 690 can be coupled to the output 615 of transformer 610 to format the output 615 into the required classification labels. Output layer 690 can perform conversions between triangular and rectangular output arrays, or thresholding, so that a weighted output component in output 615 below a threshold, can be set to zero (no relationship) in the output record 695).
Regarding claim 9, Reisswig in view of Komarov and Duta teaches The method of Claim 6, Reisswig further teaches further comprising determining whether each token of the plurality of tokens is associated with another token, wherein determining the classification for the at least one token of the plurality of tokens from the document comprises predicting a classification for each token of the plurality of tokens that is not associated with another token([0101] An additional output layer 690 can be coupled to the output 615 of transformer 610 to format the output 615 into the required classification labels. Output layer 690 can perform conversions between triangular and rectangular output arrays, or thresholding, so that a weighted output component in output 615 below a threshold, can be set to zero (no relationship) in the output record 695).
Regarding claim 10, Reisswig in view of Komarov and Duta teaches The method of Claim 1, Reisswig further teaches wherein the two tokens of each key-value pair comprise at least one token that comprises a text string ([0065] A “token” is a data item representing one or more portions of a body of text such as a document. Commonly, a “simple token” can represent a word or other sequence of printable characters delimited by whitespace or punctuation, however this is not a requirement. Some tokens (e.g. dates or hyphenated names) can represent character strings which include certain punctuation symbols. Some tokens can represent a named entity recognized in document text. Some tokens can represent combinations of words or other smaller tokens. In some examples, two or more word tokens (e.g. “DEF” and “Laboratory”) can be aggregated to form a “composite token” representing a complete Provider Name (“DEF Laboratory”), as described further herein).
Regarding claim 11, Reisswig in view of Komarov and Duta teaches The method of Claim 1, Reisswig further teaches further comprising providing the classification and the correct key-value pair to a user (Figure 10, 1030 Store Values of the Vertices defined by content of the text document in the database)
Claims 12-19 are rejected using similar reasoning seen in the current rejection of claims 1-11 due to reciting similar limitations but directed towards a processing system.
Regarding claim 20, Reisswig teaches A method for extracting information, comprising: receiving a document comprising a plurality of tokens, wherein each token in the plurality of tokens has a horizontal coordinate and a vertical coordinate ([0080] FIGS. 3A-3D are diagrams 301-304 illustrating processing of a document 310 according to an example of the disclosed technologies. To avoid clutter, reference designators for common items are not duplicated in diagrams 301-304. Diagram 301 of FIG. 3A illustrates the raw document 310, which can be an invoice for purchase of scientific equipment, in this case two glass slides and one microscope, from a provider “ABC Supplies.” Individual tokens can be extracted for text such as “ABC,” “Supplies,” “Description,” “Glass,” “Slides,” and so forth. Furthermore, certain text tokens (e.g. “Date,” “Price”) can be recognized as named entities using a named entity recognition tool. An auxiliary table can be used to store pairs of (named entity, original document text) to allow retrieval of the original document text at a later stage of document processing); determining a classification for each token of the plurality of tokens ([0074] In some deployments, a plurality of ML classifiers can be trained for respective classes of documents (e g manufacturing tracking sheets, inventory reports, orders, shipping logs, quality inspection reports, and so forth), and a received document can be determined to belong to a particular class, and the appropriate ML classifier can be invoked. In further examples, 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. The recognized fields can be used to populate one or database records with content from the input document. Alternatively or additionally, the structure graph can be stored, or returned to a requesting client); generating a plurality of key-value pairs based on the horizontal coordinate of each token of the plurality of tokens and the vertical coordinate of each token of the plurality of tokens, wherein each key-value pair comprises two tokens of the plurality of tokens with a same horizontal coordinate or a same vertical coordinate ([0081] Diagram 302 of FIG. 3B shows the document 310 after preprocessing and one level of graph extraction. For purpose of illustration, named entities are prefixed with “F:” (e.g. “F:Date:”) for text fields, or with “N:” (e.g. “N:01/01/2020”) for numeric fields. Various arrows in FIG. 3B illustrate some relationships that can be determined at this stage. Numerous unidirectional arrows identify key-value relationships. These include date token “N:01/01/2020” as a value of key “F:Date,”, “12345” as a value of key “F:Invoice No.:”, “1” and “2” as distinct values for a same key “F:Qty,” “$22” and “$100” as distinct values of key “F:Price,” “Microscope” as a value of key “F:Description,” and “$122” as a value of key “F:Total.); and providing the first correct key-value pair and the second correct key-value pair to a component of an application associated with the document(Figure 10, 1030 Store Values of the Vertices defined by content of the text document in the database).
Reisswig does not explicitly teach determining a first correct key-value pair of the plurality of key-value pairs based on the first correct key-value pair comprising a matched token that matches a type associated with the first correct key-value pair; determining an incorrect key-value pair of the plurality of key-value pairs based on the incorrect key-value pair comprising a token that does not match the type associated with the first correct key-value pair; identifying a first token of the plurality of tokens that is not associated with a correct key-value pair; generating a second correct key-value pair based on the first token and a classification assigned to the
Komarov teaches determining that a key-value pair of the plurality of key-value pairs has a value token matching a type of expected information with a key token of the key-value pair, wherein the type of expected information is determined based on a classification of the key token ([pg 3 - For example, the learning module detects a dependency between two or more tokens which are repeatedly arranged together in the same form in the data records. If a predefined criterion is met, for example if the number of repetitions exceeds a predefined threshold value, the learning module can introduce a new token type in order to take this dependency into account in the index. For example, the learning module defines one of the tokens as a token type for the other token or tokens of the corresponding arrangement. This additional token assignment is classified as a preliminary assumption. For example, a plurality of records includes the phrase “City XY”. According to embodiments, the learning module introduces the additional token type “city” and assigns the token “XY” to this additional token type “city”.); determining that the key-value pair is a correct key-value pair based on the determining that the key-value pair has the value token matching the type of expected information associated with the key token (pg 3 n addition, the learning module can be configured to assign additional tokens already included in the index to the additional token type “city”. According to execution forms, the learning module checks whether further tokens already included in the index are to be assigned to the additional token type. For this purpose, the learning module analyzes, for example, the index and / or the underlying data records. For example, the learning module analyzes all data records that include the additional token type as a token, eg the “city” token. For example, the learning module assigns all tokens to the additional token type which can be found in the data records in the same or a comparable arrangement relative to the new token type, all of them, for example the “city” token. For example, the phrase “City YZ” can be found in one or more of the data records. In this case, for example, the token “YZ” is also assigned to the token type “City” as a preliminary assumption. According to embodiments, tokens are assigned to the additional token type, which are in the same and / or a comparable arrangement in the data records can be found with a frequency which exceeds a predefined threshold value. According to embodiments, this threshold value for the assignment is less than or equal to the threshold value for the addition of an additional token type).
Komarov teaches generating, using an embedding model that is trained to generate embedding representations of documents, an embedding representation of the document; providing the embedding representation of the document as input to a classifier machine learning model, wherein the classifier machine learning model is trained to generate an output based on the embedding representation of the document, the output comprising: (FIG. 5 shows a flow diagram of an embodiment of an exemplary method for pretraining the learning module. In block 300, a plurality of initial data records are provided. These initial data sets are stored in the first data model, for example a document-oriented data model, by the multi-model database management system. In block 302, a plurality of initial tokens are generated from initial field values of the initial data records. In block 304, these initial tokens are each assigned to one or more initial token types, with all initial token assignments being defined as secured facts. Finally, in block 306, that of the searchable index is generated using the plurality of initial tokens. The generated index includes the initial tokens in combination with one or more pointers to one or more of the stored initial data records and / or fields, from whose initial field values the corresponding initial token was generated.Furthermore, the initial tokens in the index each have one or more of the token assignments established as secured facts). determining that a given key-value pair of the plurality of key-value pairs is a correct key-value pair based on the given key-value pair having a value token matching a type that corresponds to a type for a key token of the given key-value pair; ([pg 3 - For example, the learning module detects a dependency between two or more tokens which are repeatedly arranged together in the same form in the data records. If a predefined criterion is met, for example if the number of repetitions exceeds a predefined threshold value, the learning module can introduce a new token type in order to take this dependency into account in the index. For example, the learning module defines one of the tokens as a token type for the other token or tokens of the corresponding arrangement. This additional token assignment is classified as a preliminary assumption. For example, a plurality of records includes the phrase “City XY”. According to embodiments, the learning module introduces the additional token type “city” and assigns the token “XY” to this additional token type “city”.); determining that the key-value pair is a correct key-value pair based on the determining that the key-value pair has the value token matching the type of expected information associated with the key token (pg 3 n addition, the learning module can be configured to assign additional tokens already included in the index to the additional token type “city”. According to execution forms, the learning module checks whether further tokens already included in the index are to be assigned to the additional token type. For this purpose, the learning module analyzes, for example, the index and / or the underlying data records. For example, the learning module analyzes all data records that include the additional token type as a token, eg the “city” token. For example, the learning module assigns all tokens to the additional token type which can be found in the data records in the same or a comparable arrangement relative to the new token type, all of them, for example the “city” token. For example, the phrase “City YZ” can be found in one or more of the data records. In this case, for example, the token “YZ” is also assigned to the token type “City” as a preliminary assumption. According to embodiments, tokens are assigned to the additional token type, which are in the same and / or a comparable arrangement in the data records can be found with a frequency which exceeds a predefined threshold value. According to embodiments, this threshold value for the assignment is less than or equal to the threshold value for the addition of an additional token type).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Reisswig to include determining that a key-value pair of the plurality of key-value pairs has a value token matching a type of expected information with a key token of the key-value pair, wherein the type of expected information is determined based on a classification of the key token; determining that the key-value pair is a correct key-value pair based on the determining that the key-value pair has the value token matching the type of expected information associated with the key token as taught by Komarov. It would be advantageous to improve accuracy of storing key value pairs as taught by Komarov [pg 2].
Reisswig in view of Komarov does not explicitly teach and a type for each of the tokens in each key-value pair of the plurality of key- value pairs;
Duta teaches and a type for each of the tokens in each key-value pair of the plurality of key- value pairs; ([0017] In some examples, a token may include both a key and a value that should be separated. For example, a form may include “Name: John Doe.” The spacing may be such that the entire string, “Name: John Doe”, is considered a token by the tokenizer 150. To account for these types of tokens, separators may be used to split the token into a key and the key's corresponding value. In another example, the tokenizer 150 may take into account separators to separate the initial token into multiple tokens when generating tokens.)
Accordingly, it would have been obvious to one of ordinary skill in the art before the
effective filing date of the claimed invention to have mo