DETAILED ACTION
This action is responsive to communications filed on February 6, 2024. This action is made Non-Final.
Claims 1-20 are pending in the case.
Claims 1, 12, and 17 are independent claims.
Claims 1-20 are rejected.
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 .
Information Disclosure Statement
The information disclosure statement (IDS(s)) submitted on 02/06/2024 is/are in compliance with the provisions of 37 C.F.R. 1.97. Accordingly, the IDS(s) is/are being considered by the examiner.
Double Patenting
The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13.
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/patents/apply/applying-online/eterminal-disclaimer.
Claim 1-20 rejected on the ground of nonstatutory double patenting as being unpatentable over claims of U.S. Patent No. 11,928,423 indicated (via bolding) as follows:
18/433,697
Patent No. 11,928,423
1. A method, comprising:
receiving an organization identifier and description associated with a transaction,
wherein the description comprises encoded bank details of the transaction; retrieving an entity embedding comprising a vector for each entity of an
organization based on the organization identifier;
invoking a machine learning model with the entity embedding and the description, wherein the machine learning model is trained to infer a transaction embedding from the
description and compute a similarity score between the transaction embedding and each
vector of the entity embedding; and
returning a candidate entity with a similarity score that satisfies a threshold
1. (Currently Amended) A method, comprising: receiving an organization identifier and description associated with a transaction; retrieving an entity embedding comprising a vector for each entity of an organization based on the organization identifier
...
invoking a machine learning model with the entity embedding and the description, wherein the machine learning model is trained to infer a transaction embedding from the description and compute a similarity score between the trans
PNG
media_image1.png
10
7
media_image1.png
Greyscale
action embedding and each vector of the
PNG
media_image2.png
10
8
media_image2.png
Greyscale
entity embedding; and returning a candidate entity with a similarity score that satisfies a threshold.
2. The method of claim 1, wherein returning the candidate entity further comprises returning the candidate entity with the highest similarity score that satisfies the threshold
2. (Currently Amended) The method of claim 1, wherein returning the candidate entity further comprises returning the candidate entity with the a highest similarity score that satisfies the threshold.
3. The method of claim 1, further comprising filling a field in a graphical user
interface associated with the transaction with the candidate entity.
3. (Original) The method of claim 1, further comprising filling a field in a graphical user interface associated with the transaction with the candidate entity.
4. The method of claim 1, further comprising creating positive labeled training data
comprising historical user feedback including a user-selected entity for a transaction and negative
labeled training data comprising historical user feedback and a randomly selected entity different
from the user-selected entity.
4. (Original) The method of claim 1, further comprising creating positive labeled training data comprising historical user feedback including a user-selected entity for a transaction and negative labeled training data comprising historical user feedback and a randomly selected entity different from the user-selected entity.
5. The method of claim 4, further comprising preprocessing the positive and negative
labeled training data by computing a relevance score of one or more keywords in the description
and removing at least one of the one or more keywords with the lowest relevance score and
removing at least one of the one or more keywords with the highest relevance score.
5. (Currently Amended) The method of claim 4, further comprising preprocessing the positive labeled training data and the negative labeled training data by computing a relevance score of one or more keywords in the description and removing at least one of the one or more keywords with the alowest relevance score and removing at least one of the one or more keywords with thea_highest relevance score.
6. The method of claim 5, further comprising training a language model with the
positive and negative labeled training data.
6. (Currently Amended) The method of claim 5, further comprising training a language model with the positive labeled training data and the negative labeled training data.
7. The method of claim 6, wherein training the language model comprises updating a
pre-existing language model, wherein the pre-existing language model is a bidirectional encoder
representations from transformers (BERT) model.
7. (Original) The method of claim 6, wherein training the language model comprises updating a pre-existing language model, wherein the pre-existing language model is a bidirectional encoder representations from transformers (BERT) model.
8. The method of claim 6, further comprising:
determining the smallest size the entity embedding can be without reducing model
performance below a performance threshold; and
constraining the model to utilize the smallest size of the entity embedding.
8. (Currently Amended) The method of claim 6, further comprising:determining the a smallest size the entity embedding can be without reducing model performance below a performance threshold; and constraining the machine learning model to utilize the smallest size of the entity embedding.
9. detecting addition of a new entity;
determining a vector for the new entity; and
adding the vector to the entity embedding associated with the organization
identifier.
9. (Original) The method of claim 1, further comprising:detecting addition of a new entity; determining a vector for the new entity; and adding the vector to the entity embedding associated with the organization identifier.
10. The method of claim 1, further comprising:
encoding the entity embedding for each entity of the organization; and
saving the entity embedding in a key-value store with the organization identifier as
the key.
10. (Currently Amended) The method of claim 1, further comprising:encoding the entity embedding for each entity of the organization; and saving the entity embedding in a key-value store with the organization identifier as the akey.
11. The method of claim 10, wherein saving the entity embedding further comprises
decoupling data creation from storage by publishing the entity embedding to a stream with an
event streaming platform that manages storage and distribution of data streams.
1. ... from a data store, wherein the entity embedding is saved to the data store after being published to a stream with a subscriber storage application that enables stream processing and data storage in accordance with a publish/subscribe model ...
11. The method of claim 10, wherein saving the entity embedding further comprises decoupling data creation from storage by publishing the entity embedding to the stream with the subscriber storage application that receives the entity embedding and stores the entity embedding in the ke-value store in accordance with publish/subscribe model
12. ... receive an organization identifier and description associated with a transaction,
wherein the description comprises encoded bank details of the transaction;
retrieve an entity embedding comprising a vector for each entity of an organization
based on the organization identifier;
invoke a machine learning model with the entity embedding and the description,
wherein the machine learning model is trained to infer a transaction embedding from the
description and compute a similarity score between the transaction embedding and each
vector of the entity embedding; and
return a candidate entity with a similarity score that satisfies a threshold.
12. ... retrieve an entity embedding comprising a vector for each entity of an organization based on an organization identifier from a data store in response to receipt of the organization identifier and description associated with a transaction, ... model;
invoke a machine learning model with the entity embedding and the description, wherein the machine learning model is trained to infer a transaction embedding from the description and compute a similarity score between the transaction embedding and each vector of the entity embedding; and return a candidate entity with a similarity score that satisfies a threshold.
13. The system of claim 12 wherein returning the candidate entity further comprises
returning the candidate entity with the highest similarity score that satisfies the threshold.
13. (Currently Amended) The system of claim 12 wherein returning the candidate entity further comprises returning the candidate entity with the a highest similarity score that satisfies the threshold.
14. The system of claim 12, further comprising automatically filling a field in a
graphical user interface associated with the transaction with the candidate entity.
14. (Original) The system of claim 12, further comprising automatically filling a field in a graphical user interface associated with the transaction with the candidate entity.
15. The system of claim 12, wherein the organization is a business, the entity is a payee,
and the transaction is a bank transaction.
15. (Original) The system of claim 12, wherein the organization is a business, the entity is a payee, and the transaction is a bank transaction.
16. The system of claim 12, wherein retrieve the entity embedding further comprises
retrieve the entity embedding from a key-value store using the organization identifier as the key to
look up an organization-specific entity embedding.
16. (Currently Amended) The system of claim 12, wherein retrieve the entity embedding further comprises retrieve the entity embedding from a key-value store using the organization identifier as the a key to look up an organization-specific entity embedding.
17. A method of training a language model, comprising:
receiving positive labeled training data comprising historical user feedback including a
user-selected entity for a transaction with an processing-system encoded description and negative
labeled training data comprising historical user feedback and a randomly selected entity different
from the user-selected entity;
training a language model with the positive and negative labeled training data to generate
embeddings;
invoking the language model on a set of entities associated with an organization generating
entity embeddings; and
saving the entity embeddings to a key-value store, wherein an organization identifier is the
key.
17. (Currently Amended) A training method, comprising:receiving positive labeled training data comprising historical user feedback including a user-selected entity for a transaction and negative labeled training data comprising historical user feedback and a randomly selected entity different from the user- selected entity; training a language model with the positive labeled training data and the negative labeled training data to generate embeddings; invoking the language model on a set of entities associated with an organization generating entity embeddings; and saving the entity embeddings to a key-value store, wherein an organization identifier is the a key...
18. The method of claim 17, further comprising preprocessing the positive and negative
labeled training data by computing a relevance score of one or more keywords and removing at
least one of the one or more keywords with the lowest relevance score and removing at least one
of the one or more keywords with the highest relevance score.
18. (Currently Amended) The method of claim 17, further comprising preprocessing the positive labeled training data and the negative labeled training data by computing a relevance score of one or more keywords and removing at least one of the one or more keywords with the a lowest relevance score and removing at least one of the one or more keywords with the a highest relevance score.
19. determining the smallest size an entity embedding can be without reducing model
performance; and
reducing the entity embedding size to the smallest size before saving the embedding to the
key-value store.
19. (Currently Amended) The method of claim 17, further comprising:determining the a smallest size an entity embedding can be without reducing model performance; andreducing the entity embedding size to the smallest size before saving the embedding to the key-value store.
20. The method of claim 17, where training the language model comprises updating a
pre-existing natural language model with the training data.
20. (Currently Amended) The method of claim 17, where training the language model comprises updating a pre-existing natural language model with the training data ...
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) 1-3 and 12-15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Rusu et al., US Publication 2025/0390961 (“Rusu”), and further in view of Patil et al., US Publication 2019/0272482 (“Patil”).
Claim 1:
Rusu teaches or suggests a method, comprising:
recieving an organization identifier and description associated with a transaction, wherein the description comprises encoded bank details of the transaction (see Fig. 1; para. 0098 - record may include one or more attributes such as transaction data, payee details, a reference, a description, a transaction amount, transaction currency, and/or transaction type detail; para. 0099 – financial records may include truncated, merged or abbreviated strings and/or strings that are not natural language strings or strings that are not capable of being readily interpreted by a human.);
retrieving an entity embedding comprising a vector for each entity of an organization based on the organization identifier (see Fig. 1; para. 0098 - record may include one or more attributes such as transaction data, payee details, a reference, a description, a transaction amount, transaction currency, and/or transaction type detail; para. 0099 – financial records may include truncated, merged or abbreviated strings and/or strings that are not natural language strings or strings that are not capable of being readily interpreted by a human; para. 0120 - data to be vectorised, such as the financial records, account codes, entity names and/or entity attributes, may comprise multiple data strings or words and the numerical representation of the entire data; para. 0126 - data to be processed is provided to the numerical representation generation model 204. For example, the data to be processed may comprise transaction data, such as financial records, entity names, account codes, entity attribute data, such as entity type, entity geography, and/or entity industry, for example. Any of this data may be taken alone, or in combination with other data.);
invoking a machine learning model with the entity embedding and the description ... and compute a similarity score between the transaction embedding and each vector of the entity embedding (see Fig. 1, 2, 9, and 10; para. 0107 - determine a similarity measure or a confidence score by comparing pairs of the numerical representation of the candidate financial records and ones of the numerical representations of account codes with the numerical representations generated for the transaction data and associated account codes during training. Determine a similarity measure or a confidence score indicating the suitability of an account code for a specific financial record; para. 0108 - provide the determined at least one first transaction attribute associated with the candidate financial record to a client device for presentation on a user interface; para. 0109 - first and further attributes (for example, an account code recommendation and/or entity name recommendation for the financial record. In embodiments, where more than one account code and entity name is predicted and recommended to the user for a particular financial record, then the received user input may comprise a designation of a specific one of the recommended account codes and entity names; para. 0111 - based on confidence scores associated with determined transaction attributes. For example, the accounting system 110 may be configured to determine a confidence score associated with an entity name and/or account code recommendation and responsive to the confidence score meeting a threshold value; para. 0112 - numerical representation generation model 204 is configured to determine a numerical representations of inputs, such as candidate financial records, reconciled financial records and other transaction data, including accounting codes, entity names relating to transactions, and other entity attribute information such as entity type, entity geography and/or entity industry; para. 0120 - data to be vectorised, such as the financial records, account codes, entity names and/or entity attributes, may comprise multiple data strings or words and the numerical representation of the entire data; para. 0126 - data to be processed is provided to the numerical representation generation model 204. For example, the data to be processed may comprise transaction data, such as financial records, entity names, account codes, entity attribute data, such as entity type, entity geography, and/or entity industry, for example. Any of this data may be taken alone, or in combination with other data.); and
returning candidate entity with a similarity score that satisfies a threshold (see Fig. 1, 2, 9, and 10; para. 0107 - determine a similarity measure or a confidence score by comparing pairs of the numerical representation of the candidate financial records and ones of the numerical representations of account codes with the numerical representations generated for the transaction data and associated account codes during training. Determine a similarity measure or a confidence score indicating the suitability of an account code for a specific financial record; para. 0108 - provide the determined at least one first transaction attribute associated with the candidate financial record to a client device for presentation on a user interface; para. 0109 - first and further attributes (for example, an account code recommendation and/or entity name recommendation for the financial record. In embodiments, where more than one account code and entity name is predicted and recommended to the user for a particular financial record, then the received user input may comprise a designation of a specific one of the recommended account codes and entity names; para. 0111 - based on confidence scores associated with determined transaction attributes. For example, the accounting system 110 may be configured to determine a confidence score associated with an entity name and/or account code recommendation and responsive to the confidence score meeting a threshold value; para. 0112 - numerical representation generation model 204 is configured to determine a numerical representations of inputs, such as candidate financial records, reconciled financial records and other transaction data, including accounting codes, entity names relating to transactions, and other entity attribute information such as entity type, entity geography and/or entity industry; para. 0120 - data to be vectorised, such as the financial records, account codes, entity names and/or entity attributes, may comprise multiple data strings or words and the numerical representation of the entire data; para. 0126 - data to be processed is provided to the numerical representation generation model 204. For example, the data to be processed may comprise transaction data, such as financial records, entity names, account codes, entity attribute data, such as entity type, entity geography, and/or entity industry, for example. Any of this data may be taken alone, or in combination with other data; para. 0185 – entity prediction model 209 may be a natural language processing deep-learning model; para. 0187 - entity prediction model 209 is configured to determine a confidence score for each of the substrings and to determine the predicted entity identifier as the substring with the highest confidence score, or a position indicator of the substring with the highest confidence score; para. 0189 - predict entity identifiers associated with candidate financial records.).
Patil more specifically teaches or suggests wherein the machine learning model is trained to infer a transaction embedding from the description (see Fig. 2A, 3, and 5; para. 0028 - by using word2vec, the quantifier 108 captures the contexts in which words occur. Semantically similar words will occur in similar context and will thus have word2vec representations which are closer to each other. Such implementations, typically modelled as a feed forward neural network, largely involves matrix multiplication and is thus computationally efficient; para. 0029 - quantifier 108 can train a word2vec model on a relatively small set, e.g., several millions, of transaction records; para. 0035 - vectorizer 206 can provide a vector representation 208 of the transaction records to an instance sampler for further processing; para. 0042 - clustering module 304 represents each transaction record by an average of word2vec representation of each of its constituent words. The clustering module 304 then calculates a distance, e.g., a Euclidean distance, between each pair of these representations of the transaction records.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Rusu, to include wherein the machine learning model is trained to infer a transaction embedding from the description for the purpose of efficiently using a ml model such as word2vec to predict embeddings based on transaction description text, improving clustering and matching determinations, as taught by Patil (0028 and 0042).
Claim(s) 12:
Claim(s) 12 correspond to Claim 1, and thus, Rusu and Patil teach or suggest the limitations of claim(s) 12 as well.
Claim 2:
Rusu further teaches or suggests wherein returning the candidate entity further comprises returning the candidate entity with the highest similarity score that satisfies a threshold (see para. 0022 - determine a confidence score associated with the candidate financial record and each one of a plurality of account code identifiers associated with the first accounting entity; and determine the at least one first transaction attribute as the account code identifiers having the highest confidence score; para. 0033 - the entity identifier comprises the substring associated with the best match score; para. 0111 - based on confidence scores associated with determined transaction attributes. For example, the accounting system 110 may be configured to determine a confidence score associated with an entity name and/or account code recommendation and responsive to the confidence score meeting a threshold value; para. 0185 – entity prediction model 209 may be a natural language processing deep-learning model; para. 0187 - entity prediction model 209 is configured to determine a confidence score for each of the substrings and to determine the predicted entity identifier as the substring with the highest confidence score, or a position indicator of the substring with the highest confidence score; para. 0189 - predict entity identifiers associated with candidate financial records.).
Claim(s) 13:
Claim(s) 13 correspond to Claim 2, and thus, Rusu and Patil teach or suggest the limitations of claim(s) 13 as well.
Claim 3:
Rusu further teaches or suggests filling a field in a graphical user interface associated with the transaction with the candidate entity (see Fig. 7; para. 0108 - provide the determined at least one first transaction attribute associated with the candidate financial record to a client device for presentation on a user interface; para. 0109 - first and further attributes (for example, an account code recommendation and/or entity name recommendation for the financial record. In embodiments, where more than one account code and entity name is predicted and recommended to the user for a particular financial record, then the received user input may comprise a designation of a specific one of the recommended account codes and entity names.).
Claim(s) 14:
Claim(s) 14 correspond to Claim 3, and thus, Rusu and Patil teach or suggest the limitations of claim(s) 14 as well.
Claim 15:
Rusu further teaches or suggests wherein the organization is a business, the entity is a payee, and the transaction is a bank transaction (see Fig. 1; para. 0080 - Transaction data may comprise financial records, such as data from bank statements and/or bank feeds, and/or may comprise accounting or bookkeeping data, as maintained by the accounting system for a plurality of entities. In some embodiments, the numerical representation generation model 204 may also generate numerical representations of account code data 216 and/or entity data 214. Account code data 216 may comprise an account code identifier such as an account code name and/or an account code number. Entity data 214 may comprise information relating to entities including entity name, entity type, entity industry, a country the entity operates in, for example; para. 0098 - record may include one or more attributes such as transaction data, payee details, a reference, a description, a trans action amount, transaction currency, and/or transaction type detail; para. 0128 - in some embodiments, the specific attributes may include one or a combination of payee data, transaction reference, or transaction notes.).
Claim(s) 4 is/are rejected under 35 U.S.C. 103 as being unpatentable over Rusu, in view of Patil, and further in view of Light, et al., US Patent Application Publication no. US 2012/0036130 (“Light”).
Claim 4:
Light further teaches or suggests comprising positive labeled training data comprising historical user feedback including a user-selected entity for a transaction and negative labeled training data comprising historical user feedback and a randomly selected entity different from the user selected entity (see para. 0043 - parameters of the model are estimated from a corpus of training data, that is, text where a human has annotated all entity mentions or occurrences; para. 0072 - the classification of randomly selected sentences from candidate pool; para. 0078 - taking positive example sentences from classification phase and manually generating appropriate template records. The user is automatically presented with all possible templates which could be generated from the sentence and asking the user to select the one that is correct; para. 0082 - select five example sentences from each bucket randomly and mark them as either positive or negative examples. score randomly selected sentences until 500 examples of each time are identified.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Rusu, to include comprising positive labeled training data comprising historical user feedback including a user-selected entity for a transaction and negative labeled training data comprising historical user feedback and a randomly selected entity different from the user selected entity for the purpose of efficiently preprocessing transaction training data based on examples, improving entity extraction and resolution, as taught by Light (0082-0084).
Claim(s) 5, 6, and 7 is/are rejected under 35 U.S.C. 103 as being unpatentable over Rusu, in view of Patil, in view of Light, and further in view of Lee et al., US Patent no. US 8,589,399 (“Lee”).
Claim 5:
As indicated above, the combination of Rusu, Patil, and Light teach or suggest preprocessing the positive and negative labeled training data.
Rusu does not explicitly disclose by computing a relevance score of one or more keywords and removing at least one of the one or more keywords with the lowest relevance score and removing at least one of the one or more keywords with the highest relevance score.
Lee teaches or suggests by computing a relevance score of one or more keywords and removing at least one of the one or more keywords with the lowest relevance score and removing at least one of the one or more keywords with the highest relevance score (see col. 2, lines 31-37 - terms identifying features of neighborhoods; or a database storing terms associated with the identified category. Assigning weights to each of the candidate terms in the one or more 35 resources further includes one or more of filtering candidate terms with a term frequency-inverse document frequency (TF-IDF) value below a TF-IDF value threshold; col. 3, lines 32-33 - candidate terms that are above a threshold relative frequency are eliminated; col. 7, lines 56-61 - select one or more of the terms based on the scoring (e.g., terms with the highest values) as known for terms for the restaurant. In some examples, the known for system 102 can select any term with a value above a particular threshold or a particular number of terms 60 with the highest scores. terms can be selected based, at least in part, on information about the semantic meaning; col. 10, lines 13-27 - scorer 218 can filter any candidate terms with a TF-IDF value below a lower 15 TF-IDF value threshold (e.g., below a threshold relative frequency of the number of instances of the term in resources relating to the entity compared to the number of instances of the term in resources relating to entities within a selected category), for example to remove common terms in the corpus that do not apply specifically to the entity. The term scorer 218 can also filter candidate terms with a TF-IDF value above an upper TF-IDF value threshold (e.g., above a threshold relative frequency of the number of instances of the term in resources relating to the entity compared to the number of instances of the term in a corpus or in resources relating to entities within a selected category), for example to remove terms that are too unique and/or unlikely to be of interest; col. 11, lines 10, 11 - select all such terms for an entity with a weight above a relevance weight threshold.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Rusu, to include by computing a relevance score of one or more keywords and removing at least one of the one or more keywords with the lowest relevance score and removing at least one of the one or more keywords with the highest relevance score for the purpose of efficiently refining terms relating to an entity, improving entity term relevance and usefulness, as taught by Lee (col. 5 and 10).
Claim 6:
Light further teaches or suggests training a language model with positive and negative labeled training data (see para. 0043 - parameters of the model are estimated from a corpus of training data, that is, text where a human has annotated all entity mentions or occurrences; para. 0072 - the classification of randomly selected sentences from candidate pool; para. 0078 - taking positive example sentences from classification phase and manually generating appropriate template records. The user is automatically presented with all possible templates which could be generated from the sentence and asking the user to select the one that is correct; para. 0079 - from gold data set for training
data and develop extraction; para. 0082 - efficiently collect positive and negative training instances; select five example sentences from each bucket randomly and mark them as either positive or negative examples. score randomly selected sentences until 500 examples of each time are identified.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Rusu, to include comprising positive labeled training data comprising historical user feedback including a user-selected entity for a transaction and negative labeled training data comprising historical user feedback and a randomly selected entity different from the user selected entity for the purpose of efficiently preprocessing transaction training data based on examples, improving entity extraction and resolution, as taught by Light (0082-0084).
Claim 7:
As indicated above, the combination of Rusu, Patil, Lee and Light teach or suggest wherein training the language model comprises updating a pre-existing language model.
Rusu further teaches or suggests wherein the pre-existing language model is a bidirectional encoder representations from transformers (BERT) model (see para. 0079 - numerical representation generation model 204 may generate the numerical representations using a neural network trained to generate word embeddings or vectors corresponding to each token in the transaction data. In some embodiments, the numerical representation generation model 204 may incorporate one or more language models such as the Bidirectional Encoder Representations from Transformers (BERT) language model.).
Claim(s) 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Rusu, in view of Patil, in view of Light, in view of Lee, and further in view of further in view of Zhao et al., US Patent Application Publication no. US 2023/0124258 (“Zhao”).
Claim 8:
As indicated above, Rusu, Patil, Light, and Lee teach or suggest entity embedding and before saving the embedding to the key-value store.
Rusu does not explicitly disclose determining the smallest entity size an entity embedding can be without reducing model performance below a performance threshold; and constraining the model to utilize the smallest size of the entity embedding.
Zhao teaches or suggests determining the smallest entity size an entity embedding can be without reducing model performance below a performance threshold; and constraining the model to utilize the smallest size of the entity embedding (see para. 0019 - selecting the wrong dimension for the embedding may be problematic, as an embedding that is too small will lose valuable information on the feature field, and an embedding that is too big will still waste computing resources; para. 0031 - embedding with a size greater than one would be wasting resources. On the other hand, highly informative features like the item ID on a shopping website provides a highly informative value since there may be millions of items. If the item ID is embedded in a vector that cannot accommodate all the possible values, the embedding will be too small, and the item ID feature will lose informational value; para. 0034 - how to determine what the optimal embedding size is for each feature.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Rusu, to include determining the smallest entity size an entity embedding can be without reducing model performance below a performance threshold; and constraining the model to utilize the smallest size of the entity embedding for the purpose of efficiently determining appropriate embedding sizes, improving embedding resource usage and information value, as taught by Zhao (0019).
Claim(s) 9, 10, and 16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Rusu, in view of Patil, and further in view of Bruss et al., US Patent Application Publication no. US 2020/0226460 (“Bruss”).
Claim 9:
Bruss further teaches or suggests detecting addition of a new entity; determining a vector for the new entity; and adding the vector to the entity embedding associated with the organization identifier (see para. 0019 - transaction data 121 according to one or more formats, and assigning each unique entity (e.g., customer accounts and/or merchant accounts) a unique identifier; para. 0023 - each unique entity ID (e.g., customer ID and merchant ID) in the graph 110 and/or transaction data 121 is assigned a unique identifier corresponding to a row in the lookup table, such that each unique entity is represented in the embeddings layer; para. 0044 - where a unique identifier is assigned in the embeddings layer 109 for each entity (e.g., customer accounts, merchant accounts, other types of accounts); Claim 6 - embeddings layer associates each embedding value with one of the unique identifiers, wherein the vector for the first new transaction is further determined based on a unique identifier in the transaction data for the first new transaction.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Rusu, to include detecting addition of a new entity; determining a vector for the new entity; and adding the vector to the entity embedding associated with the organization identifier for the purpose of efficiently managing relevant entity and transaction information, improving transaction analysis based on embeddings, as taught by Bruss (0003 and 0011).
Claim 10:
Bruss further teaches or suggests encoding the entity embedding for each entity of the organization; and saving the entity embedding in a key-value store with the organization identifier as the key (see para. 0019 - transaction data 121 according to one or more formats, and assigning each unique entity (e.g., customer accounts and/or merchant accounts) a unique identifier; para. 0023 - each unique entity ID (e.g., customer ID and merchant ID) in the graph 110 and/or transaction data 121 is assigned a unique identifier corresponding to a row in the lookup table, such that each unique entity is represented in the embeddings layer; para. 0024 - looks up and returns the corresponding rows of the embeddings layer; para. 0041 - uses the embeddings layer 109 as a lookup table to identify the embeddings 109 for the input vector of the first new transaction, e.g., based on the account identifier; para. 0044 - where a unique identifier is assigned in the embeddings layer 109 for each entity (e.g., customer accounts, merchant accounts, other types of accounts); Claim 6 - embeddings layer associates each embedding value with one of the unique identifiers, wherein the vector for the first new transaction is further determined based on a unique identifier in the transaction data for the first new transaction.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Rusu, to include encoding the entity embedding for each entity of the organization; and saving the entity embedding in a key-value store with the organization identifier as the key for the purpose of efficiently managing relevant entity and transaction information, improving transaction analysis based on embeddings, as taught by Bruss (0003 and 0011).
Claim 16:
Bruss further teaches or suggests retrieve the entity embedding from a key-value store using the organization identifier as the key to look up an organization-specific entity embedding (see para. 0019 - transaction data 121 according to one or more formats, and assigning each unique entity (e.g., customer accounts and/or merchant accounts) a unique identifier; para. 0023 - each unique entity ID (e.g., customer ID and merchant ID) in the graph 110 and/or transaction data 121 is assigned a unique identifier corresponding to a row in the lookup table, such that each unique entity is represented in the embeddings layer; para. 0024 - looks up and returns the corresponding rows of the embeddings layer; para. 0041 - uses the embeddings layer 109 as a lookup table to identify the embeddings 109 for the input vector of the first new transaction, e.g., based on the account identifier; para. 0044 - where a unique identifier is assigned in the embeddings layer 109 for each entity (e.g., customer accounts, merchant accounts, other types of accounts); Claim 6 - embeddings layer associates each embedding value with one of the unique identifiers, wherein the vector for the first new transaction is further determined based on a unique identifier in the transaction data for the first new transaction.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Rusu, to include retrieve the entity embedding from a key-value store using the organization identifier as the key to look up an organization-specific entity embedding for the purpose of efficiently managing relevant entity and transaction information, improving transaction analysis based on embeddings, as taught by Bruss (0003 and 0011).
Claim(s) 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Light, and further in view of Bruss.
Claim 17:
Light teaches or suggests a method of training a language model, comprising:
receiving positive labeled training data comprising historical user feedback including a user-selected entity for a transaction and negative labeled training data comprising historical feedback and a randomly selected entity different from the user-selected entity (see para. 0043 - parameters of the model are estimated from a corpus of training data, that is, text where a human has annotated all entity mentions or occurrences; para. 0072 - the classification of randomly selected sentences from candidate pool; para. 0078 - taking positive example sentences from classification phase and manually generating appropriate template records. The user is automatically presented with all possible templates which could be generated from the sentence and asking the user to select the one that is correct; para. 0082 - select five example sentences from each bucket randomly and mark them as either positive or negative examples. score randomly selected sentences until 500 examples of each time are identified.);
training a language model with positive and negative labeled training data (see para. 0043 - parameters of the model are estimated from a corpus of training data, that is, text where a human has annotated all entity mentions or occurrences; para. 0054 – trained model. After the algorithm determines the most probable sequence of tags, the text, such as tagged sentence 119, where the entities are located is passed to a resolver, such as entity resolver; para. 0070 - classifier that combines selected features with selected training methods; para. 0079 - data set for training data and develop extraction; para. 0082 - efficiently collect positive and negative training instances. into a structured template record. The template record identifies the roles the named entities and tagged phrases play in the event.);
invoking the language model on a set of entities associated with an organization generating entity information (see para. 0060 - classifier 310 and a template extractor; para. 0076 - extractor 320 extracts event templates from positively classified sentences. Extracting templates from sentences involves identifying the name entities; para. 0078 – taking positive example sentences from classification phase and manually generating appropriate template records; para. 0082 - moves identified entities from a positively classified job change event sentence into a structured template record. The template record identifies the roles the named entities and tagged phrases play in the event.); and
saving the entity information to a key-value store, wherein an organization identifier is the key (see para. 0060 - classifier 310 and a template extractor; para. 0076 - extractor 320 extracts event templates from positively classified sentences. Extracting templates from sentences involves identifying the name entities; para. 0078 – taking positive example sentences from classification phase and manually generating appropriate template records; para. 0082 - moves identified entities from a positively classified job change event sentence into a structured template record. The template record identifies the roles the named entities and tagged phrases play in the event; para. 0083 - represents a data
structure. Entity ID F56748. Value Skadden & Arps.).
Light does not explicitly disclose to generate embeddings; that the information is embeddings.
Bruss teaches or suggests to generate embeddings; that the information is embeddings; saving the entity embedding to a key-value store, wherein an organization identifier is the key (see para. 0003 - provide neural embeddings of transaction data. Neural network may be trained based on training data comprising a plurality of positive entity pairs from the network graph of transaction data and a plurality of negative entity pairs not present in the network graph of transaction data, the negative entity pairs comprising artificially generated relationships between each entity in the negative entity pair, the neural network comprising an embeddings layer; para. 0019 - transaction data 121 according to one or more formats, and assigning each unique entity (e.g., customer accounts and/or merchant accounts) a unique identifier; para. 0023 - each unique entity ID (e.g., customer ID and merchant ID) in the graph 110 and/or transaction data 121 is assigned a unique identifier corresponding to a row in the lookup table, such that each unique entity is represented in the embeddings layer; para. 0024 - looks up and returns the corresponding rows of the embeddings layer; para. 0040 - trains a first neural network 105 using an ML algorithm 104 (e.g., a neural network algorithm) based on positive entity pairs and negative entity pairs as training data 107 to learn the embeddings layer; para. 0041 - apply an embedding function (or an encoding function) to the transaction data 121, thereby generating an input vector describing the first new transaction. outputs an embedding vector describing the first new transaction. uses the embeddings layer 109 as a lookup table to identify the embeddings 109 for the input vector of the first new transaction, e.g., based on the account identifier; para. 0044 - processes the positive and negative samples to generate the embeddings. where a unique identifier is assigned in the embeddings layer 109 for each entity (e.g., customer accounts, merchant accounts, other types of accounts); Claim 6 - embeddings layer associates each embedding value with one of the unique identifiers, wherein the vector for the first new transaction is further determined based on a unique identifier in the transaction data for the first new transaction.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Light, to include generate embeddings; that the information is embeddings; saving the entity embedding to a key-value store, wherein an organization identifier is the key for the purpose of efficiently managing relevant entity and transaction information, improving transaction analysis based on embeddings, as taught by Bruss (0003 and 0011).
Claim(s) 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Light, in view of Bruss, and further in view of Lee et al., US Patent no. US 8,589,399 (“Lee”).
Claim 18:
As indicated above, Light and Bruss teach or suggest preprocessing the positive and negative labeled training data.
Light does not explicitly disclose by computing a relevance score of one or more keywords and removing at least one of the one or more keywords with the lowest relevance score and removing at least one of the one or more keywords with the highest relevance score.
Lee teaches or suggests by computing a relevance score of one or more keywords and removing at least one of the one or more keywords with the lowest relevance score and removing at least one of the one or more keywords with the highest relevance score (see col. 2, lines 31-37 - terms identifying features of neighborhoods; or a database storing terms associated with the identified category. Assigning weights to each of the candidate terms in the one or more 35 resources further includes one or more of filtering candidate terms with a term frequency-inverse document frequency (TF-IDF) value below a TF-IDF value threshold; col. 3, lines 32-33 - candidate terms that are above a threshold relative frequency are eliminated; col. 7, lines 56-61 - select one or more of the terms based on the scoring (e.g., terms with the highest values) as known for terms for the restaurant. In some examples, the known for system 102 can select any term with a value above a particular threshold or a particular number of terms 60 with the highest scores. terms can be selected based, at least in part, on information about the semantic meaning; col. 10, lines 13-27 - scorer 218 can filter any candidate terms with a TF-IDF value below a lower 15 TF-IDF value threshold (e.g., below a threshold relative frequency of the number of instances of the term in resources relating to the entity compared to the number of instances of the term in resources relating to entities within a selected category), for example to remove common terms in the corpus that do not apply specifically to the entity. The term scorer 218 can also filter candidate terms with a TF-IDF value above an upper TF-IDF value threshold (e.g., above a threshold relative frequency of the number of instances of the term in resources relating to the entity compared to the number of instances of the term in a corpus or in resources relating to entities within a selected category), for example to remove terms that are too unique and/or unlikely to be of interest; col. 11, lines 10, 11 - select all such terms for an entity with a weight above a relevance weight threshold.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Light, to include by computing a relevance score of one or more keywords and removing at least one of the one or more keywords with the lowest relevance score and removing at least one of the one or more keywords with the highest relevance score for the purpose of efficiently refining terms relating to an entity, improving entity term relevance and usefulness, as taught by Lee (col. 5 and 10).
Claim(s) 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Light, in view of Bruss, and further in view of Zhao et al., US Patent Application Publication no. US 2023/0124258 (“Zhao”).
Claim 19:
As indicated above, Light and Bruss teach or suggest entity embedding and before saving the embedding to the key-value store.
Light does not explicitly disclose determining the smallest entity size an entity embedding can be without reducing model performance; and reducing the entity embedding size to the smallest size.
Zhao teaches or suggests determining the smallest entity size an entity embedding can be without reducing model performance; and reducing the entity embedding size to the smallest size before saving (see para. 0019 - selecting the wrong dimension for the embedding may be problematic, as an embedding that is too small will lose valuable information on the feature field, and an embedding that is too big will still waste computing resources; para. 0031 - embedding with a size greater than one would be wasting resources. On the other hand, highly informative features like the item ID on a shopping website provides a highly informative value since there may be millions of items. If the item ID is embedded in a vector that cannot accommodate all the possible values, the embedding will be too small, and the item ID feature will lose informational value; para. 0034 - how to determine what the optimal embedding size is for each feature.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Light, to include determining the smallest entity size an entity embedding can be without reducing model performance; and reducing the entity embedding size to the smallest size before saving for the purpose of efficiently determining appropriate embedding sizes, improving embedding resource usage and information value, as taught by Zhao (0019).
Claim(s) 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Light, in view of Bruss, and further in view of Zhu et al., US Patent Application Publication no. US 2022/0358594 (“Zhu”).
Claim 20:
Zhu further teaches or suggests wherein training the language model comprises updating a pre-existing natural language model with the training data (see para. 0008 – can be updated based on a difference between the gradient of a loss function of the first machine learning language model and the gradient of a loss function of the second machine learning language model trained with the selected counterfactual data; para. 0036 - Training a model, augmenting with counter-factual information, and updating the trained model are shown at 108. In an embodiment, a processor may train a neural network 110 such as a BERT model. E-net herein refers to an earnings call transcript network trained using a language model such as BERT. Briefly BERT (Bidirectional Encoder Representations from Transformers) is a machine learning language model for natural language processing (NLP). The neural network 110, e.g. , a BERT model, can learn language context of the earnings call data 104 and can be fine-tuned with a classification layer to predict a stock price or another financial indicator based on the language context of the earnings call data and the market data 102. For example, market data 102 and earnings call data 104 are used as ground truth data for correlating. data can pertain to a particular entity; para. 0037 - can be updated based on augmenting part of the training data; para. 0045 - data sources 206, with which the model may have been trained as augmented information.).
Accordingly, it would have been obvious to one having ordinary skill before the effective filing date of the claimed invention to modify the system and method, taught in Light, to include wherein training the language model comprises updating a pre-existing natural language model with the training data for the purpose of efficiently updating a language model using specialized training data, improving prediction functionality based on language context, as taught by Zhu (0036 and 0045).
Allowable Subject Matter
Claim 11 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Andrew T McIntosh whose telephone number is (571)270-7790. The examiner can normally be reached M-Th 8:00am-5:30pm.
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, Tamara Kyle can be reached at 571-272-4241. 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.
/ANDREW T MCINTOSH/Primary Examiner, Art Unit 2144