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 Arguments
Applicant’s arguments filed on December 01, 2025 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
The 35 U.S.C. 101 rejections for claims 1-24 have been withdrawn according to the claim amendment and the argument on the Remark received on December 01, 2025.
Response to Amendment
The amendment to the claims received on December 01, 2025has been entered.
The amendment of claims 1-2, 4, 13-14 and 16 is acknowledged.
Examiner Note
The limitation in claim 1, “ integrating a plurality of weak label augmenters of different paradigms, wherein a first of the augmenters is a rule augmenter extracting first weak labels from unlabeled data, and a second of the augmenters is a label augmenter extracting second weak labels from the unlabeled data; filtering the first and second weak labels using an instance filter to update a high- precision training set shared by the plurality of augmenters; and iteratively retraining the plurality of augmenters using the updated high-precision training set to improve recognition performance over iterations” is considered as additional elements that are sufficient to amount to significantly more than the judicial exception according to the improvement as disclosed in paragraph 25.
The limitation in claim 13, “ integrate a plurality of weak label augmenters of different paradigms, wherein a first of the augmenters is a rule augmenter generating first weak labels from unlabeled data, and a second of the augmenters is a label augmenter generating second weak labels from the unlabeled data; filter the first and second weak labels using an instance filter to update a high- precision training set shared by the plurality of augmenters; and iteratively retrain the plurality of augmenters using the updated high-precision training set to improve recognition performance over iterations.” is considered as additional elements that are sufficient to amount to significantly more than the judicial exception according to the improvement as disclosed in paragraph 25.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-3, 9, 10, 13-15, 21 and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Weakly labeled data augmentation for social media named entity recognition, and further in view of Zhao’183 (US 2021/0319183) and Asif’184 (US 2022/0101184).
With respect to claim 13, Weakly labeled data augmentation for social media named entity recognition teaches a system for training a machine-learning model to perform named-entity recognition of unlabeled text data utilizing a co-augmentation framework (Fig.1), comprising:
one or more computing devices (Fig.1) configured to:
integrate a plurality of weak label augmenters of different paradigms (pages 3 and 4), wherein a first of the augmenters is a rule augmenter generating first weak labels from unlabeled data [regarding to either an alias augmentation or a typo augmentation to generate the weak label (pages 3 and 4)], and a second of the augmenters is a label augmenter generating second weak labels from the unlabeled data [regarding to either an alias augmentation or a typo augmentation to generate the weak label (pages 3 and 4)];
Weakly labeled data augmentation for social media named entity recognition does not teach filter the first and second weak labels using an instance filter to update a high- precision training set shared by the plurality of augmenters; and
iteratively retrain the plurality of augmenters using the updated high-precision training set to improve recognition performance over iterations.
Zhao’183 filter the first and second weak labels using an instance filter to update a high-precision training set shared by the plurality of augmenters [if a label corresponds highly with an annotator (e.g., matches exactly with semantics of a programming language for a function name), the label is selected to train a machine learning mode (paragraph 25 and Fig.2, step 230). Therefore, a filter is considered being disclosed to use an instance filter to selected a label corresponds highly with an annotator (e.g., matches exactly with semantics of a programming language for a function name) from the weak labels including the first weak label and the second weak label.].
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Weakly labeled data augmentation for social media named entity recognition according to the teaching of Zhao’183 to include a filter to select the weak labels including the first weak labels and the second weak labels corresponding highly with an annotator because this will allow the first weak labels and the second weak labels to be selected more effectively.
The combination of Weakly labeled data augmentation for social media named entity recognition and Zhao’183 does not teach iteratively retrain the plurality of augmenters using the updated high-precision training set to improve recognition performance over iterations.
Asif’184 teach a machine learning model is being retrained according to the new labeled (weak-labeled) feature data (paragraph 97 and Fig.6)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Weakly labeled data augmentation for social media named entity recognition and Zhao’183 according to the teaching of Asif’184 to retrain the layers of the final perdition model according the new generated labels (iteratively retrain the plurality of augmenters using the updated high-precision training set to improve recognition performance over iterations) because this will allow the semantic entities within the target domain to be recognized more effectively.
With respect to claim 14, which further limits claim 13, Weakly labeled data augmentation for social media named entity recognition teaches wherein the plurality of weak label augmenters includes a rule augmenter, and the one or more computing devices are further configured to: extract, by a rule applier of the rule augmenter, the first weak labels based on the unlabeled data using given seed rules [an alias augmentation and a typo augmentation are used to generate more weakly
labeled data (page 3)];
use the high-precision training set, as updated based on the first weak labels filtered by the instance filter, to train a neural named entity recognition (NER) model to identify predicted labels in the unlabeled data [The NER model was trained with the weakly labeled data generated (page 5)];
extract rules from the predicted labels [The NER model was trained with the weakly labeled data generated by the method described in Section 3.1, and this training procedure is referred to as pretraining (page 5). Therefore, the rules from the predicted labels are considered being extracted to perform the pretraining]; and
add selected rules from the extracted rules to enlarge the seed rules [NER trained with weakly and manually labeled data (page 5). Therefore, the rules extracted from the weakly labels are considered being selected and then to add with the rules for the manually labeled data to train the NER].
With respect to claim 15, which further claim 14, Weakly labeled data augmentation for social media named entity recognition teaches wherein the one or more computing devices are further configured to utilize the neural NER model, once trained, to perform named- entity recognition on an unlabeled input text (Fig.1).
With respect to claim 21, which further limits claim 13, Weakly labeled data augmentation for social media named entity recognition does not teach wherein the instance filter utilizes a rule-based constraint functions to remove the weak labels that match to one or more predefined constraint rules to prevent their incorporation into the high-precision training set.
Zhao’183 teaches wherein the instance filter utilizes a rule-based constraint functions to remove the weak labels that match to one or more predefined constraint rules to prevent their incorporation into the high-precision training set [a set of weak labels from the document is being generated form the unlabeled data (paragraph 24) and if a label corresponds highly with an annotator (e.g., matches exactly with semantics of a programming language for a function name), the label is selected to train a machine learning mode (paragraph 25 and Fig.2, step 230). Therefore, the unselected generated weak labels are considered being prevented to incorporate into the high-precision training set according to a rule-based constraint functions].
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Weakly labeled data augmentation for social media named entity recognition and Asif’184 according to the teaching of Zhao’183 to include a filter to select the weak labels including the first weak labels and the second weak labels corresponding highly with an annotator because this will allow the first weak labels and the second weak labels to be selected more effectively.
With respect to claim 22, which further limits claim 13, Weakly labeled data augmentation for social media named entity recognition does not teach wherein the instance filter utilizes a neural constraint module to jointly learn and filter negative instances, where the neural constraint module only filters instances that have been added by both of the first and second augmenters.
Zhao’183 teaches wherein the instance filter utilizes a neural constraint module to jointly learn and filter negative instances, where the neural constraint module only filters instances that have been added by both of the first and second augmenters [a set of weak labels (including the first and the second weak labels) from the document is being generated form the unlabeled data (paragraph 24) and if a label corresponds highly with an annotator (e.g., matches exactly with semantics of a programming language for a function name), the label is selected to train a machine learning mode (paragraph 25 and Fig.2, step 230). Therefore, a neural constraint module is considered being disclosed to jointly learn and filter negative instances so that only the labels from the generated weak labels (including the first and the second weak labels) corresponding highly with an annotator (e.g., matches exactly with semantics of a programming language for a function name) are being selected to train the machine learning mode].
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Weakly labeled data augmentation for social media named entity recognition and Asif’184 according to the teaching of Zhao’183 to include a filter to select the weak labels including the first weak labels and the second weak labels corresponding highly with an annotator because this will allow the first weak labels and the second weak labels to be selected more effectively.
With respect to claims 1-3, 9 and 10, they are methods claim that claim how the system of claims 13-15, 21 and 22 to train a machine-learning model. Claims 1-3, 9 and 10 are obvious in view of Weakly labeled data augmentation for social media named entity recognition, Zhao’183 and Kashyapi’673 because the claimed combination operates at the same manner as described in the rejected claims 13-15, 21 and 22. In addition, the reference has disclosed a system to train a machine-learning model is inherent disclosed to be performed by a processor in the system when the system performs the operation to train a machine-learning model.
Claims 4 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Weakly labeled data augmentation for social media named entity recognition, Zhao’183 (US 2021/0319183), Asif’184 (US 2022/0101184) and further in view of Mckay’138 (US 2023/0196138)
With respect to claim 16, which further limits claim 13, Weakly labeled data augmentation for social media named entity recognition teaches wherein the plurality of weak label augmenters includes a label augmenter, and the one or more computing devices are further configured to: train the label augmenter with a robust model labeler, given input seed labels [an alias augmentation and a typo augmentation to generate more weakly labeled data and improve the NER performance in the social media
Domain (page 3). Therefore, an alias augmentation and a typo augmentation is considered being trained with a robust model labeler first before generating the weakly labeled data];
extract the second weak labels from the unlabeled data using the robust model labeler, as trained [an alias augmentation and a typo augmentation to generate more weakly labeled data (including the second weak labels) and improve the NER performance in the social media Domain (page 3)]; and
The combination of Weakly labeled data augmentation for social media named entity recognition, Zhao’183 and Kashyapi’673 does not teach use the high-precision training set, as updated based on the second weak labels filtered by the instance filter, to retrain the robust model labeler of the label augmenter.
Mckay’138 teaches to retrain the labeler according the training data (paragraph 76)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Weakly labeled data augmentation for social media named entity recognition, Zhao’183 and Kashyapi’673 according to the teaching of Mckay’138 to retrain an alias augmentation and a typo augmentation with the training data which having the high-precision training set (use the high-precision training set, as updated based on the second weak labels filtered by the instance filter, to retrain the robust model labeler of the label augmenter) because this will allow the alias augmentation and the typo augmentation to generate the weakly labeled data to be generated more effectively.
With respect to claim 4, it is a method claim that claims how the system of claims 16 to train a machine-learning model. Claims 4 is obvious in view of Weakly labeled data augmentation for social media named entity recognition, Zhao’183, Kashyapi’673 and Mckay’138 because the claimed combination operates at the same manner as described in the rejected claim 16. In addition, the reference has disclosed a system to train a machine-learning model is inherent disclosed to be performed by a processor in the system when the system performs the operation to train a machine-learning model.
Claims 11, 12, 23 and 24 are rejected under 35 U.S.C. 103 as being unpatentable over Weakly labeled data augmentation for social media named entity recognition, Zhao’183 (US 2021/0319183), Asif’184 (US 2022/0101184) and further in view of
Chen’549 (US 2021/0067549).
With respect to claim 23, which further limits claim 13, the combination of Weakly labeled data augmentation for social media named entity recognition, Zhao’183 and Asif’184 does not teach wherein the instance filter utilizes an integrated gradients-based approach to predict whether or not an entity candidate belongs to a target entity class.
Chen’549 teaches wherein the instance filter utilizes an integrated gradients-based approach to predict whether or not an entity candidate belongs to a target entity class [computing integrated gradients of a prediction score for a target class (paragraph 33)].
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Weakly labeled data augmentation for social media named entity recognition, Zhao’183 and Asif’184 according to the teaching of Chen’549 to compute integrated gradients of a prediction score for a target class because this will allow the target class to be identified and labeled more effectively.
With respect to claim 24, which further limits claim 23, the combination of Weakly labeled data augmentation for social media named entity recognition, Zhao’183, Asif’184 and Chen’549 does not teaches wherein the one or more computing devices are further configured to: responsive to a norm of the target entity class being higher including the entity candidate, discard the entity candidate from the high-precision training set; and otherwise, utilize the entity candidate in a next iteration of training for the target entity class.
Since Zhao’183 teaches a set of weak labels from the document is being generated form the unlabeled data (paragraph 24) and if a label corresponds highly with an annotator (e.g., matches exactly with semantics of a programming language for a function name), the label is selected to train a machine learning mode (paragraph 25 and Fig.2, step 230) and Chen’549 has suggested computing integrated gradients of a prediction score for a target class (paragraph 33), therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to discard the entity candidate from the high-precision training set associated with the labels corresponds highly with an annotator to train a machine learning mode and to generate a new set of weak labels from another document to train a machine learning mode when the current set of weak labels generated from a document is not applicable to train a machine learning mode (wherein the one or more computing devices are further configured to: responsive to a norm of the target entity class being higher including the entity candidate, discard the entity candidate from the high-precision training set; and otherwise, utilize the entity candidate in a next iteration of training for the target entity class) because this will allow the machine learning mode to be trained more effectively.
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Weakly labeled data augmentation for social media named entity recognition, Zhao’183, Asif’184 and Chen’549 to discard the entity candidate from the high-precision training set associated with the labels corresponds highly with an annotator to train a machine learning mode and to generate a new set of weak labels from another document to train a machine learning mode when the current set of weak labels generated from a document is not applicable to train a machine learning mode (wherein the one or more computing devices are further configured to: responsive to a norm of the target entity class being higher including the entity candidate, discard the entity candidate from the high-precision training set; and otherwise, utilize the entity candidate in a next iteration of training for the target entity class) because this will allow the machine learning mode to be trained more effectively.
With respect to claims 11 and 12, they are methods claim that claim how the system of claims 23 and 24 to train a machine-learning model. Claims 11 and 12 are obvious in view of Weakly labeled data augmentation for social media named entity recognition, Zhao’183, Asif’184 and Chen’549 because the claimed combination operates at the same manner as described in the rejected claims 23 and 24. In addition, the reference has disclosed a system to train a machine-learning model is inherent disclosed to be performed by a processor in the system when the system performs the operation to train a machine-learning model.
Claim objection
Claim 5 is objected to as being dependent upon a rejected base claim 4 because the prior art of record does not teach “wherein the label augmenter adopts a loss function that includes a weighting of components, the components including one or more of: an unlikelihood objective for class contradiction, for maximizing a probability difference between entities belonging to a correct class in the high-precision training set as compared to entities belonging to another class in the high-precision training set; a minmax entropy optimization approach for prototype re-estimation to minimize entropy given to training data and to maximize entropy on entities that cannot be labeled, to avoid bias against unlabeled entities; and/or an anchor regularizer to limit prototype drift at a current iteration to a maximum distance from prototype embedding at one or more initial iterations.” Claim 5 would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Claims 6-8 are objected to as being dependent upon an objected base claim 5. Claims 6-8 would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Claim 17 is objected to as being dependent upon a rejected base claim 16 because the prior art of record does not teach “wherein the label augmenter adopts a loss function that includes a weighting of components, the components including one or more of: an unlikelihood objective for class contradiction, for maximizing a probability difference between entities belonging to a correct class in the high-precision training set as compared to entities belonging to another class in the high-precision training set; a minmax entropy optimization approach for prototype re-estimation to minimize entropy given to training data and to maximize entropy on entities that cannot be labeled, to avoid bias against unlabeled entities; and/or an anchor regularizer to limit prototype drift at a current iteration to a maximum distance from prototype embedding at one or more initial iterations.” Claim 17 would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Claims 18-20 are objected to as being dependent upon an objected base claim 17. Claims 18-20 would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Contact
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HUO LONG CHEN whose telephone number is (571)270-3759. The examiner can normally be reached on M-F 9am - 5pm. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Tieu, Benny can be reached on (571) 272-7490. The fax phone number for the organization where this application or proceeding is assigned is (571) 273-8300. Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/HUO LONG CHEN/Primary Examiner, Art Unit 2682