DETAILED ACTION
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.
Claims 1-7 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matters:
Regarding claim 1, a system with multiple components/units, i.e., an input unit, a storage unit, a processing unit, and an output unit is being recited in claim. However, the claimed components/units can be interpreted by a person of ordinary skills in the art as software modules that carry out the claimed functions. As such, the claim is not limited to statutory subject matter and is therefore non-statutory.
Claims 2-7 fail to resolve the deficiencies of claim 1 since they only further limit the functions, or the scope of claim1. Hence, claims 2-7 are also rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matters.
.
Claims 1-4, and 8 are subject to the claim rejection under 35 U.S.C. 101 because the each claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more:
According to the 2024 Guidance Update on Patent Subject Matter Eligibility, Step 1 analysis: claim 8 is identified as identified as directed to a process;
In Step 2A, Prong One Analysis evaluates whether the claim recites a
judicial exception:
independent claim 1 recites functions of “receive document data and reference document data”, “store a classification model …”; “create first classification data, second classification data ….”, “create document comparison data….”, “determining a category of the reference document data ….” and “output the category”; which are as abstract idea of Mental processes, that can be mentally carried out solely by human, and without any particular algorithms or improvements, a person would be able to obtain document data, create or write down classification data, which can be just collection of words contained , or not contained in the document data, and a person would be able to determining a category of the reference document data ….” and “output the category”; so that these functions, under broadest reasonable interpretation (BRI), covers performance of the limitations in the mind with paper and pens, but for the recitation of generic computer components or software. Nothing in the claim precludes the limitation from being performed in the human mind and/or with the aid of pen/paper.
Step (2) Analysis, each of independent claims 1, 2, and 8 respectively recites abstract idea of Mental processes, those steps are human perception and mental processes. There is no limitation, or combination of another element, and there is no additional element. The claim does not limit any other particular device, or algorithms, or any modification or improvement of the device.
In Step 2B, Prong Two Analysis to evaluate whether the claimed additional elements amount to significantly more than the recited judicial exception itself:
There are no any additional elements recited in claim beyond the Mental processes. As no particular computer function, algorithm or configuration is practically improved, and no any particular application, or improvement in any meaningful practice is defined. Therefore, independent claims 1, 2, and 8 respectively do not have additional elements amount to significantly more than the recited judicial exception itself.
Overall, Claim 1 does not limit what’s specialty, algorithms and characteristics of a method, or a device with any specific algorithms or improvements. Therefore, Claim 1 is considered as human mental processing, and a conventional data and image mental processes, and would be conventional practice of human observation and mental processing. Therefore above mentioned steps are considered as an abstract idea of Mental processes.
Dependent claims 3-4 merely limit create vector data, and average vector data without any additional or specific algorithms or improvements. Therefore, claims 1-4, and 8 are ineligible.
On the other hands, claims 5-7, respectively recites that the classification model comprising a neural network, which is trained with first document data, second document data, and a category as teacher data. Thus, claims 5-7 are eligible in Patent Subject Matter Eligibility analysis.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103(a) which forms the basis for all obviousness rejections set forth in this Office action:
(a) A patent may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
Claims 1-8 are rejected under 35 U.S.C. 103(a) as being unpatentable over Zhang et al. (US 20220237409 A1, Date Filed: 2021-03-31), and in view of TRIPATHI (US 20210004670 A1).
Re Claim 1, Zhang et al. discloses a document classification system comprising an input unit, a storage unit, a processing unit, and an output unit (see Zhang: e.g., --A method, an electronic device, and a computer program product for processing data is disclosed. The method includes training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The method further includes determining from the individual words identification information that can identify the objects based on contributions of individual words in the reference documents to the association. Identification information that can identify objects in documents describing the objects may be determined, so that an identification information data set is automatically generated for training a machine learning model that is used to determine the identification information.--, in abstract, and, --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]);
wherein the input unit is configured to receive document data and reference document data (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]);
wherein the storage unit is configured to store a classification model (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]),
wherein the processing unit is configured to create first classification data, second classification data, and third classification data from the document data and the reference document data (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]);
wherein a word contained in the document data and not contained in the reference document data belongs to the first classification data (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]);
wherein a word contained in the document data and contained in the reference document data belongs to the second classification data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]);
wherein a word not contained in the document data and contained in the reference document data belongs to the third classification data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]);
wherein the processing unit is configured to create document comparison data from the first classification data, the second classification data, and the third classification data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]);
although Zhang discloses wherein the processing unit is configured to determine a class of the reference document data from the document comparison data using the classification model (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033] {herein associating reference document with the described object and the identification information using the classification model});
Zhang however does not explicitly disclose determining a category of the reference document data from the document comparison data using the classification model;
TRIPATHI discloses determining a category of the reference document data from the document comparison data using the classification model (see TRIPATHI: e.g., Fig. 3A, --conventional digital content management systems can capture or generate digital content and then distribute the digital content to individual client devices. To illustrate, conventional digital content management systems can classify electronic documents to determine document categories and then curate and distribute digital content to targeted client devices.
[0002] In addition, many cloud-based systems can store and manage digital content for client devices. For example, cloud-based systems can store thousands or millions of digital content items that users can access via various client devices. Some conventional digital content management systems classify electronic documents and determine document categories to assist in managing these large digital content repositories.--, in [0002]-[0003]; and, --the disclosed systems selectively pad word/phrase embeddings having a strong affinity towards a particular class/category or combination of classes. In particular, the disclosed systems can determine digital content source densities based on the ratio of distinctive words used in documents from different digital content sources relative to content neutral words from a neutral word corpus. The disclosed systems can then apply padding to word embeddings based on the density of distinctive words for each digital content source. The disclosed systems thus reduce the impact of embedding of common words (i.e., classification neutral words) when deriving embeddings of documents from the words in the documents while emphasizing word embeddings from digital content sources that contain distinctive words or phrases.
[0007] Furthermore, in one or more embodiments, the disclosed systems utilize batchwise weighted loss functions in training the machine-learning model. Specifically, the disclosed systems can set different weights for different classes for loss determination in a batch of training samples based on the proportion of samples of each class. More specifically, in one or more embodiments the disclosed systems calculate a weight for a given class based on the number of positive samples relative to the batch size for the given class. The disclosed systems can thus use dynamic assignment of weights in determining the losses to prevent numerical instability while fairly and accurately representing positive and negative samples. Using this approach, the disclosed systems can further improve accuracy and efficiency in training digital content classification models across various classes.--, in [0006]-[0007]);
Zhang and TRIPATHI are combinable as they are in the same field of endeavor: documents classification and content analysis, Therefore it would have been obvious to one of ordinary skill in the art at the time the invention was made to further modify Zhang’s system using TRIPATHI’s teachings by including determining a category of the reference document data from the document comparison data using the classification model to Zhang’s document classification in order to improve accuracy and efficiency in training digital content classification models across various classes (see TRIPATHI: e.g., in [0002]-[0003], and [0006]-[0007]);
Zhang as modified by TRIPATHI further disclose wherein the output unit is configured to output the category (see TRIPATHI: e.g., Fig. 3A, --conventional digital content management systems can capture or generate digital content and then distribute the digital content to individual client devices. To illustrate, conventional digital content management systems can classify electronic documents to determine document categories and then curate and distribute digital content to targeted client devices.
[0002] In addition, many cloud-based systems can store and manage digital content for client devices. For example, cloud-based systems can store thousands or millions of digital content items that users can access via various client devices. Some conventional digital content management systems classify electronic documents and determine document categories to assist in managing these large digital content repositories.--, in [0002]-[0003]; and, --the disclosed systems selectively pad word/phrase embeddings having a strong affinity towards a particular class/category or combination of classes. In particular, the disclosed systems can determine digital content source densities based on the ratio of distinctive words used in documents from different digital content sources relative to content neutral words from a neutral word corpus. The disclosed systems can then apply padding to word embeddings based on the density of distinctive words for each digital content source. The disclosed systems thus reduce the impact of embedding of common words (i.e., classification neutral words) when deriving embeddings of documents from the words in the documents while emphasizing word embeddings from digital content sources that contain distinctive words or phrases.
[0007] Furthermore, in one or more embodiments, the disclosed systems utilize batchwise weighted loss functions in training the machine-learning model. Specifically, the disclosed systems can set different weights for different classes for loss determination in a batch of training samples based on the proportion of samples of each class. More specifically, in one or more embodiments the disclosed systems calculate a weight for a given class based on the number of positive samples relative to the batch size for the given class. The disclosed systems can thus use dynamic assignment of weights in determining the losses to prevent numerical instability while fairly and accurately representing positive and negative samples. Using this approach, the disclosed systems can further improve accuracy and efficiency in training digital content classification models across various classes.--, in [0006]-[0007], and, --[0021] Additionally, in one or more embodiments, the classification system utilizes a batchwise weighted loss function to train the machine-learning model. Specifically, the classification system can determine weights to apply to individual classification categories when determining loss from an output of the machine-learning model for each batch of training samples. For example, the classification system can determine weights for different categories in a batch based on the proportion of positive samples in each category. The classification system can then use a batchwise weighted cross-entropy loss to modify parameters of the machine-learning model using customized weights for output categories within each separate batch of training samples. By utilizing batchwise weighted loss functions, the disclosed systems can further improve accuracy and efficiency of training digital content classification models.--, in [0021]).
Re Claim 2, Zhang discloses a document classification system comprising an input unit, a storage unit, a processing unit, and an output unit (see Zhang: e.g., --A method, an electronic device, and a computer program product for processing data is disclosed. The method includes training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The method further includes determining from the individual words identification information that can identify the objects based on contributions of individual words in the reference documents to the association. Identification information that can identify objects in documents describing the objects may be determined, so that an identification information data set is automatically generated for training a machine learning model that is used to determine the identification information.--, in abstract, and, --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]),
wherein the input unit is configured to receive document data (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]),
wherein the storage unit is configured to store reference document data and a classification model (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]),
the processing unit is configured to create first classification data, second classification data, and third classification data from the document data and the reference document data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]; and, --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]),
wherein a word contained in the document data and not contained in the reference document data belongs to the first classification data (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]),
wherein a word contained in the document data and contained in the reference document data belongs to the second classification data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]),
wherein a word not contained in the document data and contained in the reference document data belongs to the third classification data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]),
wherein the processing unit is configured to create document comparison data from the first classification data, the second classification data, and the third classification data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]),
although Zhang discloses wherein the processing unit is configured to determine a class of the reference document data from the document comparison data using the classification model (see Zhang: e.g., --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033] {herein associating reference document with the described object and the identification information using the classification model});
Zhang however does not explicitly disclose determining a category of the reference document data from the document comparison data using the classification model;
TRIPATHI discloses determining a category of the reference document data from the document comparison data using the classification model (see TRIPATHI: e.g., Fig. 3A, --conventional digital content management systems can capture or generate digital content and then distribute the digital content to individual client devices. To illustrate, conventional digital content management systems can classify electronic documents to determine document categories and then curate and distribute digital content to targeted client devices.
[0002] In addition, many cloud-based systems can store and manage digital content for client devices. For example, cloud-based systems can store thousands or millions of digital content items that users can access via various client devices. Some conventional digital content management systems classify electronic documents and determine document categories to assist in managing these large digital content repositories.--, in [0002]-[0003]; and, --the disclosed systems selectively pad word/phrase embeddings having a strong affinity towards a particular class/category or combination of classes. In particular, the disclosed systems can determine digital content source densities based on the ratio of distinctive words used in documents from different digital content sources relative to content neutral words from a neutral word corpus. The disclosed systems can then apply padding to word embeddings based on the density of distinctive words for each digital content source. The disclosed systems thus reduce the impact of embedding of common words (i.e., classification neutral words) when deriving embeddings of documents from the words in the documents while emphasizing word embeddings from digital content sources that contain distinctive words or phrases.
[0007] Furthermore, in one or more embodiments, the disclosed systems utilize batchwise weighted loss functions in training the machine-learning model. Specifically, the disclosed systems can set different weights for different classes for loss determination in a batch of training samples based on the proportion of samples of each class. More specifically, in one or more embodiments the disclosed systems calculate a weight for a given class based on the number of positive samples relative to the batch size for the given class. The disclosed systems can thus use dynamic assignment of weights in determining the losses to prevent numerical instability while fairly and accurately representing positive and negative samples. Using this approach, the disclosed systems can further improve accuracy and efficiency in training digital content classification models across various classes.--, in [0006]-[0007]);
Zhang and TRIPATHI are combinable as they are in the same field of endeavor: documents classification and content analysis, Therefore it would have been obvious to one of ordinary skill in the art at the time the invention was made to further modify Zhang’s system using TRIPATHI’s teachings by including determining a category of the reference document data from the document comparison data using the classification model to Zhang’s document classification in order to improve accuracy and efficiency in training digital content classification models across various classes (see TRIPATHI: e.g., in [0002]-[0003], and [0006]-[0007]);
Zhang as modified by TRIPATHI further disclose wherein the output unit is configured to output the category (see TRIPATHI: e.g., Fig. 3A, --conventional digital content management systems can capture or generate digital content and then distribute the digital content to individual client devices. To illustrate, conventional digital content management systems can classify electronic documents to determine document categories and then curate and distribute digital content to targeted client devices.
[0002] In addition, many cloud-based systems can store and manage digital content for client devices. For example, cloud-based systems can store thousands or millions of digital content items that users can access via various client devices. Some conventional digital content management systems classify electronic documents and determine document categories to assist in managing these large digital content repositories.--, in [0002]-[0003]; and, --the disclosed systems selectively pad word/phrase embeddings having a strong affinity towards a particular class/category or combination of classes. In particular, the disclosed systems can determine digital content source densities based on the ratio of distinctive words used in documents from different digital content sources relative to content neutral words from a neutral word corpus. The disclosed systems can then apply padding to word embeddings based on the density of distinctive words for each digital content source. The disclosed systems thus reduce the impact of embedding of common words (i.e., classification neutral words) when deriving embeddings of documents from the words in the documents while emphasizing word embeddings from digital content sources that contain distinctive words or phrases.
[0007] Furthermore, in one or more embodiments, the disclosed systems utilize batchwise weighted loss functions in training the machine-learning model. Specifically, the disclosed systems can set different weights for different classes for loss determination in a batch of training samples based on the proportion of samples of each class. More specifically, in one or more embodiments the disclosed systems calculate a weight for a given class based on the number of positive samples relative to the batch size for the given class. The disclosed systems can thus use dynamic assignment of weights in determining the losses to prevent numerical instability while fairly and accurately representing positive and negative samples. Using this approach, the disclosed systems can further improve accuracy and efficiency in training digital content classification models across various classes.--, in [0006]-[0007], and, --[0021] Additionally, in one or more embodiments, the classification system utilizes a batchwise weighted loss function to train the machine-learning model. Specifically, the classification system can determine weights to apply to individual classification categories when determining loss from an output of the machine-learning model for each batch of training samples. For example, the classification system can determine weights for different categories in a batch based on the proportion of positive samples in each category. The classification system can then use a batchwise weighted cross-entropy loss to modify parameters of the machine-learning model using customized weights for output categories within each separate batch of training samples. By utilizing batchwise weighted loss functions, the disclosed systems can further improve accuracy and efficiency of training digital content classification models.--, in [0021]).
Re Claim 3, Zhang as modified by TRIPATHI further disclose wherein the processing unit is configured to create first vector data from the word belonging to the first classification data (see TRIPATHI: e.g., -- the classification system creates document embeddings representing text documents from digital content sources. Specifically, the classification system can generate word embeddings representing padded segments corresponding to a text document from a digital content source. For instance, the classification system can utilize an encoder model to generate word embeddings that are vector representation of each padded segment in the text document. In one or more embodiments, the classification system then determines document embedding by combining (e.g., averaging) the word embeddings corresponding to the text document. Accordingly, in one or more embodiments the classification system generates a document embedding that is a vector representation of a plurality of words in the text document and that reflects the source density of the digital content source.--, in [0024]),
wherein the processing unit is configured to create second vector data from the word belonging to the second classification data (see TRIPATHI: e.g., -- the classification system creates document embeddings representing text documents from digital content sources. Specifically, the classification system can generate word embeddings representing padded segments corresponding to a text document from a digital content source. For instance, the classification system can utilize an encoder model to generate word embeddings that are vector representation of each padded segment in the text document. In one or more embodiments, the classification system then determines document embedding by combining (e.g., averaging) the word embeddings corresponding to the text document. Accordingly, in one or more embodiments the classification system generates a document embedding that is a vector representation of a plurality of words in the text document and that reflects the source density of the digital content source.--, in [0024]),
wherein the processing unit is configured to create third vector data from the word belonging to the third classification data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]; and see TRIPATHI: e.g., -- the classification system creates document embeddings representing text documents from digital content sources. Specifically, the classification system can generate word embeddings representing padded segments corresponding to a text document from a digital content source. For instance, the classification system can utilize an encoder model to generate word embeddings that are vector representation of each padded segment in the text document. In one or more embodiments, the classification system then determines document embedding by combining (e.g., averaging) the word embeddings corresponding to the text document. Accordingly, in one or more embodiments the classification system generates a document embedding that is a vector representation of a plurality of words in the text document and that reflects the source density of the digital content source.--, in [0024]), and
wherein the processing unit is configured to create the document comparison data from the first vector data, the second vector data, and the third vector data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]; and see TRIPATHI: e.g., -- the classification system creates document embeddings representing text documents from digital content sources. Specifically, the classification system can generate word embeddings representing padded segments corresponding to a text document from a digital content source. For instance, the classification system can utilize an encoder model to generate word embeddings that are vector representation of each padded segment in the text document. In one or more embodiments, the classification system then determines document embedding by combining (e.g., averaging) the word embeddings corresponding to the text document. Accordingly, in one or more embodiments the classification system generates a document embedding that is a vector representation of a plurality of words in the text document and that reflects the source density of the digital content source.--, in [0024]).
Re Claim 4, Zhang as modified by TRIPATHI further disclose wherein the processing unit is configured to create first vector data from the word belonging to the first classification data and averaging elements of the first vector data to create first average vector data (see TRIPATHI: e.g., -- [0059] The classification system 102 then creates document embeddings 210a-210c representing documents from the sources 202a-202c based on the corresponding padded segments 208a-208c. In one or more embodiments, the classification system 102 utilizes word embeddings (vector representations of n-grams from the vocabularies of the sources 202a-202c) corresponding to the padded segments 208a-208c to create the document embeddings 210a-210c. In particular, the classification system 102 can utilize an encoder (e.g., one-hot-encoding or semantic encoding such as a Word2Vec model) to generate word embeddings and then combine the word embeddings associated with a given document to generate the document embeddings 210a-210c. To illustrate, the classification system 102 creates a document embedding for a text document by combining the word embeddings from segments corresponding to the text document. In an example, the classification system 102 combines the word embeddings by averaging the word embeddings, resulting in an averaged vector representation of the n-grams in the text document. FIGS. 4A-4B and the corresponding description provide additional detail in connection with generating word embeddings and document embeddings..--, in [0059]),
wherein the processing unit is configured to create second vector data from the word belonging to the second classification data and averaging elements of the second vector data to create second average vector data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]; and see TRIPATHI: e.g., -- [0059] The classification system 102 then creates document embeddings 210a-210c representing documents from the sources 202a-202c based on the corresponding padded segments 208a-208c. In one or more embodiments, the classification system 102 utilizes word embeddings (vector representations of n-grams from the vocabularies of the sources 202a-202c) corresponding to the padded segments 208a-208c to create the document embeddings 210a-210c. In particular, the classification system 102 can utilize an encoder (e.g., one-hot-encoding or semantic encoding such as a Word2Vec model) to generate word embeddings and then combine the word embeddings associated with a given document to generate the document embeddings 210a-210c. To illustrate, the classification system 102 creates a document embedding for a text document by combining the word embeddings from segments corresponding to the text document. In an example, the classification system 102 combines the word embeddings by averaging the word embeddings, resulting in an averaged vector representation of the n-grams in the text document. FIGS. 4A-4B and the corresponding description provide additional detail in connection with generating word embeddings and document embeddings..--, in [0059]),
wherein the processing unit is configured to create third vector data from the word belonging to the third classification data and averaging elements of the third vector data to create third average vector data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]; and see TRIPATHI: e.g., -- [0059] The classification system 102 then creates document embeddings 210a-210c representing documents from the sources 202a-202c based on the corresponding padded segments 208a-208c. In one or more embodiments, the classification system 102 utilizes word embeddings (vector representations of n-grams from the vocabularies of the sources 202a-202c) corresponding to the padded segments 208a-208c to create the document embeddings 210a-210c. In particular, the classification system 102 can utilize an encoder (e.g., one-hot-encoding or semantic encoding such as a Word2Vec model) to generate word embeddings and then combine the word embeddings associated with a given document to generate the document embeddings 210a-210c. To illustrate, the classification system 102 creates a document embedding for a text document by combining the word embeddings from segments corresponding to the text document. In an example, the classification system 102 combines the word embeddings by averaging the word embeddings, resulting in an averaged vector representation of the n-grams in the text document. FIGS. 4A-4B and the corresponding description provide additional detail in connection with generating word embeddings and document embeddings..--, in [0059]), and
wherein the processing unit is configured to create the document comparison data from the first average vector data, the second average vector data, and the third average vector data (see Zhang: e.g., --training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The actions further include: based on contributions of individual words in the reference documents to the association, determining from the individual words identification information that can identify the objects.--, in [0006], and, --the solution can determine the identification information in the reference documents, thereby automatically generating an identification information data set without manually labeling the identification information. Therefore, a lot of time and resources may be saved. In addition, in this solution, by determining a specific word as the identification information based on the contributions of the words in the documents to the association of the documents with the corresponding objects, potential identification information may be determined. For example, in the case where the objects are individuals, this solution may determine potential indirect PII based on the context. This is because when manually labeling PII, some words or sentences that can indirectly identify an individual may be omitted. Therefore, the deep learning NER trained based on the general data set of manually labeled PII may not be able to recognize these types of indirect PII.
[0028] In contrast, in this solution, by training the classification model that associates the reference documents with the corresponding individuals, a contribution of each word in the documents to the identified individual may be determined according to the trained classification model. By determining the words that have great contributions to the previous association between the documents and the individual as PII, the potential indirect PII may be determined as much as possible. In addition, by adjusting a contribution threshold, PII data sets of different quality may be generated. For example, if the contribution threshold is large, only words that make a great contribution to the association can be determined as PII, so the generated PII data set has a higher quality.--, in [0027]-[0028]; and see TRIPATHI: e.g., -- [0059] The classification system 102 then creates document embeddings 210a-210c representing documents from the sources 202a-202c based on the corresponding padded segments 208a-208c. In one or more embodiments, the classification system 102 utilizes word embeddings (vector representations of n-grams from the vocabularies of the sources 202a-202c) corresponding to the padded segments 208a-208c to create the document embeddings 210a-210c. In particular, the classification system 102 can utilize an encoder (e.g., one-hot-encoding or semantic encoding such as a Word2Vec model) to generate word embeddings and then combine the word embeddings associated with a given document to generate the document embeddings 210a-210c. To illustrate, the classification system 102 creates a document embedding for a text document by combining the word embeddings from segments corresponding to the text document. In an example, the classification system 102 combines the word embeddings by averaging the word embeddings, resulting in an averaged vector representation of the n-grams in the text document. FIGS. 4A-4B and the corresponding description provide additional detail in connection with generating word embeddings and document embeddings..--, in [0059]).
Re Claims 5-7, Zhang as modified by TRIPATHI further disclose wherein the classification model comprises a neural network (see Zhang: e.g., --A method, an electronic device, and a computer program product for processing data is disclosed. The method includes training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The method further includes determining from the individual words identification information that can identify the objects based on contributions of individual words in the reference documents to the association. Identification information that can identify objects in documents describing the objects may be determined, so that an identification information data set is automatically generated for training a machine learning model that is used to determine the identification information.--, in abstract, and, --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]), and
wherein the processing unit is configured to train the classification model with first document data, second document data, and a category as teacher data (see Zhang: e.g., --A method, an electronic device, and a computer program product for processing data is disclosed. The method includes training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The method further includes determining from the individual words identification information that can identify the objects based on contributions of individual words in the reference documents to the association. Identification information that can identify objects in documents describing the objects may be determined, so that an identification information data set is automatically generated for training a machine learning model that is used to determine the identification information.--, in abstract, and, --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]).
Re Claim 8, claim 8 is the corresponding method claim to claim 1, respectively. Claim 8 thus is rejected for the similar reasons for claim 1. See above discussions with regard to claim 1 respectively. Zhang as modified by TRIPATHI further disclose a document classification method (see Zhang: e.g., --A method, an electronic device, and a computer program product for processing data is disclosed. The method includes training a classification model based on a plurality of reference documents describing different objects, the trained classification model respectively associating the plurality of reference documents with the described objects. The method further includes determining from the individual words identification information that can identify the objects based on contributions of individual words in the reference documents to the association. Identification information that can identify objects in documents describing the objects may be determined, so that an identification information data set is automatically generated for training a machine learning model that is used to determine the identification information.--, in abstract, and, --[0033] Computing device 120 may use reference data set 110 to train classification model 135 deployed in computing device 120. Classification model 135 may be a machine learning model configured to associate a document describing an object with the described object. For example, classification model 135 may be a Hierarchical Attention Network (HAN). The HAN is a commonly used machine learning model using an attention mechanism for text classification. After being trained, classification model 135 may be used to generate identification information data set 150 based on reference documents 112 in reference data set 110. The content of this aspect will be described in detail below with reference to FIG. 2 to FIG. 4.--, in [0033]).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to WEIWEN YANG whose telephone number is (571)270-5670. The examiner can normally be reached on Monday-Friday 8:30am-4:30pm east.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Amandeep Saini can be reached on 571-272-3382. 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.
/WEI WEN YANG/ Primary Examiner, Art Unit 2662