# US 12,216,618 B2 · System and a method for the classification of sensitive data elements in a file > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,216,618 B2 - Document ID: US12216618B2 - Status: Granted - Kind: B2 - Application number: 18/156,439 - Filed: 2023-01-19 - Published: 2023-07-27 - Granted: 2025-02-04 - Earliest claimed: 2022-01-21 - Assignees: SECURITI, Inc. - Applicants: SECURITI, Inc. - Inventors: Xiaolin Wang; Michael Rinehart - Web page: https://patentgrep.com/patent/US12216618B2 ## CPC - G06F 16/16: File or folder operations, e.g. details of user interfaces specifically adapted to file systems - G06F 21/6227: {where protection concerns the structure of data, e.g. records, types, queries} - G06F 21/6245: {Protecting personal data, e.g. for financial or medical purposes} - G06N 3/0475: Generative networks - G06N 3/08: Learning methods - G06N 3/045: Combinations of networks - G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound - G06N 5/025: Extracting rules from data - G06N 20/10: using kernel methods, e.g. support vector machines [SVM] - G06N 20/20: Ensemble learning - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A system and a method for classifying sensitive data elements in a file is provided. The method includes receiving and converting, the unstructured data file into a machine-readable format and generating, a plurality of sensitive data features. The plurality of sensitive data features represents single element of the sensitive data. The method includes generating, a plurality of adjacent elements corresponding to the single elements of the sensitive data and generating a plurality of feature categories. The method includes aggregating, the plurality of adjacent node features and the plurality of edge features. The method includes calculating and concatenating the plurality of aggregated adjacent nodes features and the plurality of aggregated edge features. The method includes comparing, the distance of the sensitive data from all of the adjacent sensitive data. The method includes classifying and predicting, the sensitive data to be a true positive or false positive sensitive data by using machine learning. ## Claims ### Claim 1 A system for classifying sensitive data elements in a file using machine learning, wherein the system comprises: one or more hardware processors; a memory coupled to the one or more hardware processors, wherein the memory comprises a plurality of subsystems in form of programmable instructions executable by the one or more hardware processors, and wherein the plurality of subsystems comprises: a processing subsystem hosted on a server, and configured to execute on a network to control bidirectional communications among a plurality of modules comprising: a receiving module configured to receive an unstructured data file wherein the unstructured data file is a data source with the inconsistent structure of data organized in the form of unstructured forms and unstructured natural text; a conversion module operatively coupled to the receiving module wherein the conversion module is configured to convert the unstructured data file into machine-readable format; a machine learning module operatively coupled to the conversion module wherein the machine learning module comprises: a feature generation module operatively coupled with the receiving module and configured to: generate a plurality of sensitive data features, wherein the plurality of sensitive data features represents single elements of the sensitive data; generate a plurality of adjacent elements corresponding to the single elements of the sensitive data, and analyze the relationship between the said single elements of the sensitive data and the plurality of adjacent elements; and generate a plurality of feature categories, wherein the plurality of feature categories comprises a plurality of node features, a plurality of adjacent node features, and a plurality of edge features; a feature calculation module operatively coupled to the feature generation module, wherein the feature calculation module is configured to: aggregate the plurality of adjacent node features and the plurality of edge features by calculating the average of the plurality of adjacent node features and a plurality of adjacent edge features; calculate the plurality of aggregated adjacent nodes features and the plurality of aggregated edge features for one or more adjacent nodes and adjacent node edge; and concatenate the plurality of aggregated adjacent nodes features and the plurality of aggregated edge features with the features of the individual adjacent node and the individual edge; a comparison module operatively connected to the feature calculation module, wherein the comparison module is configured to compare the distance of the sensitive data from all of the adjacent sensitive data to calculate the nearest adjacent sensitive data and select the plurality of adjacent node features; and a classification module operatively coupled with the feature generation module, wherein the classification module is configured to: classify the sensitive data and predicts the sensitive data to be a true positive or false positive sensitive data by using machine learning, in response to receiving the generated sensitive data from the feature generation module, wherein the false positive sensitive data is filtered out and the true positive personal represents accurate sensitive data. ### Claim 2 (depends on 1) The system according to claim 1, comprises a sensitive data feature generation module configured to generate sensitive data features by analysing the distance between the node text and the adjacent node text, wherein the adjacent node text acts as an anchor for the node and the edge features. ### Claim 3 (depends on 2) The system according to claim 2, wherein the sensitive data feature is a true positive sensitive data provided that the sensitive data feature is a security number type and is associated with the anchor. ### Claim 4 (depends on 1) The system according to claim 1, wherein the receiving module is configured to receive at least one of the static sensitive data, the dynamic sensitive data, and the data source information. ### Claim 5 (depends on 1) The system according to claim 1, wherein the feature generation module, feature calculation module, comparison module, and classification module generates sensitive data features by using at least one of a graph neural network and a machine learning model. ### Claim 6 (depends on 1) The system according to claim 1, wherein the feature generation module generates a plurality of node features by analysing the uniqueness and an identifiability score for each of the nodes based on the node type, wherein the identifiability score represents correctness of the estimation of the node features. ### Claim 7 (depends on 1) The system according to claim 1, wherein the plurality of node features, the plurality of adjacent node features, and the plurality of edge features are concatenated into a one-dimensional feature vector, and wherein the feature vector is a vector containing multiple elements corresponding to the sensitive data feature. ### Claim 8 (depends on 1) The system according to claim 1, classifies the sensitive data from the entire document, page, chunks, and features. ### Claim 9 (depends on 1) The system according to claim 1, wherein the plurality of edge features represents a plurality of attributes regarding the sensitive data node and related adjacent sensitive data nodes. ### Claim 10 (depends on 1) The system according to claim 1, wherein the adjacent sensitive data targets data detection such as name detection and anchor detection such as a name anchor detection. ### Claim 11 (depends on 1) The system according to claim 1, wherein the machine learning module comprises a plurality of machine learning rules, wherein the machine learning rules are configured to: determine the probability associated with the prediction of a sensitive data relating to a file; and classify the file as containing sensitive data, based on the determined probability. ### Claim 12 A method for classifying sensitive data elements in a file, the method comprises: receiving, by a receiving module of a processing subsystem comprising: one or more hardware processors; and a memory coupled to the one or more hardware processors, wherein the memory comprises a plurality of subsystems in form of programmable instructions executable by the one or more hardware processors, and wherein the plurality of subsystems comprises: an unstructured data file wherein the unstructured data file is a data source with the inconsistent structure of data organized in the form of unstructured forms and unstructured natural text; converting, the unstructured data file into machine-readable format; generating, by a feature generation module of the processing subsystem, a plurality of sensitive data features, wherein the plurality of sensitive data features represents single elements of the sensitive data; generating, by a feature generation module of the processing subsystem, a plurality of adjacent elements corresponding to the single elements of the sensitive data, and analyzing the relationship between the said single elements of the sensitive data and the plurality of adjacent elements; generating, by a feature generation module of the processing subsystem, generate a plurality of feature categories, wherein the plurality of feature categories comprises a plurality of node features, a plurality of adjacent node features, and a plurality of edge features; aggregating, by a feature calculation module of the processing subsystem, the plurality of adjacent node features and the plurality of edge features by calculating the average of the plurality of adjacent node features and a plurality of adjacent edge features; calculating, by a feature calculation module of the processing subsystem, the plurality of aggregated adjacent nodes features and the plurality of aggregated edge features for one or more adjacent nodes and adjacent node edge; concatenating, by a feature calculation module of the processing subsystem, the plurality of aggregated adjacent nodes features and the plurality of aggregated edge features with the features of the individual adjacent node and the individual edge; comparing, by a comparison, module of the processing subsystem, the distance of the sensitive data from all of the adjacent sensitive data to calculate the nearest adjacent sensitive data and select the plurality of adjacent node features; and classifying and predicting, by a classification module of the processing subsystem, the sensitive data and predicts the sensitive data to be a true positive or false positive sensitive data by using machine learning, n response to receiving the generated sensitive data from the feature generation module, wherein the false positive sensitive data is filtered out and the true positive personal represents accurate sensitive data. ### Claim 13 (depends on 12) The method according to claim 12, wherein determining the adjacent node features comprises: providing k-adjacent node individually; calculating features for each node and calculating an average of the adjacent features to generate the aggregated adjacent nodes features; calculating the aggregated adjacent nodes features for a plurality of k and concatenate; and concatenating, the calculated aggregated adjacent node features with the features of the individual adjacent nodes. ### Claim 14 (depends on 12) The method according to claim 12, comprises extracting the sensitive data from the data file and converting the extracted data into machine-learnable representation. ### Claim 15 (depends on 12) The method according to claim 12, comprises predicting the converted sensitive data for correctly predicted node features. ### Claim 16 A non-transitory computer-readable medium storing a computer program that, when executed by a processor, causes the processor to perform a method for the classification of sensitive data elements in a file, wherein the method comprises: receiving, by a receiving module of a processing subsystem, an unstructured data file wherein the unstructured data file is a data source with the inconsistent structure of data organized in the form of unstructured forms and unstructured natural text; converting, the unstructured data file into machine-readable format; generating, by a feature generation module of the processing subsystem, a plurality of sensitive data features, wherein the plurality of sensitive data features represents single elements of the sensitive data; generating, by a feature generation module of the processing subsystem, a plurality of adjacent elements corresponding to the single elements of the sensitive data, and analyzing the relationship between the said single elements of the sensitive data and the plurality of adjacent elements; generating, by a feature generation module of the processing subsystem, generate a plurality of feature categories, wherein the plurality of feature categories comprises a plurality of node features, a plurality of adjacent node features, and a plurality of edge features; aggregating, by a feature calculation module of the processing subsystem, the plurality of adjacent node features and the plurality of edge features by calculating the average of the plurality of adjacent node features and a plurality of adjacent edge features; calculating, by a feature calculation module of the processing subsystem, the plurality of aggregated adjacent nodes features and the plurality of aggregated edge features for one or more adjacent nodes and adjacent node edge; concatenating, by a feature calculation module of the processing subsystem, the plurality of aggregated adjacent nodes features and the plurality of aggregated edge features with the features of the individual adjacent node and the individual edge; comparing, by a comparison, module of the processing subsystem, the distance of the sensitive data from all of the adjacent sensitive data to calculate the nearest adjacent sensitive data and select the plurality of adjacent node features; and classifying and predicting, by a classification module of the processing subsystem, the sensitive data and predicting the sensitive data to be a true positive or false positive sensitive data by using machine learning, in response to receiving the generated sensitive data from the feature generation module, wherein the false positive sensitive data is filtered out and the true positive personal represents accurate sensitive data.