System and method for extracting structured data

A system comprising:

• a server system including a processor and a memory, the memory storing instructions which, when executed, cause the server system to perform:
• clustering, by a clustering module, stored structured data files based on visual similarity of visual representations of the stored structured data files, wherein the visual similarity includes a location of data in the visual representation associated with each file in the stored structed data files;
• after clustering the stored structured data files, associating templates with one or more corresponding clusters of structured data files;
• receiving, at an organization, a structured data file in a first data file structure, the structured data file including one or more tables;
• generating a visual representation of the structured data file, wherein the visual representation is an image file representative of an appearance of a layout and arrangement of the one or more tables within the structured data file;
• extracting, by a featureset extractor, a featureset including a plurality of features from the visual representation via an image classification model in order to classify the visual representation as belonging to a particular class;
• comparing, by a visual representation comparison module, the featureset of the visual representation to one or more featuresets extracted from one or more other visual representations of structured data files based, at least in part, on a cosine distance between the featureset and each of the one or more featuresets, wherein the cosine distance measures a visual similarity between two visual representations in order to identify a group of one or more structured data files having an appearance similar to the layout and arrangement of the one or more tables within the structured data file;
• based on a determination that the featureset is within a predetermined threshold distance from an identified featureset from among the one or more featuresets, selecting a template associated with the identified featureset, wherein the template associated with the identified featureset is identified, at least in part, by being associated with a cluster of structured data files, the identified featureset being associated with one of the structured data files included in the cluster and the template including a definition of at least one key and at least one position of data within the one or more tables, wherein the at least one key identifies a target data element for a particular data segment;
• using the selected template to identify key-value pairs within the structured data file;
• and
• automatically generating an output file in a common data file structure from information in the structured data file based, at least in part, on positions of the key-value pairs within the structured data file, the common data file structure being different from the first data file structure.

A non-transitory computer-readable storage medium comprising computer-executable instructions which, when executed, cause a computing system to perform:

• clustering, by a clustering module, stored structured data files based on visual similarity of visual representations of the stored structured data files, wherein the visual similarity includes a location of data in the visual representation associated with each file in the stored structed data files;
• after clustering the stored structured data files, associating templates with one or more corresponding clusters of structured data files;
• receiving, at an organization, a structured data file in a first data file structure, the structured data file including one or more tables;
• generating a visual representation of the structured data file, wherein the visual representation is an image file representative of an appearance of a layout and arrangement of the one or more tables within the structured data file;
• extracting, by a featureset extractor, a featureset including a plurality of features from the visual representation via an image classification model in order to classify the visual representation, as belonging to a particular class;
• comparing, by a visual representation comparison module, the featureset of the visual representation to one or more featuresets associated with one or more other visual representations of structured data files based, at least in part, on a cosine distance between the featureset and each of the one or more featuresets, wherein the cosine distance measures a visual similarity between two visual representations in order to identify a group of one or more structured data files having an appearance similar to the layout and arrangement of the one or more tables within the structured data file;
• based on a determination that the featureset is outside of a predetermined threshold distance from any of the one or more featuresets:
• prompting a user, via an application plugin, to define one or more field labels and values associated with the structured data file;
• generating, from the one or more field labels and values associated with the structured data file, one or more key-value pairs associated with positions within the structured data file;
• creating a template based, at least in part, on the one or more key-value pairs associated with positions within the structured data file;
• based on a determination that the featureset is within a predetermined threshold distance from an identified featureset from among the one or more featuresets:
• selecting a template associated with the identified featureset, wherein the template associated with the identified featureset is identified, at least in part, by being associated with a cluster of structured data files, the identified featureset being associated with one of the structured data files included in the cluster and the template including a definition of at least one key and at least one position of data within the one or more tables, wherein the at least one key identifies a target data element for a particular data segment;
• using the selected template to identify key-value pairs within the structured data file;
• and
• automatically generating an output file in a common data file structure from information in the structured data file based, at least in part, on positions of the key-value pairs within the structured data file, the common data file structure being different from the first data file structure.