System and/or method for generating clean records from imperfect data using model stack(s) including classification model(s) and confidence model(s)

A data classification system, comprising:

• an input interface configured to receive documents comprising data entries, at least some of the data entries having associated features represented directly in the documents;
• a non-transitory computer readable storage medium comprising a data warehouse configured to store curated and classified data elements;
• a model registry storing at least one model stack, each model stack including at least one classification model and at least one confidence model that is separate from the at least one classification model in the respective model stack, each classification model being configured to generate proposed classifications in accordance with a multi-element taxonomy, with each element in the multi-element taxonomy denoting a respective category, each confidence model being configured to make a trust or do not trust decision relative to a proposed classification generated by the at least one classification model in the associated model stack;
• and
• processing resources including at least one processor and a memory, the memory comprising instructions executable to at least:
• inspect documents received via the input interface to identify data entries and associated features located in the inspected documents;
• identify one or more model stacks from the model registry for execution on the identified data entries;
• execute the at least one classification model in the identified one or more model stacks to generate proposed classifications for the identified data entries;
• execute the at least one confidence model in the identified one or more model stacks to generate trust or do not trust decisions for each of the generated proposed classifications based on a likelihood of category collision occurring through execution of the at least one classification model, wherein category collision represents classification uncertainty as between two or more elements in the taxonomy for a given data entry, and wherein trust decisions indicate absences of category collisions;
• store, to the data warehouse, data entries and their associated proposed classifications, for those proposed classifications where trust decisions have been reached via execution of the at least one confidence model;
• and
• reference records in the data warehouse in response to queries from a computer terminal.