Rectifying missing or incorrect labels in unstructured data

A method comprising:

• receiving a first representation of an unstructured data entity, wherein the first representation comprises first indications or first absences of a first detection type and a second detection type, the first detection type being different from the second detection type;
• assigning the first detection type to a first group of machine learning components based on performance levels of the first group corresponding to detection of the first detection type;
• assigning the second detection type to a second group of machine learning components based on performance levels of the second group corresponding to detection of the second detection type, wherein at least a portion of the first group and the second group comprise different models, were trained according to different training data, or were queried according to different inputs besides the unstructured data entity;
• generating, using at least a subset of the first group or the second group of the machine learning components, second representations of the unstructured data entity, wherein the second representations comprise second indications or second absences of the first detection type and the second detection type;
• resolving the second indications or the second absences of the first detection type using the first group;
• resolving the second indications or the second absences of the second detection type using the second group;
• determining any first discrepancy between the first indications or the first absences of the first detection type and the second indications or the second absences of the first detection type, wherein the any first discrepancy comprises any difference in an existence or an absence of the first detection type, in a relative position of the first detection type, or in a subtype or a subclassification of the first detection type;
• determining any second discrepancy between the first indications or the first absences of the second detection type and the second indications or the second absences of the second detection type, wherein the any second discrepancy comprises any difference in an existence or an absence of the second detection type, in a relative position of the second detection type, or in a subtype or a subclassification of the second detection type;
• selectively modifying, or selectively prompting the modification of, the first representation based on the any first discrepancy or the any second discrepancy;
• and
• performing one or more downstream actions based on the selectively modified first representation, wherein performing one or more downstream actions comprises:
• generating an output based on the selectively modified first representation.

A non-transitory computer readable medium comprising instructions that, when executed, cause at least one processor of one or more processors to perform:

• receiving a first representation of an unstructured data entity, wherein the first representation comprises first indications or first absences of a first detection type and a second detection type, the first detection type being different from the second detection type;
• assigning the first detection type to a first group of machine learning components based on performance levels of the first group corresponding to detection of the first detection type;
• assigning the second detection type to a second group of machine learning components based on performance levels of the second group corresponding to detection of the second detection type, wherein at least a portion of the first group and the second group comprise different models, were trained according to different training data, or were queried according to different inputs besides the unstructured data entity;
• generating, using at least a subset of the first group or the second group of the machine learning components, second representations of the unstructured data entity, wherein the second representations comprise second indications or second absences of the first detection type and the second detection type;
• resolving the second indications or the second absences of the first detection type using the first group;
• resolving the second indications or the second absences of the second detection type using the second group;
• determining any first discrepancy between the first indications or the first absences of the first detection type and the second indications or the second absences of the first detection type, wherein the any first discrepancy comprises any difference in an existence or an absence of the first detection type, in a relative position of the first detection type, or in a subtype or a subclassification of the first detection type;
• determining any second discrepancy between the first indications or the first absences of the second detection type and the second indications or the second absences of the second detection type, wherein the any second discrepancy comprises any difference in an existence or an absence of the second detection type, in a relative position of the second detection type, or in a subtype or a subclassification of the second detection type;
• selectively modifying, or selectively prompting the modification of, the first representation based on the any first discrepancy or the any second discrepancy;
• and
• performing one or more downstream actions based on the selectively modified first representation, wherein performing one or more downstream actions comprises:
• generating an output based on the selectively modified first representation.