Predictive natural language processing using semantic feature extraction

An apparatus comprising one or more processors and at least one non-transitory memory comprising program code, wherein the at least one non-transitory memory and the program code are configured to, with the one or more processors, cause the apparatus to:

• identify an indexed representation of a first natural language data object;
• identify a low-occurrence phrase in the indexed representation;
• identify, for the low-occurrence phrase, a label phrase, a low-occurrence phrase identifier, and a label phrase identifier for the label phrase;
• update the indexed representation by replacing the low-occurrence phrase identifier with the label phrase identifier;
• determine an individual frequency for each of one or more dominant phrases in the first natural language data object;
• determine a semantically-adjusted frequency for a first dominant phrase of the one or more dominant phrases based at least in part on the individual frequency for the first dominant phrase and an individual frequency for a dependent phrase for the first dominant phrase, wherein, when the first dominant phrase is associated with the dependent phrase:
• (i) the first dominant phrase comprises the dependent phrase and is larger in size than the dependent phrase, and (ii) the dependent phrase is a second dominant phrase of the one or more dominant phrases;
• generate a first structured representation of the first natural language data object based at least in part on a semantically-adjusted frequency associated with a third dominant phrase of the one or more dominant phrases;
• provide at least a portion of the first structured representation as input to a machine learning model;
• generate, using the machine learning model, a prediction corresponding to the first natural language data object;
• and
• provide the prediction to an end user computing entity, wherein the prediction is displayed by the end user computing entity.

A non-transitory computer storage medium comprising instructions configured to cause one or more processors to:

• identify an indexed representation of a first natural language data object;
• identify a low-occurrence phrase in the indexed representation;
• identify, for the low-occurrence phrase, a label phrase, a low-occurrence phrase identifier, and a label phrase identifier for the label phrase;
• update the indexed representation by replacing the low-occurrence phrase identifier with the label phrase identifier;
• determine an individual frequency for each of one or more dominant phrases in the first natural language data object;
• determine a semantically-adjusted frequency for a first dominant phrase of the one or more dominant phrases based at least in part on the individual frequency for the first dominant phrase and an individual frequency for a dependent phrase for the first dominant phrase, wherein, when the first dominant phrase is associated with the dependent phrase:
• (i) the first dominant phrase comprises the dependent phrase and is larger in size than the dependent phrase, and (ii) the dependent phrase is a second dominant phrase of the one or more dominant phrases;
• generate a first structured representation of the first natural language data object based at least in part on a semantically-adjusted frequency associated with a third dominant phrase of the one or more dominant phrases;
• provide at least a portion of the first structured representation as input to a machine learning model;
• generate, using the machine learning model, a prediction corresponding to the first natural language data object;
• and
• provide the prediction to an end user computing entity, wherein the prediction is displayed by the end user computing entity.