Machine learning for detecting and modifying faulty controls

A method comprising:

• obtaining incident data indicative of a computing system malfunction, wherein the incident data comprises text description of an incident associated with the incident data;
• generating a first embedding based on the incident data, the first embedding comprising an incident embedding that is an embedding of at least a portion of the incident data;
• determining that the first embedding and a second embedding associated with a control satisfy a first similarity threshold, the second embedding comprising a control embedding that is an embedding of at least a portion of the control;
• based on the first embedding and the second embedding satisfying the first similarity threshold, obtaining second text associated with the control;
• generating a training sample comprising the second text and a modification to the second text for preventing future incidents;
• and
• training, based on a training dataset comprising the training sample, a machine learning model to recommend modifications for faulty controls,
• wherein training the machine learning model comprises modifying one or more weights of the machine learning model through backpropagation based on a determination that a classification of the machine learning model for a respective training sample of the training dataset does not match a label of the respective training sample.

One or more non-transitory computer-readable media comprising instructions that, when executed by one or more processors, causes operations comprising:

• obtaining incident data indicative of a computing system malfunction, wherein the incident data comprises text description of an incident associated with the incident data;
• generating a first embedding based on the incident data, the first embedding comprising an incident embedding that is an embedding of at least a portion of the incident data;
• determining that the first embedding and a second embedding associated with a control satisfy a first similarity threshold, the second embedding comprising a control embedding that is an embedding of at least a portion of the control;
• based on the first embedding and the second embedding satisfying the first similarity threshold, obtaining second text associated with the control;
• generating a training sample comprising the second text and a modification to the second text for preventing future incidents;
• and
• training, based on a training dataset comprising the training sample, a machine learning model to recommend modifications for faulty controls,
• wherein training the machine learning model comprises modifying one or more weights of the machine learning model through backpropagation based on a determination that a classification of the machine learning model for a respective training sample of the training dataset does not match a label of the respective training sample.