Automatic identification of root cause and mitigation steps for incidents generated in an incident management system

1 . A computer implemented method, comprising:

receiving a first incident record indicative of a detected incident;

accessing a set of historic incident records indicative of historic incidents detected within a threshold time period, wherein the set of historic incident records include descriptive information that describes the historic incidents;

for a set of incident records including the set of historic incident records and the first incident record, identifying a set of incident record pairs of incident records and, for each incident record pair:

tokenizing information of the incident record pair; and

generating, from the tokenized information, an embedding for each incident record of the incident record pair, the embedding representing the incident record in a multi-dimensional embedding space that encodes a semantic meaning of the respective incident record;

for each incident record pair of the set of incident record pairs:

determining a distance metric between the embeddings of the associated incident records; and

if the distance metric is below a threshold distance, generating a link and a relatedness weight for the associated incident records;

identifying a set of related incident records, from the set of incident records, that are related to the first incident record, including:

clustering the first incident record and the set of related incident records based on each incident record of the set of related incident records being at least indirectly connected to the first incident record via the generated links, to obtain a cluster of incident records;

ranking the incident records in the cluster of incident records into a set of ranked incident records based on, for each incident record in the cluster of incident records, a number of links and associated relatedness weights connected to the incident record;

selecting one of the incident records of the cluster of incident records as a root cause incident record for the cluster of incident records based on rankings of the set of ranked incident records; and

generating a mitigation step, based on the root cause incident record, to mitigate the associated incident of the root cause incident record by:

generating a prompt to a generative artificial intelligence (AI) model, the prompt including an instruction to generate the mitigation step, the prompt further including context data, indicative of the ranked incident records; and

receiving from the generative AI model, as a response to the prompt, the mitigation step.

2 . The computer implemented method of claim 1 wherein identifying the set of related incident records comprises:

generating a similarity metric indicative of a similarity of the incident records in the set of related incident records; and

comparing the similarity metric to a threshold similarity metric to determine that the set of related incident records are similar to one another.

3 . The computer implemented method of claim 2 wherein generating the similarity metric comprises:

selecting a pair of incident records;

generating an embedding for each of the pair of incident records; and

calculating a distance between the embeddings, as the similarity metric.

4 . The computer implemented method of claim 1 and further comprising:

receiving a user-generated link between the first incident record and another incident record.

5 . The computer implemented method of claim 4 wherein clustering comprises:

clustering the first incident record and the set of related incident records based on the links and based on the user-generated links.

6 . The computer implemented method of claim 5 and further comprising:

for each link, assigning a weight to the link indicative of a relatedness of the incident records linked by the link.

7 . The computer implemented method of claim 6 wherein ranking comprises:

ranking the incident records in the cluster of incident records based on the links and assigned weights.

8 . The computer implemented method of claim 7 wherein ranking comprises:

applying the links and assigned weights to an artificial neural network (ANN) ranking model; and

receiving, from the ANN ranking model, a ranking indicator indicative of a ranking of the incident records.

9 . The computer implemented method of claim 1 wherein generating the mitigation step comprises:

generating the prompt to the generative AI model, the prompt identifying the root cause incident record, and the instruction instructing the generative AI model to generate a step-by-step mitigation plan to mitigate the associated incident of the root cause incident record; and

receiving, from the generative AI model, as a response to the prompt, a set of computer executable instructions that are executed to perform the step-by-step mitigation plan.

10 . A computing system, comprising:

an incident linking system configured to receive a first incident record and access a set of historic incident records to identify a set of related historic incident records that are related to the first incident record, and wherein the set of historic incident records include descriptive information that describe associated historic incidents of the historic incident records;

an identification system configured to, for a set of incident records including the set of historic incident records and the first incident record, identify a set of incident record pairs of incident records;

a tokenizing system configured to, for each incident record pair, tokenize information of the incident record pair and generate, from the tokenized information, an embedding for each incident record of the incident record pair, the embedding representing the incident record in a multi-dimensional embedding space that encodes a semantic meaning of the respective incident record;

a distance metric determination system configured to, for each incident record pair of the set of incident record pairs, determine a distance metric between the embedding of the associated incident records and, if the distance metric is below a threshold distance, generate a link and a relatedness weight for the associated incident records;

an incident clustering system configured to cluster the first incident record and the set of related historic incident records based on the links to obtain a cluster of incident records;

an incident ranking system configured to rank the first incident record and the set of related historic incident records in the cluster of incident records into a ranked incident records; and

a root cause processing system configured to identify one incident record of the cluster of incident records as a root cause incident record for the cluster of incident records based on the ranked incident records, and to generate a mitigation step, based on the root cause incident record, to mitigate an associated root cause incident of the root cause incident record, wherein generating the mitigation step comprises:

generating a prompt to a generative artificial intelligence (AI) model, the prompt including an instruction to generate the mitigation step, the prompt further including context data, indicative of the ranked incident records; and

receiving from the generative AI model, as a response to the prompt, the mitigation step.

11 . The computing system of claim 10 wherein the incident linking system is configured to identify the set of related historic incident records, generate a similarity metric indicative of a similarity of the incident records in the set of related historic incident records, and further comprising:

a related/unrelated classifier configured to compare the similarity metric to a threshold similarity metric to determine that the related historic incident records are similar to one another.

12 . The computing system of claim 10 wherein the clustering system is configured to receive a user-generated link between the first incident record and another incident record and to cluster the first incident record and the set of related historic incident records based on the links and based on the user-generated links.

13 . The computing system of claim 12 wherein the incident linking system comprises:

a weight generator configured to, for each link, assign a weight to the link indicative of a relatedness of the incident records linked by the link.

14 . The computing system of claim 13 wherein the incident ranking system is configured to rank the incident records in the cluster of incident records based on the links and assigned weights.

15 . The computing system of claim 14 wherein the incident ranking system comprises:

an artificial neural network (ANN) ranking model configured to receive the links and assigned weights and generate a ranking indicator indicative of a ranking of the incident records.

16 . The computing system of claim 15 wherein the root cause processing system comprises an artificial intelligence (AI) model and further comprising:

a prompt generation system configured to generate a prompt to the AI model, the prompt including an instruction portion with an instruction to generate a mitigation plan for the associated root cause incident and a context data portion with context data including the ranked incident records, the AI model generating a response to the prompt, the response indicating the associated root cause incident and the mitigation plan.

17 . A computer system, comprising:

a link generator configured to generate links, each link being between incident records in a set of related incident records, wherein the set of related incident records include descriptive information that describe an associated incident of each incident record, wherein the links are links in a multi-dimensional embedding space that indicates a semantic meaning of the respective incident records;

an incident rank processor configured to rank the incident records of the set of related incident records into a ranked incident records based on a distance metric of the links in the multi-dimensional embedding space; and

a root cause identification processor configured to identify one incident record of the set of related incident records as a root cause incident record for the set of related incident records based on the ranked incident records, and to generate a mitigation step, based on the root cause incident record, wherein generating the mitigation step comprises:

generating a prompt to a generative artificial intelligence (AI) model, the prompt including an instruction to generate the mitigation step, the prompt further including context data, indicative of the ranked incident records; and

receiving from the generative AI model, as a response to the prompt, the mitigation step.

18 . The computer system of claim 17 wherein the root cause identification processor comprises:

a prompt generation system that generates a prompt with an instruction portion including an instruction to generate a mitigation plan to mitigate an associated root cause incident of the root cause incident record and a context data portion with context data including the ranked incident records; and

a large language model configured to generate the mitigation plan based on the prompt.