Multivariate Embedding System for Incident Similarity Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing systems lack the ability to efficiently identify historically similar incidents, limiting the speed and accuracy of incident resolution by relying solely on incident descriptions without considering additional relevant data points.
Innovation Solution
A system utilizing multivariate embeddings, including a large language model, processes incident data to generate embeddings based on multiple features such as short description, business category, sub-category, line of business, and incident priority, and applies a Euclidean distance formula to identify historically similar incidents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing computing systems rely solely on incident descriptions to identify similar incidents, then the system complexity remains low, but the accuracy of identifying similar incidents deteriorates
Solution Approach 1:
The patent transforms incident data from a single-dimensional description into a multi-dimensional embedding space by converting multiple incident attributes (description, business category, sub-category, line of business, priority) into numerical embeddings using a large language model. This dimensional transformation enables the system to capture complex relationships and similarities that cannot be achieved through traditional single-description comparison methods.
Solution Approach 2:
The patent introduces a large language model as an intermediary component that bridges the gap between raw incident data and similarity identification. The LLM processes multiple incident attributes and generates embeddings that serve as intermediate representations, enabling accurate similarity computation without directly comparing complex multi-attribute data structures.
2Measurement precision
If existing systems do not have the ability to find historically similar incidents, then the incident resolution speed remains fast (no search needed), but the incident resolution accuracy deteriorates
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing embeddings for historical incident data in a vector database. This allows the system to quickly query and retrieve similar incidents during actual incident resolution without performing computationally intensive embedding calculations in real-time, thus reducing resolution time while improving accuracy.
Solution Approach 2:
The patent replaces traditional mechanical search methods (manual or simple keyword search) with an automated embedding-based similarity computation system. By substituting the mechanical process of searching through descriptions with automated vector distance calculations, the system achieves both faster and more accurate incident identification.
3Measurement precision
If the system processes multiple incident attributes through embedding conversion, then the accuracy of similar incident identification improves, but the data processing complexity increases
Solution Approach 1:
The patent creates a simplified copy representation of complex incident data through embedding vectors. Instead of working with the original complex multi-attribute data structures during similarity comparison, the system uses compressed embedding vectors that capture the essential characteristics, reducing processing complexity while maintaining identification accuracy.
Data Source
AI summary
A method for finding historically similar incidents is disclosed. The method includes receiving a plurality of historical data objects corresponding to a plurality of previous incidents, each of the plurality of historical data objects indicating an occurrence of a previous incident and including previous incident multivariate information; converting the historical data objects to a plurality of historical embeddings; storing the plurality of historical embeddings in an index, the index being a vector embedding database; receiving a data object indicating an occurrence of a current incident associated with a configurable item, the data object including multivariate information; converting the data object into a first embedding by utilizing the large language model; identifying a set of historically similar incidents by applying a Euclidean distance formula to the first embedding and the plurality of historical embeddings; and determining a similarity score for each of the set of historically similar incidents.