Seeded Topic Modeling for Instrumentation Event Categorization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated data-collection-and-data-processing systems face challenges in efficiently categorizing and summarizing the high volume and complexity of events generated by instrumented web pages, leading to difficulties in extracting meaningful information for website optimization and user experience improvement.
Innovation Solution
The implementation of a seeded local/global-topic latent Dirichlet allocation method to discover topics in instrumentation-generated events, assigning topics to events, and generating topic signatures for subsequent message processing, facilitating efficient categorization and summarization of events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated data collection and processing systems are used to handle instrumentation-generated events, then data processing capability is improved, but the system becomes overwhelmed by the volume and complexity of events
Solution Approach 1:
The patent extracts only the most salient and useful events from the massive stream of instrumentation-generated events. The system filters and selects events based on their importance and relevance, extracting key information while discarding redundant data. This extraction approach allows the system to handle high event volumes without being overwhelmed, focusing processing resources on meaningful events only.
2Loss of information
If comprehensive event collection is performed to capture all user interactions, then information completeness is improved, but information extraction becomes difficult
Solution Approach 1:
The patent introduces an intermediary abstraction layer that sits between comprehensive event collection and information extraction. This intermediary layer organizes raw events into structured formats, applies filtering criteria, and prepares data for analysis. By inserting this intermediary processing stage, the system maintains complete event collection while making the extracted information much more accessible and easier to analyze.
3Measurement precision
If all instrumentation-generated events are processed in detail, then analysis accuracy is improved, but processing time increases
Solution Approach 1:
The patent segments the event processing pipeline into distinct stages: initial filtering, prioritization, detailed analysis, and aggregation. By dividing the processing workload into segments, the system can quickly filter out irrelevant events, focus detailed analysis only on high-priority events, and parallelize processing where possible. This segmentation maintains analysis accuracy for critical events while significantly reducing overall processing time.
Data Source
AI summary
The current document is directed to methods and systems that receive instrumentation-generated events and that employ statistical inference to discover event topics and to assign a topic or category to each of a number of events. In a described implementation, the events comprise key/value pairs. A seeded local/global-topic latent Dirichlet allocation methods is used to discover topics and assign topics to a set of events. The topic-assigned events are then processed to generate topic signatures, using which the methods and systems assign topics to subsequently received messages.