Conversation Knowledge Graph Building for Accurate Organisational Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for visualizing meetings and extracting organizational knowledge from conversation data fail to accurately distinguish and disregard programmatic messages, product/service mentions, and organizational structure, leading to inaccurate analysis and retrieval of conversation data.
Innovation Solution
A computer-implemented method to build a graph of organization-specific knowledge by detecting and clustering recurring words and phrases in conversation data, using natural language processing and neural networks to identify and merge nodes representing organizational knowledge, creating a hierarchical ontology-based graph.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If programmatic messages are included in conversation analysis, then complete conversation data is captured, but analysis accuracy deteriorates due to irrelevant content
Solution Approach 1:
The patent segments conversation data into distinct types by detecting and classifying programmatic messages separately from actual conversation content. This segmentation allows the system to process only relevant conversation portions for analysis, improving accuracy while maintaining awareness of complete data structure.
Solution Approach 2:
The patent extracts and removes programmatic messages (hold messages, IVR, disclaimers) from the conversation data stream before analysis. This extraction eliminates irrelevant content that would otherwise degrade analysis accuracy while preserving the complete original data for reference.
2Measurement precision
If manual configuration of organizational knowledge is required, then knowledge accuracy is improved, but system complexity and maintenance burden increase
Solution Approach 1:
The patent implements self-service by automatically building the organizational knowledge graph from the conversation data itself. The system extracts organizational entities (products, services, departments) and their relationships autonomously without requiring manual configuration, thereby maintaining detection accuracy while eliminating configuration complexity.
Solution Approach 2:
The patent performs preliminary action by pre-processing conversation data to extract and structure organizational knowledge before actual analysis tasks. This preliminary extraction creates a ready-to-use knowledge graph that improves subsequent detection accuracy without requiring manual setup at analysis time.
3Loss of information
If all conversation data is processed for analysis, then comprehensive insights are obtained, but processing time and computational resources increase
Solution Approach 1:
The patent extracts and separates programmatic messages from conversation data before processing. This extraction removes redundant content that would consume processing time without contributing to analytical insights, thereby reducing processing time while maintaining information completeness through selective processing.
Solution Approach 2:
The patent applies partial action by processing only the relevant portions of conversation data (actual speech content) rather than all data including programmatic messages. This selective processing reduces computational overhead and time while maintaining sufficient information completeness for accurate analysis.
4Reliability
If organizational knowledge is manually input, then knowledge reliability is improved, but ease of operation deteriorates
Solution Approach 1:
The patent enables the system to self-generate organizational knowledge by automatically extracting entities and relationships from conversation data. This eliminates the need for manual user input while maintaining knowledge reliability through data-driven extraction, thereby dramatically improving ease of operation.
Solution Approach 2:
The patent creates a copied representation of organizational knowledge directly from conversation data rather than requiring original manual input. This copying approach extracts organizational entities and relationships that already exist in the data, ensuring reliability while eliminating manual effort.
Data Source
AI summary
A computer-implemented method of building a graph of organisation-specific knowledge includes the step of receiving conversation data. Words and phrases within the conversation data that describe organisational knowledge are detected. Sets of candidate knowledge tuples are extracted. Each tuple includes a sequence of elements that represent respective words, phrases, or sentences that describe an item of knowledge. The elements in each tuple are ordered according to an ontology of the item of knowledge. The tuples in each set describe respective versions of the item of organisational knowledge. Exemplar tuples that correspond to candidate tuples in respective sets that contain elements that occur more than a threshold number of times in the text are created. The graph is initialised from nodes representing the elements of the exemplar tuples. Edges represent relationships between the elements. Sets of nodes that represent the same element of organisational knowledge are progressively determined and merged into exemplary nodes.