Enterprise Knowledge Graph Mining with BERT-Based NER
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing knowledge graph systems face challenges in accurately mining information from enterprises due to unique vocabularies, private project names, and non-standard use of words, leading to inaccurate search results and inefficient use of time and resources, particularly in industries with domain-specific knowledge.
Innovation Solution
The implementation of a multi-toolkit enterprise mining system using neural entity recognition and multiple AI toolkits, including an Enterprise Named Entity Recognition (ENER) model based on BERT, to mine and annotate enterprise knowledge graphs, addressing domain specificity and entity recognition challenges by leveraging diverse data sources and user context.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional NER systems are used with publicly available training corpora, then the system can recognize common entities like people, locations, and organizations, but it fails to accurately identify enterprise-specific entities such as projects and teams
Solution Approach 1:
The patent changes the training data parameter by using enterprise-specific documents and confidential information as training corpora instead of public news data. This allows the NER system to learn and recognize enterprise-specific entities like internal projects, teams, and proprietary terminology, thereby improving both accuracy for enterprise contexts and adaptability to private domain knowledge
Solution Approach 2:
The patent segments the entity recognition task into multiple specialized NER models, each trained on specific enterprise domains or entity types. This segmentation allows different models to specialize in recognizing particular kinds of enterprise entities (e.g., one model for projects, another for teams), improving overall recognition accuracy while maintaining adaptability across diverse enterprise contexts
2Adaptability or versatility
If a single NER tool is used, then the system is simple to implement, but it cannot cover diverse entity types and data sources effectively
Solution Approach 1:
The patent implements a universal NER framework that can handle multiple entity types (projects, teams, products, users) and diverse data sources (OneDrive, emails, Yammer, Teams, external wikis) through a common architecture. This multi-functional system uses standardized processing pipelines and unified data models, allowing it to adapt to different entity types and sources without requiring completely separate systems for each, thus achieving versatility while controlling complexity through design standardization
3Measurement precision
If confidential enterprise information is used to train NER models, then the system can achieve high accuracy for enterprise-specific entities, but the training data cannot be publicly shared
Solution Approach 1:
The patent introduces an intermediary layer that processes and anonymizes enterprise training data before it is used to train NER models. This intermediary layer extracts and removes sensitive confidential information while preserving the structural patterns and entity relationships needed for training. The resulting anonymized training data can then be shared or used without exposing proprietary enterprise information, maintaining both accuracy and information security
Data Source
AI summary
Examples described herein generally relate to a computer system including a knowledge graph storing a plurality of entities. A mining of a set of enterprise source documents within an enterprise intranet is performed, by an enterprise named entity recognition (ENER) model, to determine a plurality of entity names. An entity record is generated within a knowledge graph for a mined entity name from the linked entity names based on an entity schema and ones of the set of enterprise source documents associated with the mined entity name. The entity record includes attributes aggregated from the ones of the set of enterprise source documents associated with the mined entity name.