Unary Relation Extraction via Distant Supervision for Knowledge Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing binary relation extraction techniques require both entities to be nearby in the corpus, fail to infer relations based on different mentions, and cannot adapt representations to different relations, limiting their effectiveness in populating knowledge graphs.
Innovation Solution
The method employs distant supervision under a deep learning approach to train a model using a knowledge graph and unannotated corpus, predicting unary relations by identifying entities and generating context sets for a multi-class, multi-label deep neural network, which converts unary relations into binary relations for knowledge graph population.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If binary relation extraction techniques are used, then relations between entities can be extracted, but both entities must be nearby in the corpus which limits the ability to infer relations based on different mentions
Solution Approach 1:
The patent segments the relation extraction task into unary relation prediction for each entity independently, rather than requiring paired entity extraction. This allows the system to process entity mentions separately and infer relations based on individual entity contexts, enabling relation inference from different mentions throughout the corpus without requiring entities to be nearby each other.
Solution Approach 2:
The patent introduces an intermediary knowledge graph that stores pre-existing relational knowledge. This knowledge graph acts as a mediator between the corpus and relation extraction, allowing the system to infer relations by querying the knowledge graph with individual entity mentions rather than requiring direct entity pairs in the text, thus enabling relation inference from scattered mentions across the corpus.
2Productivity
If traditional relation extraction methods are used, then relations can be identified, but the representation cannot adapt to different relations which reduces effectiveness
Solution Approach 1:
The patent employs a dynamic neural network representation that can adapt to different relation types. The model uses learnable relation-specific embeddings and attention mechanisms that dynamically adjust based on the relation type being predicted, allowing the same base representation to effectively capture different semantic relations without requiring fixed hand-crafted features for each relation type.
Solution Approach 2:
The patent changes the parameters of the neural network representation based on the relation type. By using relation-specific embedding vectors and adjusting model parameters through training on different relation types, the system adapts its representation to capture the unique characteristics of each relation, improving effectiveness across diverse relation types in the knowledge graph.
3Ease of manufacture
If unary relation prediction is used, then the complexity of annotating corpora is reduced, but the model requires distant supervision which increases data processing requirements
Solution Approach 1:
The patent performs preliminary action by pre-populating a knowledge graph with existing relational data before training the unary relation prediction model. This pre-existing knowledge structure serves as training supervision, eliminating the need for manual corpus annotation while providing the model with target relations to learn from. The knowledge graph is prepared in advance to guide the learning process.
Solution Approach 2:
The system uses self-service by leveraging its own generated knowledge graph as training data. The model extracts relations from the corpus, populates the knowledge graph, and then uses this same knowledge graph to supervise further training of the unary prediction model. This self-reinforcing approach eliminates external annotation requirements while managing data processing through iterative refinement.
Data Source
AI summary
Methods and systems are described for populating knowledge graphs. A processor can identify a set of data in a knowledge graph. The processor can identify a plurality of portions of an unannotated corpus, where a portion includes at least one entity. The processor can cluster the plurality of portions into at least one data set based on the at least one entity of the plurality of portions. The processor can train a model using the at least one data set and the set of data identified from the knowledge graph. The processor can apply the model to a set of entities in the unannotated corpus to predict unary relations associated with the set of entities. The processor can convert the predicted unary relations into a set of binary relations associated with the set of entities. The processor can add the set of binary relations to the knowledge graph.