Joint Entity Relation Extraction via Recurrent Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing natural language processing methods are inadequate for handling multiple relation instances in a sentence and fail to effectively detect entity mention pairs participating in relations, as they treat relation classification as a sentence-level multi-class classification problem without considering the interdependencies between entity recognition and relation classification.
Innovation Solution
A device comprising a recurrent neural network and an entity-relation table that jointly performs entity and relation extraction, using a forward and backward neural network to provide context representations for words, allowing for the interdependencies between entity recognition and relation classification to be modeled, thereby alleviating the need for search heuristics and explicit label dependencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If relation classification is treated as a sentence-level multi-class classification problem with a single relation instance, then the classification process is simplified, but the method becomes incapable of handling multiple relation instances in a sentence
Solution Approach 1:
The patent segments the sentence-level classification problem into word-pair level classifications by creating an entity-relation table with n x n cells, where each cell corresponds to a specific word pair (wi, wj). This allows multiple relation instances to be independently classified and detected within the same sentence processing framework.
Solution Approach 2:
The patent transitions from a single sentence-level classification dimension to a two-dimensional word-pair classification dimension by introducing the entity-relation table structure. This dimensional expansion enables simultaneous representation of multiple relations and entities within the sentence without increasing overall computational complexity significantly.
2Ease of manufacture
If entity recognition and relation classification are performed as separate subtasks in a pipeline, then each subtask can be optimized independently, but the interdependencies between the two tasks are lost
Solution Approach 1:
The patent merges entity recognition and relation classification into a unified joint extraction framework using a single recurrent neural network that processes both tasks simultaneously. The shared neural network architecture allows the model to learn interdependencies between entity recognition and relation classification, improving overall extraction accuracy while maintaining efficient processing.
3Productivity
If conventional relation classification methods are used, then the processing is computationally efficient for single relations, but the methods fail to detect corresponding entity mention pairs participating in the relation
Solution Approach 1:
The recurrent neural network in the patent is designed to perform multiple functions simultaneously: it classifies relations between word pairs, identifies entity mentions, and detects their participations in relations. This multi-functional approach maintains processing efficiency while significantly improving the detection of entity mention pairs through the unified processing framework.
Data Source
AI summary
A device for relation extraction in a natural language sentence having n words is suggested, the device comprising: a recurrent neural network for joint entity and relation extractions of entities and relations between the entities in the sentence, and an entity-relation table for storing entity labels for the entities and relation labels for the relations, wherein both the entity labels and the relation labels are defined as instances of binary relationships between certain words wi and wj in the sentence, with i∈[1, . . . , n], and j∈[1, . . . , n], wherein each of the entity labels is a first binary relationship for i=j, and wherein each of the relation labels is a second binary relationship for i≠j, wherein the recurrent neural network is configured to fill the cells of the entity-relation table and includes a forward neural network and a backward neural network.