Phrase Generation Model Using Encoder Decoder for Relationship Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional relationship estimation techniques cannot generate a phrase having a relationship defined by a given input phrase and label, as they require pre-prepared phrase combinations and relational scores, limiting their ability to produce phrases from input pairs {phrase 1, label}.
Innovation Solution
A learning device and method for a phrase generation model that uses a combination of phrases and conjunctive expressions as training data, incorporating an encoder to convert phrases into vectors and a decoder to generate phrases based on relational labels, enabling the generation of phrases with specified relationships from input pairs {phrase, label}.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional relationship estimation techniques are used to output relational scores from 3-tuple inputs, then relational scores can be calculated, but it is not possible to generate a phrase 2 from {phrase 1, label} as input
Solution Approach 1:
The patent applies universality by designing a single phrase generation model that can perform multiple functions: both generating phrases with specific relationships and estimating relational scores. The model accepts different input formats (2-tuple or 3-tuple) and produces appropriate outputs (generated phrase or relational score), making it versatile enough to handle various relationship estimation tasks without requiring separate specialized systems.
Solution Approach 2:
The patent inverts the conventional approach by switching from a score-estimation paradigm to a phrase-generation paradigm. Instead of taking three inputs and outputting a score, the model takes two inputs ({phrase 1, label}) and generates the missing phrase 2. This inversion transforms the problem from evaluation to creation, enabling the system to actually produce relevant phrases rather than just scoring pre-formed combinations.
2Productivity
If pre-prepared phrase combinations are used for relationship estimation, then relational scores can be obtained, but the process requires manual preparation of phrase 2 combinations
Solution Approach 1:
The patent applies self-service by enabling the model to automatically generate the phrase 2 component that would otherwise require manual preparation. Given an input phrase 1 and a relationship label, the model autonomously produces a semantically appropriate phrase 2 that satisfies the specified relationship, eliminating the need for external manual intervention in creating training data combinations or test cases.
Solution Approach 2:
The patent applies preliminary action by pre-training the model on large datasets of phrase combinations and relationships, so that when deployed, it can immediately generate appropriate phrases without requiring on-the-spot manual preparation. The model learns from pre-prepared training data during the training phase, enabling automatic generation during the operational phase.
3Measurement precision
If conventional techniques require manual generation of 3-tuple data for learning, then relationship estimation accuracy can be improved, but the data preparation process becomes complex and time-consuming
Solution Approach 1:
The patent applies copying by using pre-existing large-scale datasets (such as CommonSenseQA or ConceptNet) as training data. Instead of manually creating 3-tuple data from scratch, the model learns from copied or adapted existing knowledge bases and datasets, significantly reducing the complexity and time required for data preparation while maintaining or improving estimation accuracy through exposure to diverse relationship patterns.
Data Source
AI summary
A learning device of a phrase generation model includes a memory; and a processor configured to execute learning the phrase generation model including an encoder and a decoder, by using, as training data, a 3-tuple. The 3-tuple includes a combination of phrases and at least one of a conjunctive expression representing a relationship between the phrases, and a relational label indicating the relationship represented by the conjunctive expression. The encoder is configured to convert a phrase into a vector from a 2-tuple. The 2-tuple includes a phrase and at least one of the conjunctive expression and the relational label. The decoder is configured to generate, from the converted vector and the conjunctive expression or the relational label, a phrase having the relationship represented by the conjunctive expression or the relational label with respect to the phrase.