Multi-Modal Word Embedding for Relational Database Semantic Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems face challenges in efficiently managing and utilizing large amounts of multi-modal data, particularly in enabling artificial intelligence capabilities through cognitive processes that can effectively handle and analyze various types of data.
Innovation Solution
A method and system for supporting inductive reasoning queries over multi-modal data from relational databases by generating text representations of features and building a multi-modal word embedding model to capture relationships between different types of data, allowing for cognitive intelligence queries that leverage semantic contextual similarities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional data management systems are used to handle multi-modal data, then data storage is achieved, but the ability to perform cognitive analysis and extract semantic relationships is insufficient
Solution Approach 1:
The patent introduces word embedding models as an intermediary layer between traditional relational databases and cognitive query systems. This mediator transforms structured data into semantic vector representations, enabling the system to capture and preserve semantic relationships without requiring complete system redesign. The embedding model acts as a bridge that translates tabular data into a format suitable for cognitive analysis.
Solution Approach 2:
The patent applies parameter changes by transforming data from its original structured format into vector space representations with different dimensional parameters. By changing the representation parameters of data (from rows/columns to vector embeddings), the system enables new types of queries and analysis that were previously impossible with traditional data structures.
2Ease of operation
If multi-modal data is stored in relational databases, then data organization is achieved, but query capabilities for inductive reasoning and semantic matching are limited
Solution Approach 1:
The patent replaces traditional mechanical query processing (SQL-based exact matching) with neural network-based semantic reasoning. Instead of relying on rigid database join operations, the system uses word embedding models to perform semantic similarity calculations and inductive reasoning, enabling more flexible and accurate query responses.
Solution Approach 2:
The patent creates a composite system that combines relational database structures with neural network embedding models. This composite approach integrates the organizational benefits of structured databases with the semantic reasoning capabilities of AI models, achieving both data organization and advanced query capabilities.
3Loss of information
If large amounts of multi-modal data are processed, then comprehensive analysis is achieved, but computational efficiency decreases
Solution Approach 1:
The patent applies preliminary action by pre-computing word embedding vectors for all data elements before queries are executed. This preprocessing step creates ready-to-use semantic representations that can be quickly compared and analyzed during query processing, avoiding the need for computationally intensive calculations at query time.
Solution Approach 2:
The patent uses copying by creating vector embedding representations of the original data without moving or transforming the actual data storage. These copied vector representations enable efficient semantic comparisons while the original structured data remains intact in the database.
Data Source
AI summary
A system, apparatus, and a method for training with multi-modal data in a relational database, including generating a first database including a multi-view of the multi-modal data, retrieving a second set of data from an external source via a network, and training a first model according the first database and the second set of data. The first model outputs relationships of the first database with the multi-view and the second set of data.