Heterogeneous Data Indexing via Relation-Aware Hashing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hashing technologies are inefficient for searching and indexing heterogeneous data, such as social media networks, which contain diverse data types and relationships, as they are primarily designed for homogeneous data.
Innovation Solution
The implementation of a Relation-aware Heterogeneous Hashing (RaHH) index that learns hash functions for each data type and their mappings, allowing for efficient retrieval of similar entities within and across different data domains by generating separate hash tables for each domain and mapping between them.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing hashing technologies are applied to heterogeneous data, then the indexing structure can be established, but the search efficiency and accuracy deteriorate due to data type differences
Solution Approach 1:
The patent segments the heterogeneous data into multiple homogeneous data domains (e.g., text, image, audio) and creates separate hash tables for each domain. This allows each hash table to be optimized for its specific data type while maintaining overall system capability to handle heterogeneous data through domain-specific indexing structures.
Solution Approach 2:
The patent implements a universal hashing framework that can handle multiple data types by learning domain-specific hash functions for each data domain and establishing mappings between them. The system achieves multi-functionality by enabling search operations across different data types through a unified architecture that respects the heterogeneity of the data.
2Device complexity
If a single hash function is used for all data types, then the indexing structure is simple, but the search precision deteriorates due to inability to capture domain-specific features
Solution Approach 1:
The patent applies local quality by learning and using domain-specific hash functions for each data domain (text, image, audio, etc.). Each hash function is optimized to capture the essential features of its specific data type, thereby improving search precision for that domain while maintaining a manageable overall structure through modular organization.
3Measurement precision
If domain-specific hash functions are learned for each data type, then the search precision improves, but the system complexity increases due to multiple hash tables and mappings
Solution Approach 1:
The patent merges multiple domain-specific hash tables and their corresponding hash functions into a unified framework. By establishing mappings between different data domains and integrating them into a single search system, the patent achieves high search precision across heterogeneous data while managing system complexity through consolidation and coordinated operation of the various components.
Data Source
AI summary
A method of performing a search of heterogeneous data based on an input query includes: generating an index including at least two hash tables, where each hash table corresponds to a different data domain of the heterogeneous data and includes hash code sets, where at least one of the hash code sets is mapped to a hash code set of another one of the tables. The method further includes performing a hash on the input query to generate a hash code, by referring to the index, determining a first hash code set that the generated hash code belongs to, and determining a second hash code set that the determined first hash code set is mapped to, and providing at least one result based on the determined second hash code set.