Similarity Index Tree for Massive High-Dimension Dataset Queries

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods are inefficient in handling massive, high-dimensional datasets for similarity-based queries, self-joins, and joins due to high computational complexity, particularly with the need to avoid the n^2 pairwise similarity computation phase.

Innovation Solution

The development of a method and apparatus for building a bulk similarity index tree, enabling fast similarity-based queries, self-joins, and joins through efficient hierarchical grouping and the use of anchor items to represent clusters, reducing computational complexity to O(n*log(n)) and O(n1*log(n1) + n2*log(n2)) respectively.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If pairwise similarity computation is performed for all data members, then similarity accuracy is improved, but computational complexity increases to O(n^2)

Engineering Contradiction:
Improvesimilarity accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the dataset into clusters using hierarchical grouping, where each cluster is represented by an anchor item. Instead of computing pairwise similarity between all n items, the system divides items into groups and only computes similarity between cluster representatives (anchors), reducing the computational scope from O(n^2) to O(n*log(n)) for query operations and O(n1*log(n1) + n2*log(n2)) for join operations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces anchor items as intermediaries between query items and the dataset. Each cluster is represented by an anchor item that serves as a mediator: similarity computation is performed between the query item and anchor items first, then results are propagated to member items within clusters. This intermediary approach avoids direct pairwise comparison between all items while maintaining similarity accuracy through the transitive property within clusters.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Speed

If hierarchical grouping with anchor items is used, then query speed is improved, but index building complexity increases

Engineering Contradiction:
Improvequery speedVSAvoidindex building complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent performs preliminary action by pre-building a hierarchical index structure with anchor items before query execution. The index construction phase groups items into clusters and selects representative anchor items in advance. This preliminary organization enables fast query execution by avoiding runtime clustering computations, trading off one-time index building complexity for sustained query speed improvements.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If all pairwise similarities are computed, then join accuracy between datasets is improved, but processing time increases significantly

Engineering Contradiction:
Improvejoin accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent extracts and utilizes the transitive property of similarity within clusters to avoid redundant computations. By computing similarity between cluster representatives (anchors) and propagating results to member items, the system extracts the essential similarity information without performing all pairwise comparisons. This extraction approach maintains join accuracy while dramatically reducing processing time from O(n1*n2) to O(n1*log(n1) + n2*log(n2)).

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP2035917B1Method and apparatus for fast similarity-based query, self-join, and join for massive, high-dimension datasets
Publication Date: 2014.05.21 NAHAVA
  • EP2035917B1 patent drawingFigure 1
  • EP2035917B1 patent drawingFigure 2
  • EP2035917B1 patent drawingFigure 3

AI summary

A method and apparatus for fast similarity-based query, self-join, and join for massive high-dimension datasets utilizing a similarity index (400).