Data Product Search with Knowledge-Graph Interestingness Scoring

Solution Overview

Problem

Data marketplaces face challenges in ranking dataset search results reliably due to insufficient user traffic, leading to unreliable conventional metrics like likes, reviews, and downloads, and 'cold start' issues with low user counts.

Innovation Solution

A system utilizing Dirichlet-Hawkes Process (DHP) to generate a lineage graph and a relational graph convolutional network (RGCN) model for semantic relevance learning, mapping user queries to knowledge and lineage graphs to generate an interestingness score for dataset recommendations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional metrics (likes, reviews, downloads) are used to rank datasets, then the ranking process is simple to implement, but the ranking accuracy is unreliable due to insufficient user traffic and cold start issues

Engineering Contradiction:

Improveranking accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces an interestingness score as an intermediary metric that bridges the gap between conventional unreliable metrics and accurate dataset ranking. The interestingness score is computed by mapping entities to knowledge graphs and lineage graphs, then calculating scores based on node matches and dataset properties. This intermediary scoring mechanism enables reliable ranking even when user traffic is insufficient for conventional metrics.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system performs preliminary actions by pre-computing and storing knowledge graphs and lineage graphs before ranking is needed. The knowledge graph contains pre-extracted entities and relationships from dataset metadata, while the lineage graph pre-stores dataset provenance information. These pre-computed structures enable fast and accurate ranking without requiring real-time complex computations when user queries are received.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If metadata and lineage information are leveraged to improve ranking accuracy, then measurement precision improves, but device complexity increases due to additional processing requirements

Engineering Contradiction:

Improveranking accuracyVSAvoidcomputational resources

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary actions by pre-computing and storing knowledge graphs and lineage graphs before ranking is needed. The knowledge graph contains pre-extracted entities and relationships from dataset metadata, while the lineage graph pre-stores dataset provenance information. These pre-computed structures enable fast and accurate ranking without requiring real-time complex computations when user queries are received.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent extracts only the essential and relevant information from dataset metadata and lineage information needed for ranking. The knowledge graph extraction focuses on identifying key entities and their relationships, while the lineage graph extraction concentrates on provenance paths relevant to dataset quality. This selective extraction reduces computational overhead while maintaining ranking accuracy.

Inventive Principle:

Principle #2Taking out (Extraction)

3Reliability

If semantic relevance learning is implemented to map queries to knowledge graphs, then the quality of search results improves, but the complexity of the search system increases

Engineering Contradiction:

Improvesearch result qualityVSAvoidsearch system complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces an interestingness score as an intermediary metric that bridges the gap between conventional unreliable metrics and accurate dataset ranking. The interestingness score is computed by mapping entities to knowledge graphs and lineage graphs, then calculating scores based on node matches and dataset properties. This intermediary scoring mechanism enables reliable ranking even when user traffic is insufficient for conventional metrics.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20250299237A1Searching and exploring data products by popularity

Publication Date: 2025.09.25 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

A computer-implemented method may include identifying an entity in a user query or a user profile; mapping, via a relational graph convolutional network model, the entity to a knowledge node in a knowledge graph; mapping, via a semantic relevance learning engine, the knowledge node of the knowledge graph to a lineage node of a lineage graph; generating a list of matched nodes from the mapping of the knowledge node of the knowledge graph to the lineage node of the lineage graph; generating a interestingness score for a dataset associated with the list of matched nodes; identifying a ranked dataset recommendation based on the interestingness score; and communicating instructions to communicate the interestingness score and the ranked dataset recommendation in a user interface.