Abstract
Helix is a system designed to address the pervasive "data silo" problem in large enterprises by transforming ontology alignment into a collaborative, "pay-as-you-go" process. Rather than requiring massive upfront manual effort to integrate heterogeneous data stores, Helix embeds alignment tasks into the standard data exploration workflow. This paper describes the Helix architecture, focusing on guided navigation, result preservation via a "shelf" mechanism, and human-in-the-loop linking to incrementally build a unified semantic model across disparate datasets.
Introduction
In large-scale enterprise environments, data is frequently trapped in a maze of heterogeneous silos, each characterized by unique formats, schemas, and query languages. To make this data accessible to non-experts, researchers typically employ ontology alignment—the process of identifying semantic correspondences between concepts in different datasets (e.g., mapping "Location" in one database to "Address" in another).
Traditional approaches to ontology alignment face a significant scalability gap. Automated tools are often noisy and error-prone, while manual reconciliation is labor-intensive. Most existing systems require these alignments to be finalized before the data becomes queryable. Helix addresses this gap by treating data integration as a continuous, user-driven byproduct of everyday data exploration.
The Helix System and Workflow
The core philosophy of Helix is "progressive alignment." The system abstracts the underlying complexity of various query languages and data formats, allowing users to contribute to the technical task of ontology mapping without requiring expertise in SQL or SPARQL. The workflow is centered around three primary features:
- Guided Navigation: Users search and navigate data through a uniform interface. The system provides suggestions based on existing semantic and schematic links to facilitate discovery across different stores.
- Result Preservation (The Shelf): Users can "bookmark" or save specific results of interest to a personal "shelf," effectively staging data for further analysis or integration.
- Guided Linking: When a user has multiple sets of saved results on their shelf, Helix guides them through the process of linking them. This human-in-the-loop approach ensures high-quality alignments that are then saved and shared, contributing to the global semantic model.
Collective Intelligence
Helix is designed to leverage collective intelligence; every user action benefits the broader community. Once a link is created between two datasets, that alignment becomes part of the system's shared knowledge base, improving future navigation and automated suggestions for all users. By embedding these tasks into the search and discovery process, the "cost" of data integration is distributed across the user base over time.
Implementation and Future Work
The Helix interface demonstrates a unified semantic model over virtualized data, successfully allowing users to query across data stores transparently. While the current system relies on users to manually initiate the linking of saved results, future development is focused on a recommender system. This enhancement will automatically suggest relevant saved results or existing alignments to users as they work, further reducing the friction of data reuse and discovery within the enterprise.
Summary updated