Multi-source dataset of e-commerce products with attributes for property matching

Area of Science:

• Data integration and knowledge representation
• Information retrieval and data management

Background:

• Schema/ontology matching is crucial for integrating heterogeneous data sources, especially with open data initiatives.
• Data property matching, a key task, faces challenges due to differing property names and complex relationships (1..n).
• Existing evaluation datasets are insufficient for diverse scenarios and training context-independent supervised techniques.

Purpose of the Study:

• To address the need for varied evaluation datasets for data property matching techniques.
• To support the development and training of robust schema/ontology matching methods.
• To facilitate context-independent supervised learning for data property matching.

Main Methods:

• A collection dataset of product records was created from four different contexts.
• Two existing datasets were transformed and processed, with noisy properties filtered out.
• The dataset comprises JSON files of product records and properties, with a separate grouping of matching properties.

Main Results:

• The processed dataset contains information on 2860 entities and 4386 properties.
• A total of 13350 pairwise matches between properties were identified.
• The dataset provides a diverse resource for evaluating and training data property matching.

Conclusions:

• The presented dataset is a valuable resource for advancing data property matching techniques.
• It enables more comprehensive evaluation and effective training of supervised learning models.
• Facilitates improved data integration across heterogeneous sources.