Semantic Search in RealFoodTrade

Andrea Calì, Roberto De Virgilio, Tommaso Di Noia, Luca Menichetti, Roberto Mirizzi, Luca Nardini, Vito Claudio Ostuni, Fabrizio Rebecca, Marco UnganiaAmw 2014

Abstract

RealFoodTrade (RFT) is an innovative mobile marketplace designed to bridge the "retail gap" for small-scale food producers. By bypassing traditional wholesalers, RFT connects producers directly to consumers via temporary "flash stands." The platform utilizes a cloud-based architecture that combines geo-location services with semantic search powered by Linked Open Data (LOD) and domain-specific ontologies. This approach ensures that consumers can discover local products even when search terms do not exactly match producer descriptions, fostering a more resilient and direct local food economy.

Introduction

Small-scale food producers, such as artisanal fishermen and independent farmers, often struggle to reach consumers directly due to a lack of marketing infrastructure. Consequently, they are frequently forced to sell to wholesalers at significantly reduced prices. Conversely, consumers find it challenging to locate fresh, local produce in real-time.

Traditional search mechanisms exacerbate this issue by relying on rigid keyword matching. For instance, a search for "sea bass" might fail to return a listing for "branzino," despite them being the same product. Furthermore, these transactions are highly time- and location-sensitive. RealFoodTrade (RFT) addresses these challenges by implementing a semantic matchmaking system that understands synonyms and taxonomic relationships, facilitating real-time discovery of "flash stands"—temporary, GPS-defined points of sale.

System Architecture and Methods

The RFT platform is built on a robust cloud-driven architecture, primarily utilizing Google App Engine for the backend and Android for the mobile client. The system's functionality rests on three core pillars:

  1. Flash Stands: Producers use the mobile application to establish temporary sales points. These are defined by specific product lists, quantities, delivery windows, and GPS coordinates.
  2. Semantic Matchmaking: The "App Engine Ontology Manager" replaces simple string matching with RDF-based taxonomies. By calculating the "semantic closeness" between queries and listings, the system identifies relevant results based on sub-species, synonyms, and related concepts.
  3. Hybrid Ontology Approach: To balance precision with breadth, RFT combines "hand-crafted" expert ontologies with the vast Linked Open Data (LOD) cloud. This allows the system to leverage broad web knowledge (e.g., DBpedia) while maintaining the high accuracy required for specific domains like fisheries.
The RealFoodTrade system architecture, illustrating the integration of mobile clients with Google App Engine and the Linked Open Data (LOD) cloud.

Semantic Search and User Experience

A critical challenge in mobile commerce is minimizing user input friction. RFT addresses this through semantic autocomplete. As a user types, the system performs real-time semantic matching against the ontology, suggesting terms that are semantically closest to the partial input. This guides users toward the "correct" terminology recognized by the system, improving search accuracy.

The system integrates several external datasets to power these features:

  • DBpedia: Provides broad, cross-domain information from the LOD cloud.
  • FAO Fisheries Ontology: Offers a fine-grained classification of species, essential for the Chilean Fish & Seafood Market case study.
  • WordNet: Serves as a lexical source for general semantic relatedness.
The RealFoodTrade Android application demonstrating semantic autocomplete (left), the flash stand map interface (center), and product details (right).

Key Results and Prototype Implementation

The authors developed a functional prototype tailored to the Chilean seafood market. Key achievements include:

  • Real-time Visualization: The system successfully generates Points of Interest (POIs) on Google Maps, representing active market announcements.
  • Taxonomic Integration: The prototype demonstrated the successful mapping of local fish names to scientific and commercial categories using the FAO fisheries ontology.
  • Mobile Viability: The implementation of a secure REST API with OAuth2 authentication proved the feasibility of a cloud-backed solution in rural and maritime environments.

Limitations and Future Work

While the prototype demonstrates technical viability, several areas for future development remain. The authors plan to expand the system to general agriculture, which will necessitate more complex, automated ontology extraction. Future iterations aim to automate knowledge acquisition directly from the LOD cloud and incorporate machine learning to refine similarity rankings based on user behavior. Finally, extensive field experiments are required to quantify the actual socio-economic impact on producers' livelihoods.

Summary updated

BibTeX
@inproceedings{cali2014semantic,
  author = {Andrea Calì and Roberto De Virgilio and Tommaso Di Noia and Luca Menichetti and Roberto Mirizzi and Luca Nardini and Vito Claudio Ostuni and Fabrizio Rebecca and Marco Ungania},
  title = {{Semantic Search in RealFoodTrade}},
  booktitle = {Amw},
  year = {2014},
}
Figure