GRAISearch (2014-2018)

The primary scientific goal of the GRAISearch project (Use of Graphics Rendering and Artificial Intelligence for Improved Mobile Search Capabilities) was to develop and integrate revolutionary graphics rendering and artificial intelligence (AI) methods into an existing social media search engine platform, creating groundbreaking mobile search capabilities with significant online commercial potential. The project leveraged pioneering technologies developed at two academic institutions -- INSA de Lyon (LIRIS UMR CNRS 5205) and Trinity College Dublin (TCD) -- in collaboration with the industry partner Tapastreet. The aim was to commercialize innovative functionalities that enhance social media search capabilities, positioning Tapastreet as a leader in this domain.

GRAISearch was at the forefront of modern data processing and analysis techniques. Social media traffic and personal mobile usage continue to grow exponentially with advancements in portable devices and mobile transmission technologies such as WiFi, 4G, and 5G. Simultaneously, the volume of publicly shared media, particularly visual content, is increasing, as users transform into independent amateur broadcasters, sharing real-time, geo-located content. This wealth of user-generated media provides invaluable insights into global events and trends, offering potential benefits in diverse fields such as smart city planning, tourism, emergency rescue operations, and other scenarios requiring up-to-date, localized information.

The project focused on developing tools for smart mobile browsing, enabling data harvested from official APIs of social media streams to be merged, analyzed, visualized, and presented in a concise, user-friendly format. This data—geo-located, time-stamped, and inherently noisy, multi-modal, and heterogeneous—necessitated advanced visualization tools to extract actionable insights. Work Packages 1 and 2 investigated technologies for video summarization of social media-reported events, combining geo-location information with 3D mapping. Prototypes were successfully developed and demonstrated, accompanied by impactful publications.

One of the critical research challenges addressed by GRAISearch was event detection in social media streams, a field with high potential for applications in areas underserved by mainstream media. Work Packages 3 and 4 focused on real-time detection of local trends in geo-located social media streams. In collaboration with Tapastreet, INSA researchers developed "Gazouille," a system for discovering local events from such streams. Gazouille features three core modules:

1. Data acquisition from social networks across several urban areas.
2. Event detection using time-series analysis to identify patterns.
3. A web interface that presents real-time, city-specific events accompanied by galleries of related social media content.

The system demonstrated superior performance compared to the state-of-the-art method for geo-located event detection.

Work Package 5 explored recommender systems, optimizing their application to the Tapastreet platform. Work Package 6 addressed the integration of the project's research outcomes into the commercial platform, ensuring the seamless transition of innovative features into a market-ready product.

Through its innovations, GRAISearch laid the groundwork for transformative capabilities in social media analysis and mobile search. The project’s tools and methodologies hold immense promise for numerous applications, driving advancements in how localized, real-time social media information is processed, visualized, and utilized for commercial and societal benefit.