Smart multi-modal crowdsensing-based system as a service oriented to the prediction of social problems
Project grant PID2020-112827GB-I00 funded by:ALLEGRO (Adaptive muLti-domain sociaL-media sEnsinG fRamewOrk) is a general-purpose multi-modal system for the development of real-time crowdsensing applications that can accurately reconstruct the state of the surrounding world as interpreted by the collective intelligence of social-media users. It is primarily focused on the data layer of crowdsensing systems, where raw data collected from social media along with other open-data sources (e.g. providers of meteorological, demographic, macroeconomic or geographical data) can be processed to extract relevant knowledge.
ALLEGRO consists of two modules (i.e. Data Analysis and Data Fusion), which make use of a multi-modal data repository and a knowledge base. In both modules, high performance computing (HPC) contributes to accomplish highly efficient real-time data processing. On the one hand, Data Analysis is comprised of a dedicated component for each type of data to be analysed in social-media posts, namely the Text Analysis module (DIAPASON, unifieD hybrId ApProach to microtext Analysis in Social-media crOwdseNsing), the Audio Analysis module (SOUND, Social-media sOUnd aNalysis moDule), and the Image Analysis module (ADAGIO, sociAl meDia imAGe analysIs mOdule). In this regard, microtext analysis is viewed as the initial process that provides the context of the problem described in each message, so that an event-based knowledge schema can be returned. In case that messages go with embedded audio and/or image content, this schema can be supplemented with context data from audio analysis and/or image analysis, which are concurrently executed, both to verify or rebut event-related information detected in the text or to complete missing information in the knowledge schema. The video content can also be taken into consideration for this context-augmentation process by analyzing its audio and image components separately in the corresponding modules. On the other hand, augmented knowledge schemas produced in this module are combined in the Data Fusion module (LAUD, sociaL mediA fUsion moDule), where the quality of aggregated data is enhanced by rejecting irrelevant information, minimizing redundancy, resolving inconsistencies, and completing missing information.
ALLEGRO takes an integral approach to smart cities, whose ultimate goal is to improve well-being and quality of life. To this end, social-media posts are analysed to reveal insights about problems that can disrupt social coexistence. Therefore, we focus not only on the quality of the urban infrastructure and services provided to citizens (e.g. environment, healthcare, lighting, traffic, and transportation, among others) but also on the understanding of the sociological dimension of the city, which is reflected through people's concerns (e.g. cultural and ethnic conflicts, economy and employment, poverty, and violence, just to name a few). As a result, in this proposal we use the term "smart city" to refer to both aspects, which can be viewed as two sides of the same coin. In this manner, our proposal offers a more complete alternative to the existing applications in smart cities, which are more focused on the first aspect.
Click on one of the following ALLEGRO modules:
Cebrian, J. M., Imbernón, B., Soto, J., Cecilia, J. M. (2021). Evaluation of Clustering Algorithms on HPC Platforms. Mathematics, 9(17), 2156. doi:10.3390/math9172156
Periñán-Pascual, C. Measuring associational thinking through word embeddings. Artificial Intelligence Review (2021). doi.org/10.1007/s10462-021-10056-6