Big Data & Geospatial Linked Open Data to generalize Context-Aware Recommender Systems
ISBN 978-85-88783-11-9
Authors
1Delgado, T.; 2Capote, J.L.; 3Gonzalez, G.; 4Cruz, R.
1ISPJAE Email: tdelgado@ind.cujae.edu.cu
2GEOCUBA Email: capote@geomix.geocuba.cu
3GEOCUBA Email: guillermo@geomix.geocuba.cu
4GEOCUBA Email: rcruz@geomix.geocuba.cu
Abstract
This paper extends the functionality of a collaborative Context-Aware Recommender System (CARS) based on geospatial semantic services by aggregating Big Data and Geospatial Linked Open Data components. Despite the efficacy for controlled scenarios of the CARS which has been developed, its generalization implies exploring further Big Data issues, as well as harnessing the power of Linked Open Data. In this sense, the first problem addressed was looking for a non-controlled scenarios solution capable of handling increasing volumes and a high variety of data. The problem to enable real-time streaming for machine learning is well-described in Big Data literature. A suitable solution can be found within the Big Data ecosystem with a combination of a Hadoop-storage mechanism, and a NoSQL-based database for the high variety of data. Social media streaming can be merged with the analysis carried out using the collaborative filtering in the Spatial CARS. Since this CARS solution uses Mahout (Apache Hadoop software) as machine-oriented learning software, no changes were proposed in the collaborative filtering software. Mahout is very sensitive to the integration with Hadoop Distributed File Systems. Because these systems run faster on private clouds, they will be included in the general architecture for scalability of this kind of CARS application. In the original Semantic CARS, an internal ontology of POIs (Points of Interest) allowed semantic searching and filtering. Its dissemination in other less controlled and more open scenarios led to the evaluation of the Geospatial Linked Open Data in order to use ontologies of POIs from anywhere. This paper proposes an extended architecture to optimize collaborative Spatial and Semantic Context-Aware Recommender Systems, by including Big Data & Linked Open Data building blocks. In the final section, future work and conclusions are provided.
Keywords
Big Data; Geospatial Linked Open Data; Recommender Systems