CEA LIST's participation at MediaEval 2013 placing task
Résumé
At MediaEval Placing Task 2013 [2], we focused on improving our last year's participation in four directions: (1) exploit a larger geotagged dataset in order to improve the quality of a standard geolocation language model, (2) model machine tags, (3) estimate the geographicity of tags associated to geolocated photos and (4) exploit user cues in order complement language models whenever these last are likely to fail. Obtained results show that all modifications proposed this year have a positive effect. A "standard" based only on the training data (cues (1)+(2)) has the poorest performance, with P@1km = 0.268, while P@1km = 0.434 when all cues are used.