SINAI at Affect Task in MediaEval 2010
José M. Perea‐Ortega, Arturo Montejo‐Ráez, María Teresa Martín Valdivia, Luís Alfonso Ureña López · 2010
In this paper, a basic approach to establish the boredom score of several videos using their speech transcriptions is presented. We carried out an analysis of the development data provided trying to establish a correlation between the terms of the transcriptions and their boredom scores predicted for each video. For the development data, we analyzed dierent weighting schemes such as TF, TF IDF, binary or correlation weights, due to the small size of the texts involved. We also studied how the use of stopper and stemmer inuences the system. Finally, we concluded that apply stopper and stemmer is interesting in this task and TF and correlation weights obtained the best results. Therefore, two experiments were run, using TF and correlation weights as weighting schemes, always preprocessing the transcriptions of the videos. The experiment that used correlation weights obtained the best result for the test videos provided using the Kendall tau distance as evaluation metric.