Tandem LSTM-SVM approach for sentiment analysis

In this paper we describe our approach to EVALITA 2016 SENTIPOLC task. We participated in all the subtasks with constrained setting: Subjectivity Classification, Polarity Classification and Irony Detection. We developed a tandem architecture where Long Short Term Memory recurrent neural network is used to learn the feature space and to capture temporal dependencies, while the Support Vector Machines is used for classification. SVMs combine the document embedding produced by the LSTM with a wide set of general-purpose features qualifying the lexical and grammatical structure of the text. We achieved the second best accuracy in Subjectivity Classification, the third position in Polarity Classification, the sixth position in Irony Detection.