Deep Neural Network Approaches for Spanish Sentiment Analysis of Short Texts

Abstract

Sentiment Analysis has b een extensively researched in the last years. While imp ortant theoretical and practical results have been obtained, there is still room for improvement. In particular, when short sentences and low resources languages are considered. Thus, in this work we focus on sentiment analysis for Spanish Twitter messages. We explore the combination of several word representations (Word2Vec, Glove, Fas-text) and Deep Neural Networks models in order to classify short texts. Previous Deep Learning approaches were unable to obtain optimal results for Spanish Twitter sentence classification. Conversely, we show promising results in that direction. Our best setting combines data augmentation, three word embeddings representations, Convolutional Neural Networks and Recurrent Neural Networks. This setup allows us to obtain state-of-the-art results on the TASS/SEPLN Spanish benchmark dataset, in terms of accuracy.

Machine Learning for Feeling Analysis in Twitter Communications: A Case Study in HEYDRU!, Perú

“…A comparative study, using KNN and Naive Bayes, highlighted in Ge (5.30 days) being significantly shorter at Gc (8 days). In addition to this, in a research on the detection of indicators for the success of an emerging company, the analysis of sentiments using text data mining estimates that Ge (1.12 days) was significantly lower than Gc (4 days) [15].…”

“…Currently, Machine Learning is one of the most popular ways to examine emotional behaviors, which generates intelligent algorithms that can learn without relying on rulebased programming. The application of machine learning has been prioritized in various fields, with the business environment as the main environment [14] [15]. In relation to what was said above, different agencies are being adapted in the application of machine learning for their different processes [8].…”

Over a Decade of Social Opinion Mining: A Systematic Review

“…The CNN algorithm -a variant of ANN-is made up of neurons that have learnable weights and biases, where each neuron receives an input, performs a dot product and optionally follows it with non-linearity. In total, 12 studies [286,293,288,232,290,294,90,158,57,295,166,296] made use of this algorithm. Notably, the authors in [158] propose a language-agnostic translation-free method for Twitter sentiment analysis.…”

“…RNNs, a powerful set of ANNs useful for processing and recognising patterns in sequential data such as natural language, were used in 8 studies [285,293,297,231,241,90,292,200]. One study in particular [298], considered a novel approach to aspect-based sentiment analysis of Russian social networks based on RNNs, where the best results were obtained by using a special network modification, the RNTN.…”

“…Furthermore, [293] made use of CNN and RNN deep learning algorithms for tweet sentiment analysis, [231,292] used the RNN and LSTM, whereas [286,290] proposed new models based on CNN and LSTM.…”