Zero-shot cross-lingual transfer learning for sentiment analysis on Swedish chat conversations
Abstract: As the field of machine learning grows, so do the publicly available datasets. However, in the field of natural language processing, datasets within specific languages and tasks can be scarce. This thesis shows the possibility of using zero-shot cross-lingual transfer learning to train a variety of machine-learning models on a strictly English dataset and then applying the models on a Swedish dataset. The task at hand is a binary sentiment analysis where the model learns to classify a text as either a personal attack or a non-personal attack. Four machine learning models are trained for this thesis, a feedforward neural network, a long short-term memory neural network, a gated re- current neural network, and an XLM-RoBERTa transformer. All models are trained or fine-tuned on an Englishdataset and tested on a Swedish dataset with an overall positive good result. This thesis shows that it is possible to use zero-shot cross-lingual transfer learning for sentiment analysis when using aligned word embeddings or a pretrained XLM- RoBERTa transformer.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)