Abstract ▼ AI-Helper

This work presents the idea, analysis and findings of the graduation project for Masters. This thesis's main task is Cross-Lingual Sentiment Analysis. We introduce a novel approach to address Cross-Lingual Sentiment analysis of languages with limited resources using advanced models and techniques wh...

This work presents the idea, analysis and findings of the graduation project for Masters. This thesis's main task is Cross-Lingual Sentiment Analysis. We introduce a novel approach to address Cross-Lingual Sentiment analysis of languages with limited resources using advanced models and techniques while profiting from advances made in other languages and tasks.
The main idea can be simplified as to using BERT models and machine translation for sentiment analysis of different languages. We choose to work on English, French, Arabic and the Moroccan dialect. We apply BERT to the original data of each language and then translate: Arabic and French to English, and the Moroccan dialect to Arabic. The translated data goes through the Sentiment Analysis models and finally we compare both results.
By doing so, in one hand, we make use of advances made in machine translation from/to English. In the other hand, the translated data profits from the processing in the target languages. Our experiments show promising results over baseline models, we achieve 1% increased accuracy on the French language, 2% over the Arabic data and a new state-of-the-art results for the Moroccan dialect.

주제어