Evaluation and Analysis Data from Twitter Data By Using Hybrid CNN & LTSM

The central theme for this thesis is the design of an aspect-based sentiment analysis model for the classification of online Italian automotive forums' comments. The work starts with designing a strategy for collecting information about target forums to make it possible to develop a machine learning-based sentiment classification model. The study involved applying the CNN and LTSM model, a state-of-the-art solution based on a parametric model that will improve the performance of a baseline algorithm, especially in case of very noisy data like the ones where this tool is supposed to be to work on. This work has been designed as a two-stage CNN and LTSM classifier in all its parts. It was compared with a one-step classifier to detect the pertinence about some topics, and eventually, the sentiment achieved an accuracy of 96.78% for all comments. The current problem passed from a typical three degrees' polarity sentiment analysis to a four labels text classification, where it will be introduced an additional category for determining whether the text is pertinent to a particular topic or not. Presenting this information, the models must be enhanced, and a cascade classification solution will be proposed. The final model is then utilized for a real-world use case. New data have been classified concerning some selected topics, finally presented exploiting a data visualization but still not satisfactory, thus making sentiment analysis an ongoing and open research subject.