Conditional_Text_Generation

The main problem of conditional text generation is that it is mainly based on the content of an input set of examples: this leads to little diversification of the generated text. To overcome this shortcoming, we have fine tuned CTRL using three different datasets. The first model has been used as a baseline for comparison, while the other two have been used to obtain more formal and informal text. The BART model has been employed for text classification to gauge formality.

Tripadvisor_Sentimental_Analysis

Sentiment analysis task, analyzing user’s textual reviews, to understand if a comment includes a positive or negative mood. In practice, a robust classification model that is able to predict the sentiment contained in a text.