lukehayes1995/paraphrase-generator-combined-with-siamese-rnn-for-sentence-similarity-detection

Paraphrase-Generator-Combined-With-Siamese-RNN-for-Sentence-Similarity-Detection

The aim of this assignment was to create paraphrased sentences from a set or original sentences. To do so first I used a pretrained model named T5. This model will return as many potential paraphrases as desired. I chose to take 10 and then to select the most suitable I used a Siames RNN to detect how similar the new sentence was from the original. I then selected the sentence that was most similar because this network analyses similarity by semantic meaning and therefore the sentences say the same thing but with different words. If any sentence was identical it was discarder and there were many difficult sentences where the T5 model gave no new predictions. For these cases I created a model myself that utilised wordNet synsets which enables me to look up similar words for all words that were not stopwords( words with little semnatic meaning). This meant that no matter what I was able to make a paraphrase prediction. For this assignment I achieved a result of 97/100.