WHAT WOULD A PRESIDENT SAY?

Give the Presidents your sentence. See how they would rephrase it.

Dataset:

US Presidential Speeches

Challenge:

Given a random sentence as input, the program outputs top 3 similar sentences from different presidents.

ML Problem:

Unsupervised, Text Similarity, Chosen Solution: Semantic Similarity

Model Comparison:

WordEmbedding by gensim with GloVe algorithm VS Sentence Embedding with sentenceBERT

Similarity Metric

Soft Cosine Measure / Cosine Similarity

Results

Sentence Embedding Model using SentenceBERT

• The results were good! The top 3 sentences accurately conveyed the meaning of the input sentence.
  • query = "I believe I our children are our future."
• The top 3 results gave a cosine similarity of 0.889 to 0.956
• The model took over 24 hours to run.
• The long run time of the model prevented further exploration of the model for example with different queries..

Word Embedding Model using GloVe

• The results did not resemble the query in spite of the soft cosine similarity being 0.982 to 1.0.
  • query = 'I believe that the young people in America are our future.'
• The model was much less complex.
• The model did not convey the meaning of the input sentences.

Limitations

• The dataset requires significant cleaning.
• Some cleaning was completed in the time aloted.