Ernest Lai

Computational semantic analysis of presidential speeches and TED talks

Started Jan. 24, 2024

Abstract or project description

TED Talks are globally recognized as an influential way to spread important and diverse ideas. The organization TED has created a repository of all TED Talks on its website and annotated them with topic tags, the essential themes present in the Talk. In this project, I aim to develop an effective and accurate model to predict the number of views on a TED Talk using its associated topic tags. This is important because it can help TED to better select TED Talk topics and speakers to reach a larger audience and impact more people. I developed three models based on machine-learning algorithms to predict the number of views, the first based on Universal Sentence Encoder (USE) vectors, the second utilizing the topic tags associated with each TED Talk, and the third using computer-generated topic tags based on the USE vectors. I found a significant improvement in accuracy from Model 1 to Model 2, and, unexpectedly, another significant increase in accuracy from Model 2 to Model 3. This demonstrates that computer-generated topic tags are more effective than manually-assigned tags at predicting the number of views on a TED Talk.