Project Abstract

YouTube is a preferred platform for the under 40s, and low-income individuals. In other words, those who need it most. Unfortunately, there is little accountability, particularly compared to other platforms. Fortunately, new innovations in NLP, combined with improved captioning accuracy, have improved the ability to use transcripts to evaluate videos.This work is building upon the established work in financial natural language understanding, sentiment analysis, and more. It uses state of the art attention models and information unique to YouTube to find a relationship with the quality of a video’s financial advice.First of all, lots of video transcripts and basic information such as upload date alongside their advised portfolios’ performances are collated. Next, models using this information are created and evaluated, iterating on the last to find as strong a relationship as possible.A relationship between the videos’ information is very clear to see, and it is evident from this investigation that further work in the field with a larger dataset and computationally expensive models would not be wasted.Essentially it is found that using only state of the art attention models with video transcripts, and simple easy-to-work-with video metadata, the quality of financial advice can be discerned.

Keywords: Natural Language Processing, Machine Learning, Text Insights

 Conference Details

Session: Poster Session B at Poster Stand 115

Location: Sir Stanley Clarke Auditorium at Wednesday 8th 09:00 – 12:30

Markers: Xianghua Xie, Pardeep Kumar

Course: BSc Computer Science, 3rd Year

Future Plans: I’m looking for work