I am trying to build a model that uses transcribed audio and on-screen text to classify a video ad as political or non-political, as well as extracts the name of the candidate and sponsor. How can I go about doing this? One possible solution I thought of was using a dictionary of commonly used (or necessitated phrases) like “I approve this message” and “This ad has been sponsored by …, Vote for”
and using Spacy NER to extract the candidate’s name from these phrases.
If anyone has any solutions/suggestions please let me know.

Trying to classify the ad as political or non-political and extract the candidate’s name, constituency, and party