Skip to content

Latest commit

History

18 Commits

Folders and files

NameName
Last commit message
Last commit date

Repository files navigation

Text analysis on Fantasy Football video transcripts

No one has time now adays to listen through the podcasts or youtube videos for the dozens of content creators out there, so this extracts the transcripts since the last gameweek and summarises their findings. Initially this will just be youtube transcripts as there doesn't appear to be a way to automate extracting transcripts through either Apple or Spotify podcasts.

The YouTube API is used to extract both the channel_id's and video_id's from each of a given set of content creators, and then the YouTubeTranscriptApi will attempt to extract the transcripts for each video after a specified time (last gameweek). Summarisation will then either be done using Facebook's BART model and tokeniser (bart-large-cnn) or by plugging it into Claude API.

The maximum token length using BART is 1024, so chunking the transcripts will be required even after preprocessing. Whereas the maximum token length through Claude will be 100,000 (including prompt) so chunking shouldn't be necessary. Preprocessing includes removing automated tags such as [music] or [applause], along with filler words such as "basically" and "I mean". More can be done to reduce the token count by creating abbreviations for common terms, for example replacing "fantasy premier league" with "fpl", but a balance needs to be made between stripping out tokens but maintaining context.

There are obvisouly going to be some issues with the automated speach-to-text transcripts that are generated by youtube, especially with regards to player names which are unlikley to be accurately transcribed, which is another area where the Claude API is likely to outperform BART.

Below shows a an example week for a set of youtubers, displaying the word count stats along with the reduction seen after preprocessing the transcripts.

channel_nameword_count_meanword_count_maxword_count_sumnum_videosword_count_process_meanword_count_process_maxword_count_process_sumpercent_reduction
fantasyfootballfixYT388172241940553714.66702185734
alwayscheating15940.51838763762414316.8164835726710
elitefpl1031712072309513977311432293195
AboveAverageFPL144011607228802213386.515097267737
FFScout_9945.5100011989129368.59446187375
FPLBlackBox15252278973050421413125788282627
FPLFocal22092428441822194.5241443890
TheArmbandFPL13227132942645421203312090240669
fplblackbox15252278973050421413125788282627
FMLFPL185141851418514116396163961639611
fplbanger10304103041030419557955795577

About

A tool to extract and summarise transcripts from Fantasy Premier League content creators

Topics

Resources

Stars

1 star

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages