All Hokie, All the Time. Period. Presented by First Bank & Trust Company

D
daveywoods OP
Mar 22, 2024 at 09:02 PM ET
AI on TSL. No kidding.
I'm working on a TSL AI, just for fun. I ran this by Will, he said OK. I'm trying to build a sentiment analysis tool to evaluate TSL message board posts for Positive, Negative, or Neutral. If it works, I'll give it to TSL (or I'll run it for TSL). It will be interesting to see how message board sentiment trends through the season. This would be run on the bulk of the board, not labeling individual messages. Something like "this week the message board is 63% more positive than last week". This is achieved by grading individual messages, but only aggregate data would be reported. But assessment of positive or negative is obviously a subjective evaluation, which is why I'm here asking for help. I'm looking for people to "grade" the AI training data. That means looking at a few hundred rows of data and marking them as positive, negative, or neutral. The way AI works, if we give it human input on some data, it will be able to go ahead on an automated basis. If I train all the data myself, the sentiment analysis will be skewed by my bias. And anyway I could use help looking at all this data. If you're interested, email me and I'll send you 300 rows of messages to look at. You can do as many or as few as you like. This is simply looking at the data I have pulled off the message board for subject line and message, and grading that message as positive, negative, or neutral. Please help if this absurd nerdery interests you. my email address is my username at gmail dot com. Also I should mention -- Pretty good chance this never works, but let's try. It's neat.

14 Replies

H
Hokie1992
TSL message boards skewed? Impossible
C
csHokie04
What are you writing this text classifier in? Using pytorch or something…
else?
D
daveywoods OP
Scikit-Learn to split the data in Jupyter Notebook
CPVT3 CPVT3
I'll bet it's often negative. Only because negative comments are >>> +
People usually comment on negatives more than positives IMO Like online reviews or whatnot
949hokie 949hokie
Interested
K
KCHokie2
"Hal, open the portal bay doors."
WestyHokie WestyHokie
Up till now, this kind of effort has always failed
Maybe AI will be different. In the past, algorithims struggled with things like negative posts about positive things, or vice versa, and sarcasm. I suspect the latter will be a challenge on TSL. This issue used to drive us crazy at my former company on automated analysis of media stories and employee engagement survey comments.
J
jesuisvtguy
I grade this post as Negative ;)
joking - sounds like a cool project and could be of really nice use to help eventually moderate as well. ** Edited by jesuisvtguy at 3/23/2024, 12:53:15 PM
dolph dolph
I can rank all of the posts from Phew, Bama and Johnny Hawk. Done.
The_Phew The_Phew
It's hard to characterize sarcasm/snark as 'positive' or 'negative'.
Usually there is a generational divide where posters 'of a certain age' don't understand internet humor and think it's some kind of attack on their worldview. It goes both ways as well; anyone under 50 is going to be puzzled at 'Boomer Facebook' humor like Cathy comic strips or whatever. ** Edited by The_Phew at 3/23/2024, 12:11:04 PM
TomTurkey TomTurkey
I'll take Pogo or L'il Abner, thank you
Vienna_Hokie Vienna_Hokie
That's why AI models are trained, they are taught who the snarks are ;)
The_Phew The_Phew
Train the model with Reddit, and 'no snark' would be considered negative.
It's a well-known fact that if you want help on Reddit, don't just politely ask your question, or your query will be ignored. Instead, make a smart-ass post explaining why a wrong answer is 'the way', and posters will line up to correct you with the actual correct answer.
P
Pride_and_Joy
Happy to help. I’m very good at gauging
positivity and negativity.