-1

I've been scanning the internet for ways to generate baseball-based labels for youtube baseball videos using text collected from a YT video's description, title, and top 50 titles, but so far, I have been unable to find a Natural Language Processing (NLP) that can achieve this.

To achieve this, I think I must analyse the sentence's sentiment (positive or negative) and understand its hypothesis. From there, I would need to compare those sentences to a baseball-themed whitelist of labels I would need, like 'match', 'baseball', and 'Yankees - Red Sox'.

We have plenty of videos that already have labels, so this would be an option for training an AI.

Finding sentiment in a sentence is already finished. The question is, how can we generate labels with text that already has sentiment values assigned to it? Is there a program we could run this through that is free?

Ian Steffy
  • 49
  • 3
  • I feel like you are asking a combination of questions as one question. Could you possibly clarify your question for us by dividing it into several questions? If you are convinced of your method then it could be 'how do i do sentiment analysis'? But it could also be 'is sentiment analysis the right method?' 'what other methods can i use'? – Robin van Hoorn Dec 21 '22 at 08:42
  • I edited the question. We have sentiment for videos. We just need a way to extract labels from the text that has a sentiment score. I think that should be it... Would there be any other parts to this i am missing that would prevent me from accomplishing this? – Ian Steffy Dec 21 '22 at 15:45
  • I'm totally confused. What does sentiment analysis have to do with label generation? – Snehal Patel Dec 23 '22 at 13:50
  • I can only produce accurate labels if i have sentiment analysis. It would help decipher what a label should be in sentences like 'I hate football' vs 'I love football'. 'I love football' would get the 'football' tag – Ian Steffy Dec 26 '22 at 15:06

2 Answers2

0

Given a number of videos, one could (as you describe) transcribe all language within the video, e.g. using Whisper. Now one has a number of videos that have a minute-by-minute transcription. Given such a transcription it would make sense to cut each video in pieces (for example using the length of the sentence as cut-off points). Due to its associated transcript, each piece could now be associated with a specific sentiment (as determined by an off-the-shelf sentiment analysis tool). One now has a video collection that is labeled with the sentiment associated with the scene.

There are however several problems I see with this approach: one is that the sentiment will likely be influenced by the favorite team of the commentator. Another one is that not all comments during the game will be associated with the game directly but might still contain positive or negative valence.

Mariusmarten
  • 383
  • 1
  • 16
0

There are a few different approaches you could take to generate labels for baseball videos using the text from the video's description, title, and top 50 titles. Here are a few ideas:

Keyword extraction: One approach would be to use a keyword extraction tool to identify the most important words or phrases in the text. These words or phrases could be used as labels for the video. There are a number of free keyword extraction tools available online, such as RAKE (Rapid Automatic Keyword Extraction) and TextRank.

Named entity recognition: Another approach would be to use a named entity recognition (NER) tool to identify specific people, organizations, and locations mentioned in the text. These named entities could be used as labels for the video. There are a number of free NER tools available, such as Stanford NER and Spacy.

Sentiment analysis: If you've already identified the sentiment of the sentences in the text, you could use this information to select labels that align with the sentiment. For example, if the sentiment is positive, you might choose labels like "exciting" or "thrilling," whereas if the sentiment is negative, you might choose labels like "disappointing" or "frustrating." Machine learning: If you have a large dataset of videos with labeled data, you could train a machine learning model to predict labels for new videos based on the text from the video's description, title, and top 50 titles. There are a number of free machine learning libraries and frameworks available that you could use to build and train your model, such as scikit-learn and TensorFlow.

It's worth noting that generating labels for videos based on their text can be a challenging task, and it may require some experimentation and fine-tuning to achieve good results.

Manindra Sammanathegreat
  • 1
  • 1