Youtube_Sentiment_Stock_Prediction is a Python package that we hope will be a robust way to begin building and understanding the effects youtube creators can have on stock prices. There are many examples of influencers on youtube that create content and either move stock prices themselves or at least provide a window from excerpt eyes on the way a product or service may affect a stock price. This python script hopes to investigate those perceived trends and highlight the effects of content on youtube against a targeted stock.
I intended to create a continuous pipeline in AWS (see below pipeline diagram) that would pull in daily youtube metrics and caption data that would then be analyzed for the search criteria (in this case, “Nvidia” and or “Nvidia Stock”). This sentiment (positive/negative) would then be used to create and continually update a model that compares this to the stock of the associated search criteria. Finally, it would also plot and allow access to all of these results on a dashboard hosted as a serverless website on either S3 or Elastic BeanStock. There are still shortfalls in the continuous process, but I hope to complete this continuous loop with future additions.
The major package used to pull the captions themselves was the youtube_transcript_api package. You can find details on it at the below link.
https://github.com/jdepoix/youtube-transcript-api
Use the package manager pip to install Youtube_Sentiment_Stock_Prediction.
pip install ***Use of this python code allows you to see results from a simple cmd line tool.
$ python youtube_sentiment_stock_prediction "APPLE Stock" 5
# APPLE Stock = Desired youtube search or ticker symbol
# 25 = Number of top search results to aggrigate sentiment score from
# OUTPUT EXAMPLE
['youtube_sentiment_stock_prediction.py', 'APPL', '5']
videoID KR0g-1hnQPA
datePub 2020-10-13 18:15:12
searchedDate 2020-11-04 14:45:14.106028
VideoTitle Apple Event — October 13
channelTitle Apple
viewCount 54349586
likeCount 925541
dislikeCount 54425
captionString ["Good Day for Dreaming"\nby Ruelle playing] H...
Name: 0, dtype: object
{'neg': 0.013, 'neu': 0.802, 'pos': 0.185, 'compound': 1.0}The stock_pred_model.py code will take the (current) caption data that the lambda fuction builds that then plot that with the associated Stock data. In this case the data and code has been written for Nvidia Youtube Searches and Stock comparisons (see below image for exmple of plot, and this codes potential usefulness)
Christopher Oblak (linkedin: www.linkedin.com/in/christopheroblak/)
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Please make sure to update tests as appropriate.