trending-topics

Pipelines & NLP for identification of trending topics in crypto, by ecosystem, over the last 24-48 hours.

How to reproduce

1. clone repo and open trending-topics.Rproj
2. You'll need a twitter api key twitter-secret.txt and chatGPT api key chatgpt-secret.txt these are gitignored and placed in the trending_topics/ directory where the scheduled Rmarkdown lives (update_topics_pipeline.Rmd).
3. You'll also need snowflake-details.json for the submitSnowflake function in the form:

{ "driver": "YOUR-LOCAL-SNOWFLAKE-DRIVER-HERE",
  "server_url": "YOUR-URL-HERE.snowflakecomputing.com",
  "username": "PUT-USERNAME-HERE",
  "password": "PUT-PASSWORD-HERE",
  "role": "INTERNAL_DEV",
  "warehouse": "DATA_SCIENCE",
  "database": ""
}

4. The reprex-for-pulltweets-ai-summarize.R contains the internal functions & example accounts for the broader pipeline. It sources from trending_topics/source_funcs_and_secrets.R which will load the required libraries, functions, and secrets to run the pipeline.

Pipeline

Full pipeline diagram included as an image + pdf.

1. Pulls target_twitter_accounts and calls pull_account_tweets.
2. Dump dataframe form in raw_tweet_dump, call add_new_tweets() proc to clean & append to processed_tweets.
3. Ingest unused tweets to call chatgpt_id_topic at the day-ecosystem level and get the subjects and summaries.
4. Update used tweets to used_in_summary = TRUE and insert into ai_summary.

Website (in-dev) offers a UI over the summaries that uses term frequency to link back to relevant tweets at the day-ecosystem level.