pgfeldman / keywordexplorer Goto Github PK

These are the counters in ModelExplorer.build_percent_row()

It looks like I need to set up a way to database posts so that they can be searched, since that is not in the intent of the API. Looks straightforward though

When reading in a file and making summaries, the system will get into a loop where it is making the same summary over and over again. Need to at least bail at that point, after making embeddings and hate speech detection

I just checked that a Google site search can point to a Mastodon instance and get result counts for a search term including date ranges. This can be the core of the term explorer. And it should work for Twitter and Reddit as well. https://developers.google.com/custom-search/v1/reference/rest/v1/cse/list

New app for more freeform interactions with LLMs

These are the values in ModelExplorer.build_param_row()

Stub is in markup/TweetEmbedExplorer.md

Create a DataFrame using the values in dict_list = self.hgpt.parse_sequence(sequence) (line 262 in run_probe_callback), then write it out as a csv using the name of the experiment as the filename

I think the logic in excluding clusters may be backwards in CorporaGenerator.get_where_options()

stub is in markup/ModelExplorer.md

The idea is to prompt the GPT-3 with a sampling of tweets from a cluster and have it return a topic name or possibly a set of topic names. The name)s) get stored in the topic_name field of the table_tweet in the twitter_v2 db

We will need a description text area and a "record" checkbox. All output gets split and sent to the db. We could also add some post processing like in TweetEmbedExplorer for Parts of Speech and sentiment, since the fields are there