chihacknight / energy-news-roundup Goto Github PK

A key objective of this project is to split out the news entries by state, and eventually by city.

To do this, lets make use of the locationtagger library: https://pypi.org/project/locationtagger/ and check against the descriptions and link URLs.

If the state isn't named directly, we can create a mapping of publications to states/cities and assign them based on that

With the basic scrape completed in #1 (nice work @SuragNuthulapaty and @HerreraRobert!), the next step is to review the output data and see if there's any changes to make to massage the data. I created a Google Sheet of the scraped digest items here as well as a summary table for counts of each category: https://docs.google.com/spreadsheets/d/1n_ruhDMz0g-b2yfhak1m9fq7Xr_0SY5rqPeamJqPHho/edit#gid=1843617235

From this, it looks like there are a few redundant / inconsistently named categories we're pulling like OIL AND GAS vs OIL & GAS etc: https://docs.google.com/spreadsheets/d/1n_ruhDMz0g-b2yfhak1m9fq7Xr_0SY5rqPeamJqPHho/edit#gid=934319479

We should try to clean up / merge these together. A few ideas for that:

• convert all text to uppercase and strip out all non-letter characters
• create a list of categories to ignore, like YOUR AD HERE and YOUR MESSAGE HERE
• for the ALSO and MEANWHILE articles, maybe associate them with the previously found category?

It seems that energynews.us has changed where they hold the new digests. All the links are in what we considered the "old link" previously. They all are held in https://energynews.us/category/digest/page/, so need to remove the requirement to use the other "new" system from before

some of the processing on the data doesn't work and there is some blurbs where there are still bullet points, this needs to be fixed. Currently, throw an error if we find this

Some entries that locationtagger misses seem pretty obvious to me:

• New Mexico’s state land office brought in more than $2 billion in oil and gas royalty revenue over the last 12 months, a record high.
• Wyoming conservationists urge state and federal regulators to tax and further restrict oil and gas facilities’ methane emissions.

Lets add a string comparison check against a list of states and also tag based on that.

The README should explain where we get our data (what newsletter) and some more introductory info.

Take the links to the digests and pre-process them so that:
<b> -> <strong>
</b> -> </strong>
<i> -> <em>
</i> -> </em>

to make it easier to work through

Make unittests for text parsing, specifically cleaning up publication name.

Turn the logic to parse the publication tag and make it into its own method to make it easier to debug.

This will make it so that we don't accidentally truncate words

As @jpvelez discussed last night, once #6 is done, we can start to analyze these 93,000 articles.

One way to start doing that is with topic modeling. Juan pointed to this tutorial as a potential way to get started: https://towardsdatascience.com/end-to-end-topic-modeling-in-python-latent-dirichlet-allocation-lda-35ce4ed6b3e0

@jpvelez, if you have more resources for @SuragNuthulapaty, please add them here.