This is an example application demonstrating access to the Twitter Streaming API.
This crate uses async/await syntax, so a nightly compiler is necessary. The beta channel does not work at the moment due to some pinned dependencies, but it is expected the ecosystem will stabilize quickly after 1.39.0 is released.
The charting functionality depends on a system library to draw text. On an Ubuntu system, you can
run sudo apt-get install libfontconfig libfontconfig1-dev libfreetype6-dev (tested on Ubuntu
19.10).
To run the application, first crate a .env file setting the following environment variables.
The required keys and secrets must be obtained from Twitter Developer Dashboard for your
application.
TWITTER_CONSUMER_KEY=<key>
TWITTER_CONSUMER_SECRET=<secret>
TWITTER_TOKEN_KEY=<key>
TWITTER_TOKEN_SECRET=<secret>
The server can now be run with cargo +nightly run --release.
The application spawns two top-level tasks. The "client task" makes a connection to the Twitter Streaming API and runs sentiment analysis on the received tweets. At 1 second intervals, this task plots a line series graph of the accumulated data and saves it to a PNG on the local filesystem.
The other top-level task is the "server task". It listens for incoming HTTP connections and serves up either a static HTML page, or the PNG image. The HTML file includes a meta tag to refresh the page every 2 seconds.
The two top-level tasks only share a single piece of state (the PNG image) through the filesystem. After writing to a new file, the client task renames the file into place so the server task never observes a partially written file.
- The current score tally logic tracks the average sentiment of each keyword since execution began. This typically results in spikes and volatility early in the chart with the data settling down to an overall average for each keyword. With the current algorithm it will likely take a long time for an overall change in sentiment to be reflected in the chart. A sliding average should be investigated. If a small window is used then low activity keywords will be sensitive to spikes due to outliers (which have less matches to be averaged against). A window of 15 seconds may be a good starting point.
- It would be nice to move the legend outside of the chart area so that most recent data is not obscured, and to have thicker line weights in the plot. Neither seems to be implemented upstream in the crate I picked.
- The page flickers with each reload, which isn't ideal. It may be possible to load the image via AJAX and then draw the PNG via canvas or a data URL. Or, the raw data could be sent for client-side rendering. The current plotting library appears to support compilation to WASM and could possibly be used there.
- A WebSocket could be used to send data incrementally for client-side rendering.
Licensed under either of these:
- Apache License, Version 2.0, (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)