Synonym Spectra

Twitter account: @SynonymSpectra

This code powers a Twitter account which posts chains of synonyms, which lead from a source word to its antonym: a spectrum of words.

Distinguishable, discernible, perceptible, faint, indistinct, indistinguishable.

I read some paper years and years ago, like a decade or so, that first told me about these structures. In their context, they were trying to use it to power sentiment analyses -- call one antonym +1 and the other -1, and call every link in the chain some fractional value between them.

Good, ample, wide, inaccurate, faulty, bad.

Usage

$ python post_spectrum.py spectra.config spectra.db

where spectra.config is the configuration file used and spectra.db is the Sqlite3 file housing the synonym chains. More detail on both below.

Useful, good, complete, dead, unprofitable, useless.

Dependencies

Synonym spectra database

Before writing this bot code, I had already created a database of the spectra. The posting code here expects to be fed that data in a Sqlite3 file, containing a table called SPECTRA:

$ sqlite3 spectra.db
SQLite version 3.8.10.2 2015-05-20 18:17:19
Enter ".help" for usage hints.
sqlite> .tables
SPECTRA
sqlite> .schema SPECTRA
CREATE TABLE SPECTRA(Id INT, PostCount INT, Start TEXT, End TEXT, Distance Int, Chain TEXT);
CREATE INDEX ROW_IDX on SPECTRA(Id);
sqlite> select count(*) from SPECTRA;
5577
sqlite>

For reference:

• Id is just a handy numerical identifier, unique to each chain
• PostCount is a field I use to track the number of times each chain has been tweeted, so I can avoid tweeting the same ones over and over
• Start and End encode the antonyms of the chain
• Distance is the number of hops in the chain -- when Distance is 3, that means the overall chain has four words in it (including the antonyms)
• Chain is the actual spectrum itself, with each word separated by a space

This database file doesn't actually appear in this repo. I can send it to you on request, I suppose, if you're interested.

I was very proud of myself, starting with just a massive collection of synonym pairs and antonym pairs, then staying up late implementing Djikstra's algorithm over top of it to find the shortest path from one antonym word to its partner by hopping synonym edges. I got to build a priority queue! Then I told my wife about it and she said why didn't I just use breadth-first search since the edges were all unweighted. And I had no good answer for her. Oh well.

Looking at this now, shortest path was probably the wrong way to go.
Encouraging short chains seems like it also encourages using the most tenuous synonym-links available. Maybe I'll relax that shortness constraint some day and see what unfolds.

Ongoing, continuing, last, past.

Config file

Like a lot of my bots, I set up a text file, spectra.config, to manage the details of posting. It looks like:

CONSUMER_KEY = [app key]
CONSUMER_SECRET = [app secret]
ACCESS_SECRET = [account secret]
ACCESS_KEY = [account key]

Those four fields are parsed and used to create a Tweepy Twitter API object, which is used to post the spectra to the account.

Secret, clandestine, fraudulent, unfair, raw, overt.

Libraries

I'm using Python 2.7 to run this.

This really only depends on the Tweepy library. Here's what the virtual environment I use to run the CRON job that tweets these spectra looks like:

$ pip freeze
argparse==1.2.1
tweepy==2.3.0
wsgiref==0.1.2

Order, management, control, subdue, overpower, rout, uproar, pandemonium, chaos.

Photo credit for the twitter account's profile and header: Mark Deckers.