Brendan Sudol
June 2017

United Statemojis of Twitter

How does emoji usage on Twitter differ across the United States? Great question! Let’s explore…

Here’s a larger version of this map, created with d3.js. In addition, all of the code for this analysis is on GitHub.


Using Twitter’s streaming API, I collected ~1.8 million tweets (over a week last month) that were geolocated within the USA. Of those, ~410k tweets contained at least one emoji. From there, I split up the tweets based on the state they came from and then tallied up all of the emojis.

A couple initial thoughts: 1) the emoji dictionary is expansive—there were 1,067 different emojis used across all of the tweets. 2) The ”face with tears of joy” emoji (😂) is crazy popular. Of the 821k total emojis (many tweets contain more than one), 16% (134k) were 😂s; this is more than the second (😭), third (❤️), fourth (😍), and fifth (🔥) most popular emojis, combined.

If you simply plotted the most commonly used emoji for each state (which I did at first), it’s a pretty boring and repetitive map with 49 😂s out of 51 (including DC). To overcome the monotony, we can compute the “term frequency–inverse document frequency” (tf–idf for short) for each emoji. This is a numerical technique to determine words (emojis) that are significant to a document (tweets within a state) within a larger corpus (all USA tweets). As explained by Wikipedia, ”the tf-idf value increases proportionally to the number of times a word appears in the document, but is often offset by the frequency of the word in the corpus, which helps to adjust for the fact that some words appear more frequently in general.”

With emojis ranked by tf-idf, the maps (top 1, top 4, also see image above) are a lot more interesting. I love seeing emojis for a state match up with things I associate with that state, like the mountains (🏔) in Alaska and Colorado, volcano (🌋) in Hawaii, shamrock (☘) in Massachusetts, and checkered flag (🏁) in Indiana. I also like seeing the creative use of emojis that resemble letters, like the 〰️ (Wavy Dash) for “W” in Washington or the 🔰 (Japanese Symbol for Beginner) for “V” in Virginia. Some of these don’t make as much sense to me, like the mouse (🐁) in South Dakota or nose (👃) in West Virginia, but that might be more of a reflection on me and my unculturedness 😜.

How do you feel about the emojis that represent your state? I hope you think they’re 👌.


If you’re curious about how I collected and aggregated the data, or want to extend this analysis or do something similar, see below for some code snippets.

import json
import os

from collections import defaultdict
from time import time

from emoji import UNICODE_EMOJI
from IPython.core.display import display, HTML
from tweepy import OAuthHandler, Stream, StreamListener, API as TwApi

from util.tfidf import tfidf
CONSUMER_KEY = os.environ.get('TW_CONSUMER_KEY')

DATA_DIR = 'data'
USA_BBOX = [-175.1, 22.4, -59.8, 72.3]  # via
# a few helper functions

def extract_emojis(txt):
    return [c for c in txt if c in UNICODE_EMOJI]

def sort_values(data):
    return sorted(data.items(), key=lambda x: x[1], reverse=True)

def get_json(fname):
    with open('{}/{}'.format(DATA_DIR, fname)) as f:
        return json.load(f)
def save_to_json(data, fname):
    with open('{}/{}'.format(DATA_DIR, fname), 'w') as f:
        json.dump(data, f, ensure_ascii=False)

step 1: fetch tweets (within USA)

# this uses Twitter's streaming API
# for demo purposes, this only fetches 1k tweets
# (in reality, this ran over several days and collected millions of tweets)

class MyListener(StreamListener):
    def __init__(self):
        self.ct = 0
        self.started = time()

    def on_status(self, data):
        if hasattr(data, 'retweeted_status'):

            with open('{}/tweets-all.json'.format(DATA_DIR), 'a') as f:
        except Exception as e:
            print('error: {}'.format(str(e)))

        self.ct += 1
        if (self.ct % 100 == 0):
            print('🚨 {} tweets... ({} secs elapsed)'.format(
                int((time() - self.started))

        # stop stream after 1k results (for demo)
        if self.ct > 1000:
            return False
    def on_error(self, status):
        print('uh-oh! ({})'.format(status))
api = TwApi(auth)

stream = Stream(auth, MyListener())
🚨 100 tweets... (4 secs elapsed)
🚨 200 tweets... (8 secs elapsed)
🚨 300 tweets... (13 secs elapsed)
🚨 400 tweets... (17 secs elapsed)
🚨 500 tweets... (21 secs elapsed)
🚨 600 tweets... (26 secs elapsed)
🚨 700 tweets... (29 secs elapsed)
🚨 800 tweets... (34 secs elapsed)
🚨 900 tweets... (38 secs elapsed)
🚨 1000 tweets... (42 secs elapsed)

step 2: filter tweets to ones containing emojis

filtered = []

with open('{}/tweets-all.json'.format(DATA_DIR)) as f:
    for i, line in enumerate(f):
        d = json.loads(line)
        emojis = extract_emojis(d['text'])

        if i % 200000 == 0:
            print('done with {}...'.format(i))

        if not len(emojis):

            'id': d['id_str'],
            'time': d['created_at'],
            'user': d['user']['screen_name'],
            'text': d['text'],
            'coordinates': d['coordinates'],
            'place': d['place'],
            'emojis': emojis,
            'emojis_names': [UNICODE_EMOJI[e] for e in emojis],

print('{} tweets with emojis'.format(len(filtered)))
done with 0...
done with 200000...
done with 400000...
done with 600000...
done with 800000...
done with 1000000...
done with 1200000...
done with 1400000...
done with 1600000...
done with 1800000...
413433 tweets with emojis
save_to_json(filtered, 'tweets-w-emojis.json')

step 3: aggregate by emoji and state

data = get_json('tweets-w-emojis.json')

{'coordinates': None,
 'emojis': ['👀', '😏'],
 'emojis_names': [':eyes:', ':smirking_face:'],
 'id': '866461716569350145',
 'place': {'attributes': {},
  'bounding_box': {'coordinates': [[[-85.605166, 30.355644],
     [-85.605166, 35.000771],
     [-80.742567, 35.000771],
     [-80.742567, 30.355644]]],
   'type': 'Polygon'},
  'country': 'United States',
  'country_code': 'US',
  'full_name': 'Georgia, USA',
  'id': '7142eb97ae21e839',
  'name': 'Georgia',
  'place_type': 'admin',
  'url': ''},
 'text': "I'm here for Drake and Vanessa... Dranessa 👀😏",
 'time': 'Mon May 22 01:12:25 +0000 2017',
 'user': 'LongLiveDenzy'}
state_cts, emoji_cts = defaultdict(int), defaultdict(int)
state_emoji_cts = defaultdict(lambda: defaultdict(int))

for d in data:
    place = d['place']

    if not place:

    country, ptype = place['country_code'], place['place_type']
    if country != 'US' or ptype not in ['city', 'admin']:

    state = place['name']
    if ptype == 'city':
        state = STATE_LOOKUP[place['full_name'][-2:].upper()]

    if state not in STATES:

    state_cts[state] += 1
    for e in d['emojis_names']:
        if e not in SKIN_TONES:
            emoji_cts[e] += 1
            state_emoji_cts[state][e] += 1
[('Texas', 55023),
 ('California', 48004),
 ('Florida', 26849),
 ('New York', 20858),
 ('Ohio', 18172),
 ('Georgia', 17905),
 ('Illinois', 13705),
 ('Louisiana', 13586),
 ('North Carolina', 11725),
 ('Pennsylvania', 10869)]
[(':face_with_tears_of_joy:', 133771),
 (':loudly_crying_face:', 47571),
 (':red_heart:', 30301),
 (':smiling_face_with_heart-eyes:', 28715),
 (':fire:', 18188),
 (':female_sign:', 17195),
 (':weary_face:', 15559),
 (':skull:', 13885),
 (':face_with_rolling_eyes:', 13747),
 (':person_shrugging:', 12666)]
# show most popular emojis by state 

results = []

for state, emojis in sorted(state_emoji_cts.items()):
        'state': state,
        'emojis': dict(emojis),
        'top_emojis': sort_values(emojis)[:10],

for r in results[:5]:
    print('{}:\n{}'.format(r['state'], r['top_emojis'][:3]))
[(':face_with_tears_of_joy:', 2882), (':loudly_crying_face:', 590), (':female_sign:', 498)]
[(':face_with_tears_of_joy:', 80), (':red_heart:', 50), (':loudly_crying_face:', 33)]
[(':face_with_tears_of_joy:', 1985), (':loudly_crying_face:', 745), (':red_heart:', 718)]
[(':face_with_tears_of_joy:', 958), (':fire:', 258), (':red_heart:', 221)]
[(':face_with_tears_of_joy:', 13933), (':loudly_crying_face:', 5584), (':red_heart:', 4001)]
save_to_json(results, 'emojis-by-state.json')

step 4: add tf-idf (term frequency–inverse document frequency)

data = get_json('emojis-by-state.json')
counts_list = [d['emojis'] for d in data]

data_tfidf = []

for d in data:    
    counts = d['emojis']
    scores = {word: tfidf(word, counts, counts_list) for word in counts}
    sorted_words = sort_values(scores)

        'state': d['state'],
        'top_counts': d['top_emojis'],
        'top_tfidf': sorted_words[:10],
# preview results

for d in data_tfidf[:5]:

    for i, result in enumerate(d['top_tfidf'][:3]):
        word, score = result
        print("\t{}. {} tf-idf: {}".format(i + 1, word, round(score, 5)))
    1. :double_exclamation_mark: tf-idf: 0.00104
    2. :speaking_head: tf-idf: 0.00094
    3. :cat_face_with_tears_of_joy: tf-idf: 0.00074
    1. :snow-capped_mountain: tf-idf: 0.00971
    2. :mount_fuji: tf-idf: 0.0053
    3. :passenger_ship: tf-idf: 0.00486
    1. :A_button_(blood_type): tf-idf: 0.00311
    2. :deciduous_tree: tf-idf: 0.00198
    3. :cactus: tf-idf: 0.00138
    1. :cloud: tf-idf: 0.01442
    2. :double_exclamation_mark: tf-idf: 0.00209
    3. :water_wave: tf-idf: 0.00207
    1. :heavy_minus_sign: tf-idf: 0.00147
    2. :thermometer: tf-idf: 0.0009
    3. :cat_face_with_tears_of_joy: tf-idf: 0.00068