Scraping data from Twitter’s Streaming API

Authenticating

Before we can start collecting Twitter data, we need to create an OAuth token that will allow us to authenticate our connection and access our personal data.

After the new API changes, getting a new token requires submitting an application for a developer account, which may take a few days. For teaching purposes only, I will temporarily share one of my tokens with each of you, so that we can use the API without having to do the authentication.

However, if in the future you want to get your own token, check the instructions at the end of this file.

library(ROAuth)
load("~/my_oauth")

To check that it worked, try running the line below:

library(tweetscores)

## Loading required package: R2WinBUGS

## Loading required package: coda

## Loading required package: boot

## ##
## ## tweetscores: tools for the analysis of Twitter data

## ## Pablo Barbera (LSE)

## ## www.tweetscores.com
## ##

getUsers(screen_names="LSEnews", oauth = my_oauth)[[1]]$screen_name

## [1] "LSEnews"

If this displays LSEnews then we’re good to go!

Some of the functions below will work with more than one token. If you want to save multiple tokens, see the instructions at the end of the file.

Collecting data from Twitter’s Streaming API

Collecting tweets filtering by keyword:

library(streamR)

## Loading required package: RCurl

## Loading required package: bitops

## Loading required package: rjson

## Warning: package 'rjson' was built under R version 3.4.4

## Loading required package: ndjson

## Warning: package 'ndjson' was built under R version 3.4.4

filterStream(file.name="~/data/trump-streaming-tweets.json", track="trump", 
    timeout=20, oauth=my_oauth)

## Capturing tweets...

## Connection to Twitter stream was closed after 20 seconds with up to 1348 tweets downloaded.

Note the options: - file.name indicates the file in your disk where the tweets will be downloaded
- track is the keyword(s) mentioned in the tweets we want to capture. - timeout is the number of seconds that the connection will remain open
- oauth is the OAuth token we are using

Once it has finished, we can open it in R as a data frame with the parseTweets function

tweets <- parseTweets("~/data/trump-streaming-tweets.json")

## 790 tweets have been parsed.

tweets[1,]

##                                                                                                                                            text
## 1 RT @1Romans58: And just like that they figured out how to stop the caravan...\n\nOrganizer of Migrant Caravan Detained After President Trump…
##   retweet_count favorite_count favorited truncated              id_str
## 1           279            353     FALSE     FALSE 1052602890311884803
##   in_reply_to_screen_name
## 1                    <NA>
##                                                               source
## 1 <a href="http://twitter.com" rel="nofollow">Twitter Web Client</a>
##   retweeted                     created_at in_reply_to_status_id_str
## 1     FALSE Wed Oct 17 16:51:00 +0000 2018                      <NA>
##   in_reply_to_user_id_str lang listed_count verified location
## 1                    <NA>   en            9    FALSE     <NA>
##          user_id_str description geo_enabled
## 1 703631194601222145      \u274c       FALSE
##                  user_created_at statuses_count followers_count
## 1 Sat Feb 27 17:22:04 +0000 2016          33960            1838
##   favourites_count protected user_url       name time_zone user_lang
## 1            51074     FALSE     <NA> DEE RIVERA        NA        en
##   utc_offset friends_count screen_name country_code country place_type
## 1         NA          1821    nicher66         <NA>    <NA>         NA
##   full_name place_name place_id place_lat place_lon lat lon expanded_url
## 1      <NA>       <NA>     <NA>       NaN       NaN  NA  NA         <NA>
##    url
## 1 <NA>

If we want, we could also export it to a csv file to be opened later with Excel

write.csv(tweets, file="~/data/trump-streaming-tweets.csv", row.names=FALSE)

And this is how we would capture tweets mentioning multiple keywords:

filterStream(file.name="~/data/politics-tweets.json", 
    track=c("graham", "sessions", "trump", "clinton"),
    tweets=20, oauth=my_oauth)

Note that here I choose a different option, tweets, which indicates how many tweets (approximately) the function should capture before we close the connection to the Twitter API.

We can also filter tweets in a specific language:

filterStream(file.name="~/data/spanish-tweets.json", 
    track="trump", language='es',
    timeout=20, oauth=my_oauth)

tweets <- parseTweets("~/data/spanish-tweets.json")
sample(tweets$text, 10)

And we can filter tweets by / retweeting / mentioning a specific user:

filterStream(file.name="~/data/trump-follow-tweets.json", 
    follow=25073877, timeout=10, oauth=my_oauth)

tweets <- parseTweets("~/data/trump-follow-tweets.json")
sample(tweets$text, 10)

We now turn to tweets collect filtering by location instead. To be able to apply this type of filter, we need to set a geographical box and collect only the tweets that are coming from that area.

For example, imagine we want to collect tweets from the United States. The way to do it is to find two pairs of coordinates (longitude and latitude) that indicate the southwest corner AND the northeast corner. Note the reverse order: it’s not (lat, long), but (long, lat).

In the case of the US, it would be approx. (-125,25) and (-66,50). How to find these coordinates? You can use Google Maps, and right-click on the desired location. (Just note that long and lat are reversed here!)

filterStream(file.name="~/data/tweets_geo.json", locations=c(-125, 25, -66, 50), 
    timeout=30, oauth=my_oauth)

## Capturing tweets...

## Connection to Twitter stream was closed after 30 seconds with up to 584 tweets downloaded.

We can do as before and open the tweets in R

tweets <- parseTweets("~/data/tweets_geo.json")

## 560 tweets have been parsed.

And use the maps library to see where most tweets are coming from. Note that there are two types of geographic information on tweets: lat/lon (from geolocated tweets) and place_lat and place_lon (from tweets with place information). We will work with whatever is available.

library(maps)
tweets$lat <- ifelse(is.na(tweets$lat), tweets$place_lat, tweets$lat)
tweets$lon <- ifelse(is.na(tweets$lon), tweets$place_lon, tweets$lon)
tweets <- tweets[!is.na(tweets$lat),]
states <- map.where("state", tweets$lon, tweets$lat)
head(sort(table(states), decreasing=TRUE))

## states
##      texas california    florida       ohio    georgia   illinois 
##         66         63         41         25         20         20

We can also prepare a map of the exact locations of the tweets.

library(ggplot2)

## Warning: package 'ggplot2' was built under R version 3.4.4

## First create a data frame with the map data 
map.data <- map_data("state")

# And we use ggplot2 to draw the map:
# 1) map base
ggplot(map.data) + geom_map(aes(map_id = region), map = map.data, fill = "grey90", 
    color = "grey50", size = 0.25) + expand_limits(x = map.data$long, y = map.data$lat) + 
    # 2) limits for x and y axis
    scale_x_continuous(limits=c(-125,-66)) + scale_y_continuous(limits=c(25,50)) +
    # 3) adding the dot for each tweet
    geom_point(data = tweets, 
    aes(x = lon, y = lat), size = 1, alpha = 1/5, color = "darkblue") +
    # 4) removing unnecessary graph elements
    theme(axis.line = element_blank(), 
        axis.text = element_blank(), 
        axis.ticks = element_blank(), 
        axis.title = element_blank(), 
        panel.background = element_blank(), 
        panel.border = element_blank(), 
        panel.grid.major = element_blank(), 
        panel.grid.minor = element_blank(), 
        plot.background = element_blank())

## Warning: Removed 3 rows containing missing values (geom_point).

And here’s how to extract the edges of a network of retweets (at least one possible way of doing it):

tweets <- parseTweets("~/data/trump-streaming-tweets.json")

## 790 tweets have been parsed.

# subset only RTs
rts <- tweets[grep("RT @", tweets$text),]
library(stringr)
edges <- data.frame(
  node1 = rts$screen_name,
  node2 = str_extract(rts$text, 'RT @[a-zA-Z0-9_]+'),
  text = rts$text,
  stringsAsFactors=F
)
edges$node2 <- str_replace(edges$node2, 'RT @', '')

# plotting largest connected component
library(igraph)

## 
## Attaching package: 'igraph'

## The following objects are masked from 'package:stats':
## 
##     decompose, spectrum

## The following object is masked from 'package:base':
## 
##     union

g <- graph_from_data_frame(d=edges, directed=TRUE)
comp <- decompose(g, min.vertices=20)
plot(comp[[1]])

Finally, it’s also possible to collect a random sample of tweets. That’s what the “sampleStream” function does:

sampleStream(file.name="~/data/tweets_random.json", timeout=30, oauth=my_oauth)

## Capturing tweets...

## Connection to Twitter stream was closed after 30 seconds with up to 2366 tweets downloaded.

Here I’m collecting 30 seconds of tweets. And once again, to open the tweets in R…

tweets <- parseTweets("~/data/tweets_random.json")

## Warning in stream_in_int(path.expand(path)): Parsing error on line 1445

## 1234 tweets have been parsed.

What is the most retweeted tweet?

tweets[which.max(tweets$retweet_count),]

##                                              text retweet_count
## 270 RT @BTS_twt: 마음만은 https://t.co/mIYCOUD4KA        361506
##     favorite_count favorited truncated              id_str
## 270         878411     FALSE     FALSE 1052603140036587520
##     in_reply_to_screen_name
## 270                    <NA>
##                                                                                   source
## 270 <a href="http://twitter.com/download/android" rel="nofollow">Twitter for Android</a>
##     retweeted                     created_at in_reply_to_status_id_str
## 270     FALSE Wed Oct 17 16:51:59 +0000 2018                      <NA>
##     in_reply_to_user_id_str lang listed_count verified location
## 270                    <NA>   ko            0    FALSE     <NA>
##             user_id_str description geo_enabled
## 270 1031644815711248385        <NA>       FALSE
##                    user_created_at statuses_count followers_count
## 270 Mon Aug 20 20:51:05 +0000 2018            578               6
##     favourites_count protected user_url name time_zone user_lang
## 270             1696     FALSE     <NA>  Bbb        NA        ar
##     utc_offset friends_count screen_name country_code country place_type
## 270         NA            16 Bbb53507778         <NA>    <NA>         NA
##     full_name place_name place_id place_lat place_lon lat lon expanded_url
## 270      <NA>       <NA>     <NA>       NaN       NaN  NA  NA         <NA>
##      url
## 270 <NA>

What are the most popular hashtags at the moment? We’ll use regular expressions to extract hashtags.

library(stringr)
ht <- str_extract_all(tweets$text, '#[A-Za-z0-9_]+')
ht <- unlist(ht)
head(sort(table(ht), decreasing = TRUE))

## ht
##         #PCAs          #BTS #HipHopAwards     #TheGroup           #ad 
##             6             4             3             3             2 
##           #Be 
##             2

And who are the most frequently mentioned users?

handles <- str_extract_all(tweets$text, '@[0-9_A-Za-z]+')
handles_vector <- unlist(handles)
head(sort(table(handles_vector), decreasing = TRUE), n=10)

## handles_vector
##         @BTS_twt         @YouTube   @jairbolsonaro @realDonaldTrump 
##                8                8                5                4 
##   @ABrunchMember      @boujeesIut         @AbemaTV     @AmazingPhil 
##                3                3                2                2 
##     @BetoORourke          @BP6311 
##                2                2

How many tweets mention Justin Bieber?

length(grep("bieber", tweets$text, ignore.case=TRUE))

## [1] 1

These are toy examples, but for large files with tweets in JSON format, there might be faster ways to parse the data. For example, the ndjson package offers a robust and fast way to parse JSON data:

library(ndjson)
json <- stream_in("~/data/tweets_geo.json")
json

## Source: local data table [560 x 1,379]
## 
## # A tibble: 560 x 1,379
##    contributors coordinates created_at display_text_ra… display_text_ra…
##           <int>       <int> <chr>                 <dbl>            <dbl>
##  1           NA          NA Wed Oct 1…               21               27
##  2           NA          NA Wed Oct 1…               NA               NA
##  3           NA          NA Wed Oct 1…               14               24
##  4           NA          NA Wed Oct 1…               NA               NA
##  5           NA          NA Wed Oct 1…               13               66
##  6           NA          NA Wed Oct 1…               NA               NA
##  7           NA          NA Wed Oct 1…               NA               NA
##  8           NA          NA Wed Oct 1…              117              140
##  9           NA          NA Wed Oct 1…               NA               NA
## 10           NA          NA Wed Oct 1…               NA               NA
## # ... with 550 more rows, and 1,374 more variables:
## #   entities.hashtags <int>, entities.symbols <int>, entities.urls <int>,
## #   entities.user_mentions.0.id <dbl>,
## #   entities.user_mentions.0.id_str <chr>,
## #   entities.user_mentions.0.indices.0 <dbl>,
## #   entities.user_mentions.0.indices.1 <dbl>,
## #   entities.user_mentions.0.name <chr>,
## #   entities.user_mentions.0.screen_name <chr>,
## #   entities.user_mentions.1.id <dbl>,
## #   entities.user_mentions.1.id_str <chr>,
## #   entities.user_mentions.1.indices.0 <dbl>,
## #   entities.user_mentions.1.indices.1 <dbl>,
## #   entities.user_mentions.1.name <chr>,
## #   entities.user_mentions.1.screen_name <chr>, favorite_count <dbl>,
## #   favorited <lgl>, filter_level <chr>, geo <int>, id <dbl>,
## #   id_str <chr>, in_reply_to_screen_name <chr>,
## #   in_reply_to_status_id <dbl>, in_reply_to_status_id_str <chr>,
## #   in_reply_to_user_id <dbl>, in_reply_to_user_id_str <chr>,
## #   is_quote_status <lgl>, lang <chr>, place.attributes <int>,
## #   place.bounding_box.coordinates.0.0.0 <dbl>,
## #   place.bounding_box.coordinates.0.0.1 <dbl>,
## #   place.bounding_box.coordinates.0.1.0 <dbl>,
## #   place.bounding_box.coordinates.0.1.1 <dbl>,
## #   place.bounding_box.coordinates.0.2.0 <dbl>,
## #   place.bounding_box.coordinates.0.2.1 <dbl>,
## #   place.bounding_box.coordinates.0.3.0 <dbl>,
## #   place.bounding_box.coordinates.0.3.1 <dbl>,
## #   place.bounding_box.type <chr>, place.country <chr>,
## #   place.country_code <chr>, place.full_name <chr>, place.id <chr>,
## #   place.name <chr>, place.place_type <chr>, place.url <chr>,
## #   quote_count <dbl>, reply_count <dbl>, retweet_count <dbl>,
## #   retweeted <lgl>, source <chr>, text <chr>, timestamp_ms <chr>,
## #   truncated <lgl>, user.contributors_enabled <lgl>,
## #   user.created_at <chr>, user.default_profile <lgl>,
## #   user.default_profile_image <lgl>, user.description <chr>,
## #   user.favourites_count <dbl>, user.follow_request_sent <int>,
## #   user.followers_count <dbl>, user.following <int>,
## #   user.friends_count <dbl>, user.geo_enabled <lgl>, user.id <dbl>,
## #   user.id_str <chr>, user.is_translator <lgl>, user.lang <chr>,
## #   user.listed_count <dbl>, user.location <chr>, user.name <chr>,
## #   user.notifications <int>, user.profile_background_color <chr>,
## #   user.profile_background_image_url <chr>,
## #   user.profile_background_image_url_https <chr>,
## #   user.profile_background_tile <lgl>, user.profile_banner_url <chr>,
## #   user.profile_image_url <chr>, user.profile_image_url_https <chr>,
## #   user.profile_link_color <chr>,
## #   user.profile_sidebar_border_color <chr>,
## #   user.profile_sidebar_fill_color <chr>, user.profile_text_color <chr>,
## #   user.profile_use_background_image <lgl>, user.protected <lgl>,
## #   user.screen_name <chr>, user.statuses_count <dbl>,
## #   user.time_zone <int>, user.translator_type <chr>, user.url <chr>,
## #   user.utc_offset <int>, user.verified <lgl>,
## #   entities.urls.0.display_url <chr>, entities.urls.0.expanded_url <chr>,
## #   entities.urls.0.indices.0 <dbl>, entities.urls.0.indices.1 <dbl>,
## #   entities.urls.0.url <chr>, entities.user_mentions.2.id <dbl>,
## #   entities.user_mentions.2.id_str <chr>,
## #   entities.user_mentions.2.indices.0 <dbl>, …

Creating your own token

Follow these steps to create your own token after your application has been approved:

Go to https://developer.twitter.com/en/apps and sign in.
If you don’t have a developer account, you will need to apply for one first. Fill in the application form and wait for a response.
Once it’s approved, click on “Create New App”. You will need to have a phone number associated with your account in order to be able to create a token.
Fill name, description, and website (it can be anything, even http://www.google.com). Make sure you leave ‘Callback URL’ empty.
Agree to user conditions.
From the “Keys and Access Tokens” tab, copy consumer key and consumer secret and paste below
Click on “Create my access token”, then copy and paste your access token and access token secret below

library(ROAuth)
my_oauth <- list(consumer_key = "CONSUMER_KEY",
   consumer_secret = "CONSUMER_SECRET",
   access_token="ACCESS_TOKEN",
   access_token_secret = "ACCESS_TOKEN_SECRET")
save(my_oauth, file="~/my_oauth")

load("~/my_oauth")

What can go wrong here? Make sure all the consumer and token keys are pasted here as is, without any additional space character. If you don’t see any output in the console after running the code above, that’s a good sign.

Note that I saved the list as a file in my hard drive. That will save us some time later on, but you could also just re-run the code in lines 22 to 27 before conecting to the API in the future.