rstudio::conf tweets

A repository for tracking tweets about rstudio::conf 2018. Read more about the Rstudio conference at rstudio.com/conference/.

• Data
  • rtweet
• Twitter APIs
  • Stream
  • Search
• Explore
  • Tweet frequency over time
  • Positive/negative sentiment
  • Semantic networks
  • Tidyverse vs. Shiny
  • Word clouds

Data

Two data collection methods are described in detail below. Hoewver, if you want to skip straight to the data, run the following code:

## download status IDs file
download.file(
  "https://github.com/mkearney/rstudioconf_tweets/blob/master/data/search-ids.rds?raw=true",
  "rstudioconf_search-ids.rds"
)

## read status IDs fromdownloaded file
ids <- readRDS("rstudioconf_search-ids.rds")

## lookup data associated with status ids
rt <- rtweet::lookup_tweets(ids$status_id)

rtweet

Whether you lookup the status IDs or search/stream new tweets, make sure you've installed the rtweet package. The code below will install [if it's not already] and load rtweet.

## install rtweet if not already
if (!requireNamespace("rtweet", quietly = TRUE)) {
  install.packages("rtweet")
}

## load rtweet
library(rtweet)

Twitter APIs

There are two easy [and free] ways to get lots of Twitter data, filtering by one or more keywords. Each method is described and demonstrated below.

Stream

The first way is to stream the data (using Twitter's stream API). For example, in the code below, a stream is setup to run continuously from the moment its executed until the Saturday at midnight (to roughly coincide with the end of the conference).

## set stream time
timeout <- as.numeric(
  difftime(as.POSIXct("2018-02-04 00:00:00"),
  Sys.time(), tz = "US/Pacific", "secs")
)

## search terms
rstudioconf <- c("rstudioconf", "rstudio::conf",
  "rstudioconference", "rstudioconference18",
  "rstudioconference2018", "rstudio18",
  "rstudioconf18", "rstudioconf2018",
  "rstudio::conf18", "rstudio::conf2018")

## name of file to save output
json_file <- file.path("data", "stream.json")

## stream the tweets and write to "data/stream.json"
stream_tweets(
  q = paste(rstudioconf, collapse = ","),
  timeout = timeout,
  file_name = json_file,
  parse = FALSE
)

## parse json data and convert to tibble
rt <- parse_stream(json_file)

Search

The second easy way to gather Twitter data using one or more keywords is to search for the data (using Twitter's REST API). Unlike streaming, searching makes it possible to go back in time. Unfortunately, Twitter sets a rather restrictive cap–roughly nine days–on how far back you can go. Regardless, searching for tweets is often the preferred method. For example, the code below is setup in such a way that it can be executed once [or even several times] a day throughout the conference.

## search terms
rstudioconf <- c("rstudioconf", "rstudio::conf",
  "rstudioconference", "rstudioconference18",
  "rstudioconference2018", "rstudio18",
  "rstudioconf18", "rstudioconf2018",
  "rstudio::conf18", "rstudio::conf2018")

## use since_id from previous search (if exists)
if (file.exists(file.path("data", "search.rds"))) {
  since_id <- readRDS(file.path("data", "search.rds"))
  since_id <- since_id$status_id[1]
} else {
  since_id <- NULL
}

## search for up to 100,000 tweets mentionging rstudio::conf
rt <- search_tweets(
  paste(rstudioconf, collapse = " OR "),
  n = 1e5, verbose = FALSE,
  since_id = since_id,
  retryonratelimit = TRUE
)

## if there's already a search data file saved, then read it in,
## drop the duplicates, and then update the `rt` data object
if (file.exists(file.path("data", "search.rds"))) {

  ## bind rows (for tweets AND users data)
  rt <- do_call_rbind(
    list(rt, readRDS(file.path("data", "search.rds"))))

  ## determine whether each observation has a unique status ID
  kp <- !duplicated(rt$status_id)

  ## only keep rows (observations) with unique status IDs
  users <- users_data(rt)[kp, ]

  ## the rows of users should correspond with the tweets
  rt <- rt[kp, ]

  ## restore as users attribute
  attr(rt, "users") <- users
}

## save the data
saveRDS(rt, file.path("data", "search.rds"))

## save shareable data (only status_ids)
saveRDS(rt[, "status_id"], file.path("data", "search-ids.rds"))

Explore

To explore the Twitter data, go ahead and load the tidyverse packages.

suppressPackageStartupMessages(library(tidyverse))

Tweet frequency over time

In the code below, the data is summarized into a time series-like data frame and then plotted in order depict the frequency of tweets–aggregated using 2-hour intevals–about rstudio::conf over time.

rt %>%
  filter(created_at > "2018-01-29") %>%
  ts_plot("2 hours", color = "transparent") +
  geom_smooth(method = "loess", se = FALSE, span = .1,
  size = 2, colour = "#0066aa") +
  geom_point(size = 5,
    shape = 21, fill = "#ADFF2F99", colour = "#000000dd") +
  theme_minimal(base_size = 15, base_family = "Roboto Condensed") +
  theme(axis.text = element_text(colour = "#222222"),
    plot.title = element_text(size = rel(1.7), face = "bold"),
    plot.subtitle = element_text(size = rel(1.3)),
    plot.caption = element_text(colour = "#444444")) +
  labs(title = "Frequency of tweets about rstudio::conf over time",
    subtitle = "Twitter status counts aggregated using two-hour intervals",
    caption = "\n\nSource: Data gathered via Twitter's standard `search/tweets` API using rtweet",
    x = NULL, y = NULL)

 

Positive/negative sentiment

Next, some sentiment analysis of the tweets so far.

## clean up the text a bit (rm mentions and links)
rt$text2 <- gsub(
  "^RT:?\\s{0,}|#|@\\S+|https?[[:graph:]]+", "", rt$text)
## convert to lower case
rt$text2 <- tolower(rt$text2)
## trim extra white space
rt$text2 <- gsub("^\\s{1,}|\\s{1,}$", "", rt$text2)
rt$text2 <- gsub("\\s{2,}", " ", rt$text2)

## estimate pos/neg sentiment for each tweet
rt$sentiment <- syuzhet::get_sentiment(rt$text2, "syuzhet")

## write function to round time into rounded var
round_time <- function(x, sec) {
  as.POSIXct(hms::hms(as.numeric(x) %/% sec * sec))
}

## plot by specified time interval (1-hours)
rt %>%
  mutate(time = round_time(created_at, 60 * 60)) %>%
  group_by(time) %>%
  summarise(sentiment = mean(sentiment, na.rm = TRUE)) %>%
  mutate(valence = ifelse(sentiment > 0L, "Positive", "Negative")) %>%
  ggplot(aes(x = time, y = sentiment)) +
  geom_smooth(method = "loess", span = .1,
    colour = "#aa11aadd", fill = "#bbbbbb11") +
  geom_point(aes(fill = valence, colour = valence), 
    shape = 21, alpha = .6, size = 3.5) +
  theme_minimal(base_size = 15, base_family = "Roboto Condensed") +
  theme(legend.position = "none",
    axis.text = element_text(colour = "#222222"),
    plot.title = element_text(size = rel(1.7), face = "bold"),
    plot.subtitle = element_text(size = rel(1.3)),
    plot.caption = element_text(colour = "#444444")) +
  scale_fill_manual(
    values = c(Positive = "#2244ee", Negative = "#dd2222")) +
  scale_colour_manual(
    values = c(Positive = "#001155", Negative = "#550000")) +
  labs(x = NULL, y = NULL,
    title = "Sentiment (valence) of rstudio::conf tweets over time",
    subtitle = "Mean sentiment of tweets aggregated in one-hour intervals",
    caption = "\nSource: Data gathered using rtweet. Sentiment analysis done using syuzhet")

 

Semantic networks

The code below provides a quick and dirty visualization of the semantic network (connections via retweet, quote, mention, or reply) found in the data.

## unlist observations into long-form data frame
unlist_df <- function(...) {
  dots <- lapply(list(...), unlist)
  tibble::as_tibble(dots)
}

## iterate by row
row_dfs <- lapply(
  seq_len(nrow(rt)), function(i)
    unlist_df(from_screen_name = rt$screen_name[i],
      reply = rt$reply_to_screen_name[i],
      mention = rt$mentions_screen_name[i],
      quote = rt$quoted_screen_name[i],
      retweet = rt$retweet_screen_name[i])
)

## bind rows, gather (to long), convert to matrix, and filter out NAs
rdf <- dplyr::bind_rows(row_dfs)
rdf <- tidyr::gather(rdf, interaction_type, to_screen_name, -from_screen_name)
mat <- as.matrix(rdf[, -2])
mat <- mat[apply(mat, 1, function(i) !any(is.na(i))), ]

## get rid of self references
mat <- mat[mat[, 1] != mat[, 2], ]

## filter out users who don't appear in RHS at least 3 times
apps1 <- table(mat[, 1])
apps1 <- apps1[apps1 > 1L]
apps2 <- table(mat[, 2])
apps2 <- apps2[apps2 > 1L]
apps <- names(apps1)[names(apps1) %in% names(apps2)]
mat <- mat[mat[, 1] %in% apps & mat[, 2] %in% apps, ]

## create graph object
g <- igraph::graph_from_edgelist(mat)

## calculate size attribute (and transform to fit)
matcols <- factor(c(mat[, 1], mat[, 2]), levels = names(igraph::V(g)))
size <- table(screen_name = matcols)
size <- (log(size) + sqrt(size)) / 3

## reorder freq table
size <- size[match(names(size), names(igraph::V(g)))]

## plot network
par(mar = c(12, 6, 15, 6))
plot(g,
  edge.size = .4,
  curved = FALSE,
  margin = -.05,
  edge.arrow.size = 0,
  edge.arrow.width = 0,
  vertex.color = "#ADFF2F99",
  vertex.size = size,
  vertex.frame.color = "#003366",
  vertex.label.color = "#003366",
  vertex.label.cex = .8,
  vertex.label.family = "Roboto Condensed",
  edge.color = "#0066aa",
  edge.width = .2,
  main = "")
par(mar = c(9, 6, 9, 6))
title("Semantic network of users tweeting about rstudio::conf",
  adj = 0, family = "Roboto Condensed", cex.main = 6.5)
mtext("Source: Data gathered using rtweet. Network analysis done using igraph",
  side = 1, line = 0, adj = 1.0, cex = 3.8,
  family = "Roboto Condensed", col = "#222222")
mtext("User connections by mentions, replies, retweets, and quotes",
  side = 3, line = -4.25, adj = 0,
  family = "Roboto Condensed", cex = 4.9)

 

Ideally, the network visualization would be an interactive, searchable graphic. Since it's not, I've printed out the node size values below.

nodes <- as_tibble(sort(size, decreasing = TRUE))
nodes$rank <- seq_len(nrow(nodes))
nodes$screen_name <- paste0(
  '<a href="https://twitter.com/', nodes$screen_name, 
  '">@', nodes$screen_name, '</a>')
dplyr::select(nodes, rank, screen_name, log_n = n)

rank	screen_name	log_n
1	@hadleywickham	11.0164053
2	@robinson_es	10.7205729
3	@drob	9.8275619
4	@rstudio	9.0122754
5	@juliasilge	8.3156539
6	@RLadiesGlobal	7.4782542
7	@LucyStats	7.4112162
8	@AmeliaMN	7.1228108
9	@dataandme	7.0632396
10	@sharlagelfand	6.7674473
11	@d4tagirl	6.7547438
12	@JennyBryan	6.4124132
13	@Voovarb	6.0284818
14	@romain_francois	6.0284818
15	@ellisvalentiner	5.9837260
16	@EmilyRiederer	5.7999183
17	@CivicAngela	5.7842362
18	@sharon000	5.7842362
19	@sconvers	5.7049294
20	@astroeringrand	5.6888842
21	@eamcvey	5.6888842
22	@stephhazlitt	5.6403685
23	@kearneymw	5.5581979
24	@SK_convergence	5.5248529
25	@visnut	5.3186034
26	@CMastication	5.2831274
27	@kara_woo	5.2652635
28	@datapointier	5.1929426
29	@thmscwlls	5.1562458
30	@tanyacash21	5.1191779
31	@njogukennly	5.0056364
32	@old_man_chester	4.8683901
33	@minebocek	4.8281541
34	@elhazen	4.5540652
35	@CorradoLanera	4.5097581
36	@ijlyttle	4.4192141
37	@AlexisLNorris	4.3961566
38	@kierisi	4.3961566
39	@juliesquid	4.3961566
40	@malco_bearhat	4.3259141
41	@thomas_mock	4.3259141
42	@nj_tierney	4.3021301
43	@jimhester_	4.3021301
44	@edzerpebesma	4.3021301
45	@thomasp85	4.2296099
46	@ds_floresf	4.2296099
47	@rudeboybert	4.2050318
48	@dvaughan32	4.1045644
49	@shermstats	4.0267962
50	@jent103	4.0267962
51	@grrrck	3.9195611
52	@theRcast	3.9195611
53	@jasongrahn	3.8920720
54	@_RCharlie	3.8920720
55	@alice_data	3.8642952
56	@jafflerbach	3.8362222
57	@ajmcoqui	3.8362222
58	@therriaultphd	3.8078439
59	@Bluelion0305	3.7791511
60	@sgrifter	3.7501339
61	@RLadiesBA	3.7501339
62	@yutannihilation	3.7207821
63	@taraskaduk	3.6910847
64	@cantoflor_87	3.6910847
65	@jonmcalder	3.6910847
66	@jcheng	3.6306068
67	@nic_crane	3.5998014
68	@seankross	3.5998014
69	@ma_salmon	3.5686007
70	@DataActivism	3.5369905
71	@cpsievert	3.5049555
72	@gdequeiroz	3.5049555
73	@kevin_ushey	3.5049555
74	@daattali	3.4724797
75	@Dorris_Scott	3.4724797
76	@Blair09M	3.4395462
77	@PyDataBA	3.3722321
78	@sellorm	3.3722321
79	@claytonyochum	3.3378116
80	@simecek	3.3378116
81	@MangoTheCat	3.3028532
82	@lariebyrd	3.2673334
83	@krlmlr	3.2312268
84	@jessenleon	3.1945063
85	@paylakatel	3.1191041
86	@drvnanduri	3.0803567
87	@bhive01	3.0803567
88	@zevross	3.0408634
89	@RLadiesMVD	3.0408634
90	@aindap	3.0005839
91	@ucdlevy	3.0005839
92	@ntweetor	3.0005839
93	@tnederlof	2.9594743
94	@JonathanZadra	2.9594743
95	@Denironyx	2.9594743
96	@just_add_data	2.9594743
97	@OilGains	2.9594743
98	@duto_guerra	2.9174869
99	@dmi3k	2.9174869
100	@jarvmiller	2.9174869
101	@javierluraschi	2.8745690
102	@danielphadley	2.8745690
103	@bizScienc	2.8745690
104	@SanghaChick	2.8745690
105	@CVWickham	2.8745690
106	@nicoleradziwill	2.8745690
107	@patsellers	2.8306631
108	@RhoBott	2.8306631
109	@RLadiesOrlando	2.8306631
110	@deekareithi	2.8306631
111	@GioraSimchoni	2.7857054
112	@BaumerBen	2.7857054
113	@mmmpork	2.7857054
114	@alandipert	2.7396253
115	@NovasTaylor	2.7396253
116	@millerdl	2.7396253
117	@sheilasaia	2.7396253
118	@RLadiesNYC	2.6923444
119	@conjja	2.6923444
120	@SHaymondSays	2.6437752
121	@PinnacleSports	2.6437752
122	@MineDogucu	2.5938194
123	@Nujcharee	2.5938194
124	@S_Owla	2.5938194
125	@theotheredgar	2.5938194
126	@chrisderv	2.5938194
127	@mfairbrocanada	2.5938194
128	@plzbeemyfriend	2.5423660
129	@dnlmc	2.5423660
130	@egolinko	2.5423660
131	@data_stephanie	2.4892894
132	@jabbertalk	2.4892894
133	@jamie_jezebel	2.4892894
134	@OHIscience	2.4892894
135	@webbedfeet	2.4344460
136	@ParmutiaMakui	2.3776708
137	@R_by_Ryo	2.3776708
138	@SDanielZafar1	2.3776708
139	@hrbrmstr	2.3776708
140	@LuisDVerde	2.3187730
141	@_jwinget	2.3187730
142	@ROfficeHours	2.2575296
143	@rweekly_org	2.2575296
144	@b23kelly	2.2575296
145	@kyrietree	2.2575296
146	@hugobowne	2.2575296
147	@ibddoctor	2.1936778
148	@hspter	2.1936778
149	@jhollist	2.1936778
150	@jakethomp	2.1936778
151	@markroepke	2.1936778
152	@pacocuak	2.1936778
153	@DaveQuartey	2.1269049
154	@DoITBoston	2.1269049
155	@wmlandau	2.1269049
156	@RLadiesColumbus	2.1269049
157	@RLadiesNash	2.1269049
158	@rick_pack2	2.0568335
159	@jo_hardin47	2.0568335
160	@butterflyology	2.0568335
161	@RiinuOts	2.0568335
162	@darokun	2.0568335
163	@klausmiller	1.9830028
164	@dantonnoriega	1.9830028
165	@katie_leap	1.9830028
166	@kpivert	1.9830028
167	@brad_cannell	1.9830028
168	@dobbleobble	1.9830028
169	@_ColinFay	1.9830028
170	@math_dominick	1.9048400
171	@RobynLBall	1.9048400
172	@ericcolson	1.9048400
173	@KurggMantra	1.9048400
174	@tcbanalytics	1.9048400
175	@flatironhealth	1.8216209
176	@eirenkate	1.8216209
177	@revodavid	1.8216209
178	@harrismcgehee	1.7324082
179	@jherndon01	1.7324082
180	@_NickGolding_	1.7324082
181	@samhinshaw	1.7324082
182	@DJShearwater	1.7324082
183	@runnersbyte	1.7324082
184	@ImagineBos	1.7324082
185	@modmed	1.7324082
186	@rdpeng	1.7324082
187	@JonTheGeek	1.7324082
188	@chrisalbon	1.7324082
189	@n_ashutosh	1.7324082
190	@jblistman	1.6359562
191	@balling_cc	1.6359562
192	@__mharrison__	1.6359562
193	@TrestleJeff	1.6359562
194	@EarlGlynn	1.6359562
195	@OmniaRaouf	1.6359562
196	@jdblischak	1.6359562
197	@JeanetheFalvey	1.6359562
198	@harry_seunghoon	1.6359562
199	@alichiang13	1.6359562
200	@volha_tryputsen	1.5305538
201	@abresler	1.5305538
202	@aaronchall	1.5305538
203	@AllenDowney	1.5305538
204	@tonyfujs	1.5305538
205	@maryclaryf	1.5305538
206	@uncmbbtrivia	1.5305538
207	@KirkD_CO	1.4137497
208	@rsangole	1.4137497
209	@msciain	1.4137497
210	@jebyrnes	1.4137497
211	@ledell	1.4137497
212	@ben_d_best	1.4137497
213	@canoodleson	1.4137497
214	@benjamingreve	1.4137497
215	@tonmcg	1.4137497
216	@zymla	1.4137497
217	@strnr	1.2818353
218	@LauraBBalzer	1.2818353
219	@JTLewis5	1.2818353
220	@clairemcwhite	1.2818353
221	@s_pearce	1.2818353
222	@jomilo75	1.2818353
223	@nwstephens	1.2818353
224	@Emil_Hvitfeldt	1.2818353
225	@kwbroman	1.2818353
226	@VParrillaAixela	1.2818353
227	@ClausWilke	1.1287648
228	@sgpln	1.1287648
229	@lorenzwalthert	1.1287648
230	@hoxo_m	1.1287648
231	@obergr	1.1287648
232	@drewconway	1.1287648
233	@iainmwallace	1.1287648
234	@arjunsbaghela	1.1287648
235	@_lionelhenry	1.1287648
236	@olgavitek	1.1287648
237	@crozierrj	1.1287648
238	@msimas	1.1287648
239	@CaltechChemLib	1.1287648
240	@AriLamstein	1.1287648
241	@bogdanrau	1.1287648
242	@RLadiesTC	1.1287648
243	@jduckles	0.9435544
244	@Md_Harris	0.9435544
245	@aronatkins	0.9435544
246	@awunderground	0.9435544
247	@ukacz	0.9435544
248	@itsrainingdata	0.9435544
249	@EricLeeKrantz	0.7024536
250	@PeterSForbes	0.7024536
251	@wahalulu	0.7024536
252	@jkassof	0.7024536
253	@nmhouston	0.7024536
254	@alathrop	0.7024536
255	@bj_bloom	0.7024536
256	@jaredlander	0.7024536

Tidyverse vs. Shiny

This code identifies tweets by topic, detecting mentions of the tidyverse [packages] and shiny. It then plots the frequency of those tweets over time.

rt %>%
  filter(created_at > "2018-02-01" & created_at < "2018-02-05") %>%
  mutate(
    text = tolower(text),
    tidyverse = str_detect(
      text, "dplyr|tidyeval|tidyverse|rlang|map|purrr|readr|tibble"),
    shiny = str_detect(text, "shiny|dashboard|interactiv")
  ) %>%
  select(created_at, tidyverse:shiny) %>%
  gather(pkg, mention, -created_at) %>%
  mutate(pkg = factor(pkg, labels = c("Shiny", "Tidyverse"))) %>%
  filter(mention) %>%
  group_by(pkg) %>%
  ts_plot("2 hours") +
  geom_point(shape = 21, size = 3, aes(fill = pkg)) + 
  theme_minimal(base_family = "Roboto Condensed") + 
  scale_x_datetime(timezone = "America/Los_Angelos") + 
  theme(legend.position = "bottom",
    legend.title = element_blank(),
    legend.text = element_text(size = rel(1.1)),
    axis.text = element_text(colour = "#222222"),
    plot.title = element_text(size = rel(1.7), face = "bold"),
    plot.subtitle = element_text(size = rel(1.3)),
    plot.caption = element_text(colour = "#444444")) +
  scale_fill_manual(
    values = c(Tidyverse = "#2244ee", Shiny = "#dd2222")) +
  scale_colour_manual(
    values = c(Tidyverse = "#001155", Shiny = "#550000")) +
  labs(x = NULL, y = NULL,
    title = "Frequency of tweets about Tidyverse and Shiny during rstudio::conf",
    subtitle = "Tweet counts aggregated for each topic in two-hour intervals",
    caption = "\nSource: Data gathered using rtweet. Made pretty by ggplot2.")

 

Word clouds

I didn't want to add a bunch more code, so here I'm sourcing the prep work/code I used to get word lists.

source(file.path("R", "words.R"))

Shiny word cloud

This first word cloud depicts the most popular non-stopwords used in tweets about Shiny.

par(mar = c(0, 0, 0, 0))
wordcloud::wordcloud(
  shiny$var, shiny$n, min.freq = 3,
  random.order = FALSE,
  random.color = FALSE,
  colors = gg_cols(5)
)

 

Tidyverse word cloud

The second word cloud depicts the most popular non-stopwords used in tweets about the tidyverse.

par(mar = c(0, 0, 0, 0))
wordcloud::wordcloud(
  tidyverse$var, tidyverse$n, min.freq = 5,
  random.order = FALSE,
  random.color = FALSE,
  colors = gg_cols(5)
)