Justin Chun-ting Ho
5/25/2017
This vignette presents a Keyness Analysis of the posts and comments of the Scottish National Party (SNP) and the UK Independence Party (UKIP). Keyness refers to the quality a word (or a feature) has of being "key" in differentiating one corpus from another. In other words, keywords are "items of unusual frequency in comparison with a reference corpus".
This vignette is going to use these five packages:
require(readtext)
require(quanteda)
require(magrittr)
require(dplyr)
require(wordcloud)Next, we are going to load the corpura. The corpura were collected in May 2017 from SNP and UKIP's public Facebook page, unfortunately, the corpura are too large to be uploaded onto Gihub, but the codes in this vignette can be used on most other corpura.
snp <- corpus(readtext("SNP_fullstats.csv", textfield = "post_message"))
docvars(snp, "Party") <- "SNP"We could use cat() to have a peck.
cat(snp[1:3]) # Displaying the first 3 documents## John Swinney SNP met young people today to talk votes at 16 and 17 In 21st century Scotland it is ridiculous that anyone should suffer from food poverty. The UK Government s welfare cuts have placed some of the most vulnerable people and increasing numbers of working people and their families too into poverty. - SNP MSP Christina McKelvie Sad news today with the death of Charlie Kennedy.
The folowing codes do two things: first, tokenising the text into tokens; second, creating a Document-Feature Matrix from the tokens.
snptokens <- tokens(snp, remove_punct = TRUE, remove_numbers = TRUE, verbose = FALSE, remove_url = TRUE)
snpdfm <- dfm(snptokens, remove = c(stopwords('english'), "s", "http")) %>%
dfm_trim(min_doc = 5, min_count = 10) %>%
dfm_weight(type = 'tfidf')We can use topfeatures() to see the most common words of the corpus.
topfeatures(snpdfm)## scotland snp scottish government nicola sturgeon
## 324.9589 305.3087 299.2164 254.0677 250.3998 246.2121
## will party national read
## 234.8084 232.3340 231.6727 223.1298
And, of course, we can do wordcloud too.
textplot_wordcloud(snpdfm)
Now we do the same thing with the UKIP corpus.
ukip <- corpus(readtext("ukip_fullstats.csv", textfield = "post_message"))
cat(ukip[1:3])## Voters will be asked 'Should the United Kingdom remain a member of the European Union? when they go to the polls for the in/out referendum it has been confirmed. In response UKIP Leader Nigel Farage said: It is a simple straightforward unambiguous question. That much is clear. However that Cameron is opting to give the pro-EU side the positive Yes suggests strongly that his negotiations are so much fudge. He has already decided which way he wants the answer to be given without a single power repatriated. UKIP Deputy Leader Paul Nuttall MEP: "Sepp Blatter's position as FIFA president was already on shaky ground but now surely no one will argue that it is in any way tenable. "Pretty much all of the crimes being alleged happened on his watch. If he knew about it then he should be prosecuted. If he didn't know about it then he wasn't doing his job properly and should be removed. Grassroots football in some of the world's poorest countries and fans everywhere are the apparent victims of these crimes. "Only a root and branch clear out at FIFA will satisfy them that we're cleaning up the game once and for all." UKIP MEP Louise Bours said: The EU continues to suffer economic and unemployment disaster yet the EC wastes taxpayers money to decide how much coffee adults should drink. This study by the EU food safety quango is absurd and wasteful. Its findings are patronising for pregnant women. They have medical advice. They don t need bossy Brussels.
docvars(ukip, "Party") <- "UKIP"
ukiptokens <- tokens(ukip, remove_punct = TRUE, remove_numbers = TRUE, verbose = FALSE, remove_url = TRUE)
ukipdfm <- dfm(ukiptokens, remove = c(stopwords('english'), "s", "http")) %>%
dfm_trim(min_doc = 5, min_count = 10) %>%
dfm_weight(type = 'tfidf')
topfeatures(ukipdfm)## eu will people uk nigel farage european leader
## 329.3331 298.1561 227.6134 223.4786 217.1207 210.0337 209.7742 207.2822
## mep can
## 204.7251 192.2940
Now, we are all set up for the Keyness Analysis, in the quantenda package, the code would be textstat_keyness().
kwdspage <- textstat_keyness(rbind(snpdfm, ukipdfm), target = seq_along(snptokens))These are the top 20 SNP words:
head(kwdspage, 20)## chi2 p n_target n_reference
## snp 469.41812 0 305.30873 0.00000
## nicola 384.21911 0 250.39982 0.00000
## sturgeon 377.72699 0 246.21205 0.00000
## scottish 369.71247 0 299.21639 28.46064
## scotland 335.16918 0 324.95895 56.49398
## scotland's 204.54171 0 134.30168 0.00000
## shared 179.38832 0 181.70650 34.45871
## national 138.60282 0 231.67272 92.79983
## pledge 137.09242 0 90.61299 0.00000
## #bothvotessnp 132.94242 0 87.92297 0.00000
## read 129.80989 0 223.12976 91.84224
## mhairi 123.86770 0 82.03994 0.00000
## results 119.51884 0 79.22024 0.00000
## black 112.99535 0 74.99006 0.00000
## holyrood 110.16541 0 73.15479 0.00000
## cuts 110.04552 0 73.07704 0.00000
## tories 109.48939 0 128.43313 32.10025
## angus 102.97553 0 68.49151 0.00000
## poverty 99.81500 0 66.44139 0.00000
## robertson 99.28685 0 66.09878 0.00000
These are the top 20 UKIP words:
tail(kwdspage, 20)## chi2 p n_target n_reference
## whittle -54.73329 1.380007e-13 0 86.63295
## spread -57.77242 2.942091e-14 0 91.29813
## peter -58.51605 2.020606e-14 0 92.43953
## believe -59.59318 1.165734e-14 0 94.09272
## says -63.55804 1.554312e-15 0 100.17725
## migration -64.48608 9.992007e-16 0 101.60125
## borders -67.69474 2.220446e-16 0 106.52415
## defence -75.28798 0.000000e+00 0 118.17100
## immigration -78.47525 0.000000e+00 0 123.05849
## control -82.20145 0.000000e+00 0 128.77142
## spokesman -88.69143 0.000000e+00 0 138.71925
## nuttall -94.65383 0.000000e+00 0 147.85569
## paul -100.01647 0.000000e+00 0 156.07088
## britain -106.73507 0.000000e+00 0 166.36034
## london -107.67552 0.000000e+00 0 167.80038
## ukip -110.17431 0.000000e+00 0 171.62625
## british -123.45259 0.000000e+00 0 191.94895
## mep -131.80442 0.000000e+00 0 204.72514
## farage -135.27562 0.000000e+00 0 210.03374
## nigel -139.91057 0.000000e+00 0 217.12074
Not surprisingly, SNP uses Scotland, Scotish and the name of Nicola Sturgeon distinctively. We can also see the UKIP favours words like control, immigration, migration, borders, and defence.
Let's turn to comments, same process.
snpcom <- corpus(readtext("snp_comments.csv", textfield = "comment_message"))##
Read 14.0% of 571079 rows
Read 29.8% of 571079 rows
Read 45.5% of 571079 rows
Read 63.0% of 571079 rows
Read 85.8% of 571079 rows
Read 571079 rows and 11 (of 11) columns from 0.303 GB file in 00:00:08
docvars(snpcom, "Party") <- "SNP"
snpcomtokens <- tokens(snpcom, remove_punct = TRUE, remove_numbers = TRUE, verbose = FALSE, remove_url = TRUE)
snpcomdfm <- dfm(snpcomtokens, remove = c(stopwords('english'), "s", "http")) %>%
dfm_trim(min_doc = 5, min_count = 10) %>%
dfm_weight(type = 'tfidf')
topfeatures(snpcomdfm) # This is not necessarily but it is a good practice to check anyway## scotland will eu snp uk people scottish
## 108309.14 90830.64 90092.99 89424.29 78747.24 76896.57 62582.38
## vote just get
## 61597.05 60872.98 58624.02
ukipcom <- corpus(readtext("ukip_comments.csv", textfield = "comment_message"))##
Read 10.6% of 660106 rows
Read 34.8% of 660106 rows
Read 71.2% of 660106 rows
Read 660106 rows and 11 (of 11) columns from 0.468 GB file in 00:00:06
docvars(ukipcom, "Party") <- "UKIP"
ukipcomtokens <- tokens(ukipcom, remove_punct = TRUE, remove_numbers = TRUE, verbose = FALSE, remove_url = TRUE)
ukipcomdfm <- dfm(ukipcomtokens, remove = c(stopwords('english'), "s", "http")) %>%
dfm_trim(min_doc = 5, min_count = 10) %>%
dfm_weight(type = 'tfidf')
topfeatures(ukipcomdfm) # This is not necessarily but it is a good practice to check anyway## eu will people ukip get vote country
## 111568.08 110107.28 92962.62 77370.46 76432.67 73609.10 72282.08
## us just can
## 70046.42 65556.07 65304.12
kwds <- textstat_keyness(rbind(snpcomdfm, ukipcomdfm), target = seq_along(snpcomtokens))Again, these are the top 20 SNP words in comments:
head(kwds, 20)## chi2 p n_target n_reference
## scotland 87701.108 0 108309.141 10048.7205
## snp 78186.327 0 89424.293 5966.6822
## scottish 52514.700 0 62582.383 5030.6388
## independence 29727.543 0 47712.266 8507.5686
## westminster 29369.630 0 39757.327 4879.9413
## nicola 28708.892 0 30743.953 1346.7619
## sturgeon 21367.725 0 26692.613 2559.5433
## scots 20399.767 0 26142.905 2729.8967
## indy 19031.128 0 18453.437 176.3451
## independent 18397.411 0 29158.515 5076.3223
## scotland's 17706.387 0 17920.135 435.7083
## oil 16763.220 0 23803.282 3292.8498
## unionist 13183.916 0 13092.237 232.0713
## wee 12033.249 0 13319.651 731.4149
## unionists 9852.095 0 9600.033 106.9893
## england 9188.910 0 35976.746 15707.0689
## deficit 8924.705 0 12673.795 1751.9780
## holyrood 8123.195 0 8192.752 188.3765
## salmond 7846.873 0 8446.644 382.8388
## devolved 7545.925 0 8274.777 427.3872
These are the top 20 UKIP words in comments:
tail(kwds, 20)## chi2 p n_target n_reference
## scum -6023.550 0 0.0000 6385.105
## country -6047.861 0 42627.5595 72282.083
## immigrants -6198.431 0 7254.6547 21407.006
## jo -6367.329 0 973.3675 9397.341
## immigration -6685.851 0 7742.2024 22948.042
## nuttall -7826.041 0 119.1854 8659.602
## muslim -8298.993 0 2247.7873 14528.127
## muslims -8555.205 0 2046.8837 14367.830
## dutta -8796.708 0 0.0000 9323.125
## wiseman -9555.915 0 0.0000 10127.396
## argha -9672.758 0 0.0000 10251.171
## british -11993.265 0 18701.2234 49169.705
## britain -12128.124 0 14601.7433 42566.012
## cameron -16172.198 0 12135.2110 43807.392
## fuhrer -17705.271 0 181.4220 19316.995
## mein -18146.532 0 147.8826 19682.155
## bigot -23291.104 0 1813.2021 29884.800
## farage -29398.089 0 2809.5532 39093.984
## nigel -51019.847 0 1772.0113 59307.093
## ukip -56421.099 0 6319.1029 77370.458
To visualise it, we can generate a comparison cloud. In the following chunck, I am selecting all the keywords with p-value equals to or small than 0.05.
kwdssig <- data.frame(term = row.names(kwds), chi2 = kwds$chi2, p=kwds$p) %>%
filter(kwds$p <= 0.05) %>%
select(term, chi2)
row.names(kwdssig) <- kwdssig$term
kwdssig$SNP <- kwdssig$chi2
kwdssig$UKIP <- kwdssig$chi2
kwdssig$UKIP[kwdssig$UKIP > 0] <- 0
kwdssig$SNP[kwdssig$SNP < 0] <- 0
kwdssig <- kwdssig[,-1:-2]To plot the comparison cloud.
set.seed(1024)
comparison.cloud(kwdssig, random.order=FALSE, colors = c("goldenrod1","blueviolet"),scale=c(6,.6), title.size=3, max.words=500)