Unmanned Store Analysis

Goals

Step 1: Upload & clean text into quanteda gui

Step 2: Analysis

Sentiment, least priority (can be done in SPSS)

• General sentiment
• Sentiment on individual accounts (variable in row O)
• Sentiment by continent (variable row K or L)
• Sentiment by country (variable row M)

Word usage

• Most used words minus stop words
• Most used words per continent
• Most used negative words (but most entries are neutral)
• Most used positive words (but most entries are neutral)

KWIC

• Payment & checkout
• Payment & checkout => sentiment
• Access & check-in

Load the dataset

library("quanteda")

Package version: 2.9.9000
Unicode version: 10.0
ICU version: 61.1
Parallel computing: 12 of 12 threads used.
See https://quanteda.io for tutorials and examples.

library("dplyr")

Attaching package: ‘dplyr’

The following objects are masked from ‘package:stats’:

    filter, lag

The following objects are masked from ‘package:base’:

    intersect, setdiff, setequal, union

corp <- read.csv("~/Dropbox/_Whittlesey Street/Unmanned Stores Data.csv") %>%
    corpus(text_field = "Full.Text")

Descriptives

Page type by sentiment (as classified by Brandwatch):

Source	(Brandwatch) Sentiment			Total
	negative	neutral	positive	n
blog	13.5	80.0	6.5	934
forum	25.0	67.3	7.7	651
instagram	3.6	54.0	42.5	252
news	7.3	88.0	4.7	10049
reddit	29.5	57.3	13.1	525
review	26.7	46.7	26.7	30
tumblr	19.9	70.2	9.9	1114
twitter	2.7	94.7	2.6	32079
youtube	16.8	66.7	16.5	381

Country breakdowns:

corp$Country2 <- ifelse(corp$Country %in% c("Germany", "United States of America", "United Kingdom"),
                        corp$Country, "Other")

docvars(corp) %>%
    group_by(Country2) %>%
    summarize(n = n()) %>%
    mutate(pct = round(n / sum(n), 1)) %>%
    kbl(col.names = NULL) %>%
    add_header_above(c("Country", "Total", "%"), align = c("l", "r", "r")) %>%
    kable_styling("striped")

Country	Total	%
Germany	592	0.0
Other	30562	0.7
United Kingdom	2468	0.1
United States of America	12393	0.3

Sentiment by country

Sentiment by country:

tab <- docvars(corp) %>%
    select(Country2, Sentiment) %>%
    with(., prop.table(table(Country2, Sentiment), margin = 1) * 100) %>%
    round(1)
cbind(tab, n = table(corp$Sentiment)) %>%
    kbl() %>%
    add_header_above(c("Country", "(Brandwatch) Sentiment" = 3, "Total" = 1), align = c("l", "c", "r")) %>%
    kable_styling("striped")

number of rows of result is not a multiple of vector length (arg 2)

Country	(Brandwatch) Sentiment			Total
	negative	neutral	positive	n
Germany	2.9	88.3	8.8	2350
Other	5.1	91.4	3.4	41887
United Kingdom	4.5	90.4	5.1	1778
United States of America	5.2	90.3	4.4	2350

library("ggplot2")
as.data.frame(tab) %>%
    ggplot(aes(x = Country2, y = Freq, fill = Sentiment)) +
    geom_bar(stat = "identity", position = position_dodge()) +
    scale_fill_brewer(palette = "Paired") +
    xlab("") +
    ylab("Percentage") +
    theme_minimal()

Comparing sentiment on balance:

library("quanteda.tidy")

Attaching package: ‘quanteda.tidy’

The following object is masked from ‘package:stats’:

    filter

net_sentiment <- docvars(corp) %>%
    select(Sentiment, Country2) %>%
    group_by(Sentiment, Country2) %>%
    summarise(n = n()) %>%
    filter(Sentiment != "neutral") %>%
    rename(Country = Country2)

`summarise()` has grouped output by 'Sentiment'. You can override using the `.groups` argument.

library("tidyr")
net_sentiment <- pivot_wider(net_sentiment, names_from = Sentiment, values_from = n) %>%
    mutate(Sentiment = log(positive / negative))

ggplot(net_sentiment, aes(x = Country, y = Sentiment, color = Country, fill = Country)) + 
    geom_bar(stat = "identity") + 
    ylab("Net Sentiment") +
    # scale_colour_brewer(palette = "Paired") +
    theme_minimal()

Keywords by sentiment

library("quanteda.textstats")
library("quanteda.textplots")

toks <- tokens(corp, remove_punct = TRUE)
dfmat <- dfm(toks)

Germany

colls_DE <- tokens_subset(toks, Country == "Germany") %>%
    tokens_remove(c(stopwords("de"), stopwords("en")), padding = TRUE) %>%
    textstat_collocations()
toks_DE <- tokens_subset(toks, Country == "Germany") %>%
    tokens_compound(colls_DE[1:200, ])
keywords_de <- dfm(toks_DE) %>%
    dfm_subset(Sentiment != "neutral" & Country == "Germany") %>%
    dfm_remove(c(stopwords("de"), stopwords("en"))) %>%
    dfm_group(groups = Sentiment) %>%
    textstat_keyness(target = "positive", measure = "lr")

textplot_keyness(keywords_de)

United Kingdom

colls_UK <- tokens_subset(toks, Country == "United Kingdom") %>%
    tokens_remove(stopwords("en"), padding = TRUE) %>%
    textstat_collocations()
toks_UK <- tokens_subset(toks, Country == "United Kingdom") %>%
    tokens_compound(colls_UK[1:200, ])
keywords_uk <- dfm(toks_UK) %>%
    dfm_subset(Sentiment != "neutral" & Country == "United Kingdom") %>%
    dfm_remove(stopwords("en")) %>%
    dfm_group(groups = Sentiment) %>%
    textstat_keyness(target = "positive", measure = "lr")

textplot_keyness(keywords_uk)

United States