library(tidyverse)
library(tidytext)
library(broom)
library(reshape2)
library(gutenbergr)
library(wordcloud)
library(scriptuRs)
options(repr.plot.width=15, repr.plot.height=10)

── Attaching packages ─────────────────────────────────────── tidyverse 1.3.1 ──

✔ ggplot2 3.3.5     ✔ purrr   0.3.4
✔ tibble  3.1.3     ✔ dplyr   1.0.7
✔ tidyr   1.1.3     ✔ stringr 1.4.0
✔ readr   2.0.0     ✔ forcats 0.5.1

── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag()    masks stats::lag()


Attaching package: ‘reshape2’


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

    smiths


Loading required package: RColorBrewer

The Holy Bible sentiment analysis

We first make a more detailed analysis on the Holy Bible, which is the most read book in the world.
The version used here is the King James Version.

bible <- kjv_bible() %>%
  select(volume_title, text) %>%
  mutate(book = "The Holy Bible") %>%
  relocate(book, volume_title, text)

Creating a tidy table listing all words in the text:

words <- bible %>%
  mutate(text = str_to_lower(text)) %>%
  mutate(text = str_replace_all(text, "'s", "")) %>%
  unnest_tokens(word, text)

Removing stop words, blank lines, numbers and punctuation from the text:

pattern <- c('thou', 'thy', 'ye', 'thee', 'hath')
filtered_words = words %>% 
  filter(!word %in% c(stop_words$word, pattern))%>%
  filter(word != "") %>%
  filter(!str_detect(word, "^\\d+$")) %>%
  filter(!str_detect(word, "[[:punct:]]"))

Arranging the words that are more frequent in absolute frequency:

filtered_words %>%
  count(word, volume_title) %>%
  spread(volume_title, n, fill = 0) %>%
  mutate(total = `Old Testament` + `New Testament`) %>%
  relocate(word, `Old Testament`, `New Testament`, total) %>%
  arrange(desc(total)) %>%
  head(10)

A tibble: 10 × 4

word	Old Testament	New Testament	total
<chr>	<dbl>	<dbl>	<dbl>
lord	7236	728	7964
god	3100	1372	4472
israel	2501	75	2576
king	2451	89	2540
son	1970	422	2392
people	1910	233	2143
house	1817	206	2023
children	1660	160	1820
day	1420	319	1739
land	1667	50	1717

Calculating the odds ratio of words appearing in Old Testament or New Testament, with correction of plus 0.5:

odds_ratio <- filtered_words %>%
  count(word, volume_title) %>%
  spread(volume_title, n, fill = 0) %>%
  relocate(word, `Old Testament`, `New Testament`) %>%
  mutate(or = (`Old Testament` + 0.5) / (sum(`Old Testament`) - `Old Testament` + 0.5) /
           ( (`New Testament` + 0.5) / (sum(`New Testament`) - `New Testament` + 0.5)))

Showing the words that appears with higher frequency in Old Testament and New Testament, each:

odds_ratio %>% arrange(desc(or)) %>% head(10)

A tibble: 10 × 4

word	Old Testament	New Testament	or
<chr>	<dbl>	<dbl>	<dbl>
hosts	299	0	158.08979
judah	825	1	145.62472
philistines	254	0	134.30660
beth	225	0	118.98531
joshua	216	0	114.23132
families	174	0	92.05171
moab	174	0	92.05171
zion	153	0	80.96538
jeremiah	147	0	77.79828
manasseh	147	0	77.79828

odds_ratio %>% arrange(or) %>% head(10)

A tibble: 10 × 4

word	Old Testament	New Testament	or
<chr>	<dbl>	<dbl>	<dbl>
jesus	0	984	0.0001313515
christ	0	571	0.0002280728
paul	0	162	0.0008083798
peter	0	162	0.0008083798
john	0	133	0.0009845237
gospel	0	104	0.0012584316
pharisees	0	87	0.0015034105
disciples	1	243	0.0016159434
church	0	77	0.0016977202
simon	0	76	0.0017199451

Since we want to consider a larger variety of sentiments, it is used the nrc lexicon for sentiment analysis:

nrc <- get_sentiments("nrc") %>%
  select(word, sentiment)

Calculating here the quantitative analysis of sentiments, in absolute numbers:

sentiment_counts <- filtered_words %>%
  left_join(nrc, by = "word") %>%
  count(volume_title, sentiment) %>%
  spread(volume_title, n) %>%
  relocate(sentiment, `Old Testament`, `New Testament`) %>%
  mutate(sentiment = replace_na(sentiment, replace = "none"))
sentiment_counts %>%
  arrange(desc(`Old Testament` + `New Testament`))

A tibble: 11 × 3

sentiment	Old Testament	New Testament
<chr>	<int>	<int>
none	142225	35527
positive	34336	10101
trust	25080	7893
negative	24089	5433
fear	13514	4214
joy	11422	4777
anticipation	11157	4885
disgust	13220	2528
sadness	8525	2603
anger	8134	2140
surprise	4004	1389

It is important to note that the Old Testament contains much more words than New Testament, as follows:

words %>% group_by(volume_title) %>% summarize(n = n())

A tibble: 2 × 2

volume_title	n
<chr>	<int>
New Testament	180392
Old Testament	610017

Due to this difference, the odds ratio of each sentiment is calculated.
An odds ratio bigger than 1 represents more appearance in the Old Testament:

sentiment_counts %>%
  mutate(`Old Testament` = `Old Testament` / (sum(`Old Testament`) - `Old Testament`),
         `New Testament` = `New Testament` / (sum(`New Testament`) - `New Testament`),
         or = `Old Testament`/`New Testament`) %>%
  arrange(desc(or))

A tibble: 11 × 4

sentiment	Old Testament	New Testament	or
<chr>	<dbl>	<dbl>	<dbl>
disgust	0.04679878	0.03201540	1.4617584
negative	0.08868738	0.07143327	1.2415417
none	0.92666193	0.77294781	1.1988674
anger	0.02828509	0.02696912	1.0487953
positive	0.13136932	0.14149239	0.9284551
sadness	0.02968511	0.03299656	0.8996425
fear	0.04788938	0.05453181	0.8781917
trust	0.09267402	0.10724622	0.8641239
surprise	0.01372634	0.01734061	0.7915719
joy	0.04017813	0.06227106	0.6452135
anticipation	0.03920942	0.06376868	0.6148695

The purpose is to show which sentiments are more expressed in each testament, compared to the other one.
The Old Testament has more use of: disgust, negative and anger sentiments
The New Testament have more expression of: anticipation, joy and surprise

For each sentiment, the odds ratio and confidence interval is calculated:

log_or <- sentiment_counts %>%
  mutate( log_or = log( (`Old Testament` / (sum(`Old Testament`) - `Old Testament`)) / (`New Testament` / (sum(`New Testament`) - `New Testament`))),
          se = sqrt( 1/`Old Testament` + 1/(sum(`Old Testament`) - `Old Testament`) + 1/`New Testament` + 1/(sum(`New Testament`) - `New Testament`)),
          conf.low = log_or - qnorm(0.975)*se,
          conf.high = log_or + qnorm(0.975)*se) %>%
  arrange(desc(log_or))
log_or

A tibble: 11 × 7

sentiment	Old Testament	New Testament	log_or	se	conf.low	conf.high
<chr>	<int>	<int>	<dbl>	<dbl>	<dbl>	<dbl>
disgust	13220	2528	0.37964009	0.022077520	0.3363689457	0.42291124
negative	24089	5433	0.21635391	0.015569290	0.1858386634	0.24686916
none	142225	35527	0.18137728	0.007965602	0.1657649912	0.19698958
anger	8134	2140	0.04764216	0.024623368	-0.0006187525	0.09590308
positive	34336	10101	-0.07423329	0.012081300	-0.0979122070	-0.05055438
sadness	8525	2603	-0.10575783	0.022751542	-0.1503500353	-0.06116563
fear	13514	4214	-0.12989039	0.018104858	-0.1653752610	-0.09440552
trust	25080	7893	-0.14603910	0.013559115	-0.1726144815	-0.11946373
surprise	4004	1389	-0.23373460	0.031394347	-0.2952663874	-0.17220281
joy	11422	4777	-0.43817397	0.017704235	-0.4728736311	-0.40347430
anticipation	11157	4885	-0.48634530	0.017632539	-0.5209044452	-0.45178616

The analysis performed calculates the log of the odd ratios, so it is easier to notice which sentiments are more expressed in each testament: positive numbers are more frequent in the Old Testament.
A graphical visualization shows how all of these sentiments are overrepresented in one or another testament:

log_or %>%
  mutate(sentiment = reorder(sentiment, log_or),) %>%
  ggplot(aes(x = sentiment, ymin = conf.low, ymax = conf.high)) +
  geom_errorbar() +
  geom_point(aes(sentiment, log_or)) +
  ylab("Log odds ratio for association between Old Testament and sentiment") +
  theme(axis.text = element_text(size=14)) +
  coord_flip()

It is possible to visualize which words are driving more extreme sentiments like disgust or joy:

odds_ratio %>% inner_join(nrc, by = "word") %>%
  filter(sentiment == "disgust" & `Old Testament` + `New Testament` > 10) %>%
  arrange(desc(or)) %>% head(10)

A tibble: 10 × 5

word	Old Testament	New Testament	or	sentiment
<chr>	<dbl>	<dbl>	<dbl>	<chr>
fat	130	0	68.825870	disgust
pestilence	47	0	25.041208	disgust
burnt	361	5	17.350655	disgust
oppression	24	0	12.914512	disgust
slayer	19	0	10.278642	disgust
hanging	18	0	9.751483	disgust
terrible	51	1	9.049990	disgust
boil	16	0	8.697182	disgust
derision	15	0	8.170040	disgust
prolong	14	0	7.642902	disgust

odds_ratio %>% inner_join(nrc, by = "word") %>%
  filter(sentiment == "joy" & `Old Testament` + `New Testament` > 10) %>%
  arrange(or) %>% head(10)

A tibble: 10 × 5

word	Old Testament	New Testament	or	sentiment
<chr>	<dbl>	<dbl>	<dbl>	<chr>
church	0	77	0.001697720	joy
faith	2	245	0.002671210	joy
charity	0	28	0.004620887	joy
heavenly	0	23	0.005604585	joy
marriage	2	17	0.037635424	joy
star	2	13	0.048790348	joy
promise	8	45	0.049190981	joy
salute	7	32	0.060779964	joy
ministry	4	18	0.064081365	joy
marry	5	17	0.082799169	joy

The 5 words most used to express each sentiment, for each testament, can be plotted in a graph.
Here blue symbolizes New Testament and red symbolizes Old Testament:

odds_ratio %>%
  inner_join(nrc, by = "word") %>%
  mutate(sentiment = factor(sentiment, levels = log_or$sentiment)) %>%
  mutate(log_or = log(or)) %>%
  filter(`Old Testament` + `New Testament` > 10 & abs(log_or)>1) %>%
  mutate(word = reorder(word, log_or)) %>%
  arrange(log_or) %>%
  group_by(sentiment) %>%
  filter(row_number()<=5 | row_number()>=n()-4) %>%
  ungroup() %>%
  ggplot(aes(word, log_or, fill = log_or < 0)) +
  facet_wrap(~sentiment, scales = "free_x", nrow = 2) + 
  geom_bar(stat="identity", show.legend = FALSE) +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))