nlp.R

library(tidyverse)
library(lubridate)
library(tidytext)
library(wordcloud)
library(reshape2)
data("stop_words")

# read in sentiments
sentiments_bing <- get_sentiments("bing")


# read in conference talks from 1971 - 2018
text <- read_csv("conference_talks.csv")

# add date field
text %>% 
    mutate(date = dmy(paste("01", conference))) %>%
    group_by(date) %>%
    summarise(number_talks = n()) %>%
    ggplot(aes(x = date, y = number_talks)) +
        geom_point() +
        theme_bw() +
        labs(x = "", y = "Number of Talks per Conference") +
        geom_smooth()

# count number of words
text %>% 
    mutate(date = dmy(paste("01", conference))) %>%
    group_by(date) %>%
    summarise(number_words = ) %>%
    ggplot(aes(x = date, y = number_talks)) +
    geom_point() +
    theme_bw() +
    labs(x = "", y = "Number of Talks per Conference") +
    geom_smooth()

# wordcloud titles
text %>%
    unnest_tokens(word, title) %>%
    anti_join(stop_words) %>%
    count(word, sort = T) %>%
    with(wordcloud(word, n, max.words = 100))

    # with sentiment
text %>%
    unnest_tokens(word, title) %>%
    anti_join(stop_words) %>%
    count(word, sort = T) %>%
    inner_join(sentiments_bing) %>%
    acast(word ~ sentiment, value.var = "n", fill = 0) %>%
    comparison.cloud(colors = c("gray20", "gray80"),
                     max.words = 100)
# wordcloud speakers


# examine sentiment over time of each talk or each conference
# could look at sentiment by sentence or paragraph