topic-model-isec2022.Rmd

---
title: "Topic modelling of ISEC2022 abstracts"
author: "Olivier Gimenez"
date: '2022-07-04'
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

# Motivation

This is a quick and dirty text mining analysis of the corpus of [ISEC 2022](https://www.isec2022.org/) abstracts (talks and posters) using topic modelling. 

# Clean up data

```{r}
library(tidyverse)
theme_set(theme_light(base_size = 14))
```

Read in text file of all abstracts, and [split on empty line](https://stackoverflow.com/questions/38958597/r-split-text-on-empty-line):
```{r}
allabstracts <- readLines("txt/ISEC-2022-Abstract-Book.txt")
nvec <- length(allabstracts)
breaks <- which(grepl("^[[:space:]]*$", allabstracts))
nbreaks <- length(breaks)
if (breaks[nbreaks] < nvec) {
  breaks <- c(breaks, nvec + 1L)
  nbreaks <- nbreaks + 1L
}
if (nbreaks > 0L) {
  allabstracts <- mapply(function(a,b) paste(allabstracts[a:b], collapse = " "),
                   c(1L, 1L + breaks[-nbreaks]),
                   breaks - 1L)
}
```

Fix encoding:
```{r}
Encoding(allabstracts) <- "latin1"
allabstracts <- iconv(allabstracts, "latin1", "UTF-8",sub='')
abstracts_tbl <- as_tibble(allabstracts)
```

Get rid of empty lines:
```{r}
dat <- abstracts_tbl %>%
  rowwise() %>%
  mutate(val = str_length(value)) %>%
  filter(val > 1)
```

Add a column of id for each abstract:
```{r}
dat$id <- seq(1, nrow(dat))
```

Split columns into tokens:
```{r}
library(tidytext)
tidy_abstracts <- dat %>%
  unnest_tokens(word, value)
```

Get rid of common stop words:
```{r}
data(stop_words)
tidy_abstracts <- tidy_abstracts %>%
  anti_join(stop_words)
```

Create DocumentTermMatrix object:
```{r}
abstracts_dtm <- tidy_abstracts %>%
  group_by(id, word) %>%
  summarize(count = n()) %>%
  mutate(document = id, term = word) %>%
  cast_dtm(document, term, count)
```

# Counting

Count word occurrence:
```{r}
tidy_abstracts %>%
  count(word, sort = TRUE)
```

Visualize:
```{r}
counts <- tidy_abstracts %>%
  count(word, sort = TRUE) %>%
  filter(n > 150) %>%
  mutate(word = reorder(word, n)) %>%
  ggplot(aes(n, word)) +
  geom_col() +
  labs(y = NULL)
counts
```

Save plot:
```{r}
ggsave(filename = "fig/words-isec2022.png",
       plot = counts,
       dpi = 600)
```

# Topic modeling

Estimate LDA model:
```{r}
library(topicmodels)
isec_lda <- LDA(x = abstracts_dtm, 
                k = 10, 
                control = list(seed = 1234))
```

Tidy up coefficients:
```{r}
library(tidytext)
topics <- tidy(isec_lda, matrix = "beta")
```

Get top terms:
```{r}
topterms <- topics %>%
  group_by(topic) %>%
  slice_max(beta, n = 10) %>% 
  ungroup() %>%
  arrange(topic, -beta)
```

Visualize:
```{r}
topicplot <- topterms %>%
  mutate(term = reorder_within(term, beta, topic)) %>%
  ggplot(aes(beta, term, fill = factor(topic))) +
  geom_col(show.legend = FALSE) +
  facet_wrap(~ topic, scales = "free") +
  scale_y_reordered()
topicplot
```

Save plot:
```{r}
ggsave(filename = "fig/topic-isec2022.png",
       plot = topicplot,
       dpi = 600,
       width = 14,
       height = 10)
```