afd_topics.Rmd

---
title: "AfD Topics"
author: "Anton Könneke"
output: github_document
---

```{r setup, include=FALSE}
pacman::p_load(quanteda,
               readtext,
               seededlda,
               tidyverse,
               quanteda.textstats,
               quanteda.textplots,
               LSX)
afdt <- readtext(file = "texts/afd_articles.csv")
```

Format dates
```{r}
Sys.setlocale(locale = "de_DE")
afdt <- afdt %>% 
  mutate(day = str_extract(date, "^\\d+"),
         mon = str_match(date, "(?<=\\d{1}\\.\\s)\\w*")[,1],
         year = str_extract(date, "\\d{4}$"),
         date_fmt = paste(year, mon, day, sep =  "-"),
         date_fmt2 = as.Date(date_fmt, format = "%Y-%B-%e")) %>% 
  filter(!is.na(fulltext))
Sys.setlocale(locale = "en_US")
```

prepare data
```{r}
afdt_corp <- corpus(afdt, text_field = "fulltext")
afdt_tok <- tokens(afdt_corp, remove_punct = TRUE, remove_numbers = TRUE, remove_symbol = TRUE)
afdt_tok <- tokens_remove(afdt_tok, pattern = c(stopwords("de")))
afdt_tok <- tokens_remove(
  afdt_tok,
  pattern =
    as.character(tokens(
      paste0(unique(afdt$author), collapse = " "),
      remove_punct = TRUE,
      remove_numbers = TRUE,
      remove_symbol = TRUE
    ))
)
afdt_dfmat <- dfm(afdt_tok) %>% 
              dfm_trim(min_termfreq = 0.8, termfreq_type = "quantile",
                       max_docfreq = 0.1, docfreq_type = "prop")
```

topic model
```{r}
ldas <- lapply(seq(5, 20, 5), function(x){
  textmodel_lda(afdt_dfmat, k = x)
})

lapply(ldas, function(x) terms(x, 10))

lda15k <- ldas[[3]]


```

plot
```{r}
bind_cols(afdt,
          as.data.frame(lda15k$theta)) %>% 
  pivot_longer(starts_with("topic")) %>% 
  filter(year >= 2020 & !is.na(date_fmt2)) %>% 
  mutate(month = lubridate::floor_date(date_fmt2, "month"),
         topic_n = as.integer(str_extract(name, "\\d+")),
         facet = case_match(topic_n,
                            c(1:3) ~ 1,
                            c(4:7) ~ 2,
                            c(8:11) ~3,
                            c(11:15)~4,
                            .default = NA)) %>% 
  summarise(value = mean(value, na.rm = T), .by = c(month, name, facet)) %>% 
  arrange(month) %>% 
  mutate(value = zoo::rollmean(value, k = 3, fill = NA, align = "center"),
         .by = name) %>% 
  filter(facet == 1) %>%
  ggplot(aes(month, value, color = name))+
  geom_line()+
  facet_wrap(~facet)+
  theme_bw() 

```