06-week6demo.Rmd

# Week 6 Demo

## Setup
First, we'll load the packages we'll be using in this week's brief demo. 

```{r, message = F, warning = F}
library(topicmodels)
library(dplyr)
library(tidytext)
library(ggplot2)
library(ggthemes)
```

Estimating a topic model requires us first to have our data in the form of a document-term-matrix. This is another term for what we have referred to in previous weeks as a document-feature-matrix.

We can take some example data from the `topicmodels` package. This text is from news releases by the Associated Press. It consists of around 2,200 articles (documents) and over 10,000 terms (words).

```{r}

data("AssociatedPress", 
     package = "topicmodels")

```

To estimate the topic model we need only to specify the document-term-matrix we are using, and the number (`k`) of topics that we are estimating. To speed up estimation, we are here only estimating it on 100 articles.

```{r}

lda_output <- LDA(AssociatedPress[1:100,], k = 10)

```

We can then inspect the contents of each topic as follows.

```{r}
terms(lda_output, 10)
```

We can then use the `tidy()` function from `tidytext` to gather the relevant parameters we've estimated. To get the $\beta$ per-topic-per-word probabilities (i.e., the probability that the given term belongs to a given topic) we can do the following.

```{r}

lda_beta <- tidy(lda_output, matrix = "beta")

lda_beta %>%
  arrange(-beta)

```

And to get the $\gamma$ per-document-per-topic probabilities (i.e., the probability that a given document (here: article) belongs to a particular topic) we do the following.


```{r}

lda_gamma <- tidy(lda_output, matrix = "gamma")

lda_gamma %>%
  arrange(-gamma)


```

And we can easily plot our $\beta$ estimates as follows.

```{r}

lda_beta %>%
  group_by(topic) %>%
  top_n(10, beta) %>%
  ungroup() %>%
  arrange(topic, -beta) %>%
  mutate(term = reorder_within(term, beta, topic)) %>%
  ggplot(aes(beta, term, fill = factor(topic))) +
  geom_col(show.legend = FALSE) +
  facet_wrap(~ topic, scales = "free", ncol = 4) +
  scale_y_reordered() +
  theme_tufte(base_family = "Helvetica")

```

Which shows us the words associated with each topic, and the size of the associated $\beta$ coefficient.