Mini-Hackathon2.md

Mini-Hackathon 2: LDA Topic Modeling

Mariken van der Velden & Kasper Welbers

Mini-Hackathons are performed and submitted in pairs of two. You must hand in your assignment on Canvas the next week before Tuesday Midnight.

Use this RMarkdown template on the canvas page for this mini-hackathon to complete your hackathon. When you are finished, knit the file into a pdf with the knit button in the toolbar (or using Ctrl+Shift+K). For this you need to have the knitr and printr packages installed, and all your code needs to work (see the R course companion for more instructions). If you cannot knit the .Rmd file, there is probably an error in your R code, therefore add eval=FALSE to the code chunk: {r, eval = FALSE}, so you are still able to knit and upload the file.

This Mini-Hackathon

This mini-hackathon builds upon the tutorial on LDA Models. For this hackathon, you can (and are strongly recommended to) use code from this week’s tutorial as well as provided here. If you aim to conduct additional analyses in R, we of course encourage this. Nevertheless, it is important to not do additional analysis just for the sake of running more code chunks. For that reason, please provide a justification for these additional analyses.

Also, we recommend to use parameters for RMarkdown codeblocks, in particular the cache = TRUE parameter for codeblocks that take long to compute (e.g., downloading data from AmCAT). A brief explanation of some usefull parameters is given in the the tutorial of two weeks ago.

Important to take into account is that this week you laptop will be doing the heavy lifting. Accordingly, in the assignment, you should be carefull not to choose a concept that has too many articles.

Hackathon Challenges

Challenge 1

Choose a data set and run a LDA topic model to inquire the topics of the media agenda. You can either work with the AmCAT data set with newspaper articles on the US presidential candidates in 2020 or the Guardian data set with newspaper articles in politics, economy, society and international sections from 2012 to 2016. To get the data, one can run the code to download the data from quanteda.corpora or download the .csv file with AmCAT data from Canvas. Take a look at respectively the tutorial of two weeks ago or last week (here and here) to see how to obtain the data.

For later challenges in this hackathon, please already add a year variable to the corpus by running the following code if you use Guaridan data:

library(quanteda)

df <- convert(corp, to = "data.frame") %>%
  mutate(year = substr(date, 1, 4))
corp <- corpus(df)

If you use AmCAT data, run the following code to add the variable:

d <-d %>%
  mutate(month = substr(date, 6, 7))

To conduct the analysis using LDA topic models, we have to create a document-term matrix (DTM), for more information about DTM, see the tutorial on basic text analysis with quanteda. Look at the tutorials of last week (here and here) how you can shape the AmCAT data or the Guardian data into a dtm.

After turning the corpus into a dtm, run LDA topic model from a dfm – have a look at the tutorial to see how to do this.

Don’t forget to filter the words, otherwise the procedure of running an LDA might take forever, or even makes your computer crash.

I will run a LDA topic model with 10 topics, and set the seed to the date of the tutorial.

This can take some time if you do not have a powerful computer, don’t forget cache=T therefore.

Challenge 2

Validation is very important when applying topic models to measure any agenda. Inspect the results of your LDA model using terms, and make a wordcloud for at least 3 topics. Give an interpretation for each topic elaborated with an example. Have a look at the tutorial for the code.

library(wordcloud)

terms(m_G, 5)

Topic 1	Topic 2	Topic 3	Topic 4	Topic 5	Topic 6	Topic 7	Topic 8	Topic 9	Topic 10
said	said	people	one	block-time	said	said	party	trump	said
police	people	said	can	published-time	eu	us	said	said	year
court	one	health	says	gmt	government	government	labour	clinton	$
told	two	can	like	bst	minister	security	government	new	business

topic <- 5
words_G <- posterior(m_G)$terms[topic, ]
topwords_G <- head(sort(words_G, decreasing = T), n=50)
wordcloud(names(topwords_G), topwords_G)

terms(m_A, 5)

Topic 1	Topic 2	Topic 3	Topic 4	Topic 5	Topic 6	Topic 7	Topic 8	Topic 9	Topic 10
like	court	mr	said	said	$	mr	said	de	bst
one	senate	biden	united	police	said	trump	coronavirus	la	coronavirus
people	said	trump	china	black	companies	president	health	que	�
can	republican	said	mr	people	new	said	new	en	cases
just	election	campaign	states	protests	economic	house	virus	el	updated

topic <- 5
words_A <- posterior(m_A)$terms[topic, ]
topwords_A <- head(sort(words_A, decreasing = T), n=50)
wordcloud(names(topwords_A), topwords_A)

Finally, see in which years certain topics were more prevelent.

Challenge 3

Reflect on the validity of the current analysis: to what extent do the results of the LDA model measure the media agenda?

Challenge 4

The method used in the tutorial is pretty crude. Can you think of any problem in particular, and any ways to mend them? Are there problems which cannot be solved, or would be very hard to solve?