05-group-exercise.qmd

---
format:
  revealjs:
    logo: "https://cms-cdn.lmu.de/assets/img/Logo_LMU.svg"
    footer: "[Advanced Text Analysis - SICSS-Munich 2023 - Valerie Hase](https://github.com/valeriehase)"
    center-title-slide: false
    theme: ["theme/q-theme.scss"]
    highlight-style: atom-one
    code-fold: show
    code-copy: true
    self-contained: true
    number-sections: false
    smaller: true
    progress: true
execute:
  echo: true
bibliography: "references/references.bib"
csl: references/apa.csl
editor: 
  markdown: 
    wrap: 72
---

<h1>Advanced Text Analysis</h1>

<h2>SICSS-Munich, Day 4</h2>

<hr>

Session: Group Exercise

Valerie Hase (LMU Munich)

## Agenda

Today we covered a broad range of techniques for analyzing texts beyond
bag-of-words approaches.

The main goal was to get you to think about ...

-   what kind of social science questions we can answer with these
    methods 🤔

-   how to apply them

-   critically reflect on their limitations ❌

Now, let's apply this knowledge! 🔨

## Group Exercise

-   Get together in 5 groups (by counting from 1 to 5)
-   In the group, decide whether you want to work on Task 1 or Task 2
-   Work as a team: The goal is **not** to produce the most efficient
    code the quickest, but to tackle the task as a team of
    interdisciplinary researchers with varying experience in coding
-   Regardless of which task you choose, produce one visualization that
    describes your main findings 🖼️
-   Be ready to give a three-minute pitch in which you present your
    results at 3.30 PM ⏲️

## Option 1 - Using advanced text analysis for analyzing news coverage

*Task1.RDATA* contains German news coverage, with *N* = 500 texts (see "../data").
    
Either Option 1a: 

-   If we look at the data, we can see that it contains duplicates. As a
    team, **discuss** which of the methods you just learned about you
    could use to identify (close-to) similar texts?
-   As a team, **apply** one of the methods to deduplicate the corpus.

Or... Option 1b:

-   Next, we want to identify sources of direct quotes in the
    corpus using syntactic information. By **applying** (for example) *dependency parsing*, can you
    identify common sources? Who is cited most often?
-   Try to install the `spacyr()` via the following
    [vignette](https://cloud.r-project.org/web/packages/spacyr/vignettes/using_spacyr.html).
    Compare `spacyr` and `udpip`: Does any of the packages do a better
    job (at either dependency parsing or PoS tagging)? How could you benchmark results?

## Option 2 - Extending your knowledge on word embeddings

Load the data frame of TED talks that you already analyzed yesterday (see "../data").
    
Either Option 2a:

-   By **applying** our knowledge on *word embeddings*, can you explain
    and illustrate how speakers talk about a key issue (e.g.,
    technology)?
    
Or... Option 2b:

-   Using the `context()` package (for a vignette, see
    [here](https://cran.r-project.org/web/packages/conText/vignettes/quickstart.html)), 
    can you explain and illustrate
    whether there are any gender differences in how speakers talk about
    the issue you selected? (i.e., when using gender as a covariate)