Lesson 2. What is content analysis?

From: Privy Council Office

Introduction

This lesson will look at content analysis as a possible solution to the challenge of analyzing large amounts of consultation data. Content analysis is a qualitative research technique used to find patterns in communication in a systematic way.

Many techniques can be used to do content analysis, including:

• word counting
• topic modelling
• sentiment analysis

Together, these techniques let you explore large bodies of text and recognize certain patterns and big pictures. Looking at your data in this way may help you gain new and valuable insights.

Word counting

In content analysis, counting words lets you look for themes and patterns and find interesting keywords to explore. After you remove trivial words, like “the” or “but”, you can gain insight from the most frequently occurring words in your dataset. Look at the text of Lesson 1, for example. The top 3 words are:

1. data (11)
2. consultation (11)
3. qualitative (6)

These words seem to fit, given that the goal of the lesson was to introduce working with qualitative consultation data. Yet top-ranking words can often come as a surprise. Uncovering this information and raising questions about its potential meaning are important parts of content analysis.

Topic modelling

Topic modelling is an exploratory process. It is a way of looking for patterns and themes in text that could guide research. Human beings are able to read and naturally make sense of text. You can see how words are connected and draw conclusions about those relationships. You can make these connections because you know about the world.

Computers don’t have this knowledge. If you give them a list of words, they can’t make the logical leaps that enable you to recognize thematic categories. They only know what you tell them.

Instead of making connections like humans, topic modelling uses statistical models to look for similar “parts” within a dataset. The goal of topic modelling is to locate “parts” that humans can interpret as logical “topics.” You won’t know right away how many different “topics” there are in a given dataset or how many words would fit into each one. The only way to find these things out is to try different options.

Sentiment analysis

Sentiment analysis identifies the emotional content of words. At its most basic, sentiment analysis counts negative and positive words, which suggests whether the dataset tends to one or the other. Assigning a negative or positive label to words is based on human interpretation, so the results are subjective. To conduct this analysis, a list of positive and negative words must be given to the computer.

Sentiment analysis is a developing field. More nuanced tools are available, especially when it comes to understanding text from social media and “big data”.

Conclusion

These are some of the techniques you can use to conduct your content analysis. In the lessons that follow, you will see how to use them and how they fit into the policy development process.