Categorization workflows: Sequential

This approach is about building topics bottom-up, one after the other, starting with the most obvious/visible ones and ending with the outliers and exceptions.

This workflow works with most datasets and is suited whenever you have a good idea of most of the topics - because you are expert in the field or because you have followed the conversation - but are not clear about the right granularity.

Typically, answers coming from an open question in a survey or a team workshop fit well into that approach.

Using these models is a guarantee of high relevance of analysis and good control on the end result (size, # of topics…)

On a broader perspective, this is the most universal and deemed "always valid" approach. 

Selecting the first topics

There are several valid ways to select your first topics, all made easy by SimpleX features

Starting with keywords list

During import, SimpleX has extracted all keywords. By clicking on #Keywords in the bottom left side bar, you display this list of keywords. Sort by decreasing frequency by clicking on the count field label. On the top of the list you will find most frequent keywords. While the first ones are likely to be generic (eg projects, feeling) , you will spot rapidly one specific keyword relevant for a first topic (eg reward for a topic around the feeling of recognition of employees)

Looking at the keyword word cloud

The word cloud is a visual way to get in a holistic and intuitive way what the dataset is all about. A good look to it is a safe way to dientif toe list of the main topics.

Unfolds the Menu left sidebar and select Chart. The word cloud will appear in the dashboard. You can magnify it by clicikng ont he arrow on the top right corner.

Search and display relevant answers

Once the topic you want build identified , it is time to build the topic and feed it with text answers.

In order to so, we need to search and display answers relevant to this topic.

Using keywords synonyms

The simplest and most intuitive approach is to use the search per keyword. Open the #keywords dropdown in the right sidebar, unfold it and find the reward keyword as well as keyword that are synonym (eg recognition).

Find similar…

Another way to search answers relevant to the topic is to use the powerful semantic search feature of SimpleX.

By clicking the initial keyword, you will display all the text answers containing this keyword #reward. Pick one text answer that fits very well to your first topic (eg "feeling of not being rewarded")

Click "Find similar…" and SimpleX displays the 50 most similar answers present in the dataset. It will display the vast majority of the ideas containing the keyword ("little reward") BUT also the answers with the same meaning but not using the word "reward" (eg "If we are not lucky to work on "visible" projects, we don't get any recognition.")

Feeding topics with answers

Once displayed, select all or some of the answers you judge relevant, select "Move to" in the Grouped action dropdown, select new topic… as a destination. Click on Apply.

Enter the label of the topic (eg "Recognition feeling") and click Move.

You just created your first topic and fed it with some semantically relevant answers.

Suggest content

A great way to check whether there are not some more text answers to attribute to a given topic, use the powerful Suggest content… function.

Select in the left sidebar the suggest content for one given topic. It will display the remaining answers most similar to the answers already in the topic. If the first 15 do not seem relevant, it is probably because the topic is complete.

When clicking on suggest content,

for mono-label categorisation, not grouped

for multi-label, select All answers (outside this topic)  

Proceed following the same steps for the second and subsequent topics. You can opt for a mono (one answer can belong to only one topic) or multi label approach (one answer can belong to several topics)

After having defined and fed all your topics, there might still be some unattributed text answers left. It might be the opportunity to create a last topic or to put them in a "Miscellaneous" topic

Sequential bottom up approach is well suited when

• you are dealing with a medium/large dataset > 1-3k
• it is critical for you to build a coding plan that fits perfectly with this specific dataset
• you have a partial view on what the end-result should look like 

Pros:

• Quite intuitive and natural, step-by-step approach,
• Easy to share and explain, suited for collaborative work
• Universal

Cons:

• Workload grows as dataset size grows
• End results depends on who is in charge of analysis