Building topics bottom up

A topic in SimpleX is a group of text answers. It usually revolves around a common theme, a similar meaning but in fact it can be any underlying grouping logic. 

The fastest way to identify the underlying topics is through the InsightBot, using the /discover command. It will in seconds identify up to 10 topics within the text corpus, loosely ranked by importance. 

Yet, sorting actually the different text answers into topics gives access to a synthetic, quantified and more rigorous view of the corpus. Hence, building topics is a powerful way to highlight the key insights from a given dataset. Once this done, powerful charts and tables representing topics and their relative weights are automatically generated within SimpleX Open.

SimpleX Open provides accelerated ways, powered by AI, to cluster answers.

There are several ways to build topics with a given dataset, depending on your situation.

Reading every single text answer and moving it into the relevant topic could prove very painful and time-consuming.

Let’s assume you have some idea of the different topics you want to create, based on your own understanding to the subject, and have at least a view on the most obvious ones.

Let’s take the example of Hotel reviews. You know that some reviews will mention (positively or negatively) Cleanliness and you want to build such a topic

SimpleX offers 3 ways to dramatically accelerate this task:

• with existing answers
• with keywords
• with samples

With existing answers

Say you have picked manually a few relevant answers focused around your Cleanliness issue and moved them into your newly created topic, as explained above.

THEN, click on the ... next to the topic in the left sidebar and click on Suggest content.

SimpleX will suggest relevant new answers from the dataset, based on meaning.

Select those you agree with and save them into the topic.

with keywords

Click on Add a new topic

Select the Keywords radio button

Think of the different keywords that would indicate that a review qualifies for this topic: for example, dirty, clean, hygiene, litter, cleaning.

Indicate your set of keywords and it will automatically create a topic populated with all the text answers containing one of the selected keywords, including verbal forms and plural.

Of course, the more specific the keywords, the more relevant the approach. in any cases, reviewing the topics is advised.

Note that you can only select within the large set of keywords extracted by SimpleX. If you don’t find one specific keyword, you can always process it separately, using the full text research and moving the search results into the topic.

With samples

Click on Add a new topic

Select the Samples radio button

From a UI perspective, the simplest way to add a new topic is to select at least one text answer in the center console, to select in the Grouped Action dropdown the Move to action. Select on the right new topic… and click on Apply.

You will prompt to enter a topic label (ie name). Do so and validate by click on Move and a the new topic will be created , populated with the selection of answers.

Selecting "Move to…" will move the selected answers into the new topic, and out from any existing topic they were previously. Select "Copy to" if you prefer the answers to belong to the new topic and any other topic they were previously.

Once created, the topic will be displayed on the left sidebar. You can move to/copy to any new selection of answers by using Grouped action or via Drag & Drop (NB: select on the bottom bar how you would prefer drag & drop to behave)

Think of the different simple reviews that would qualifie for this topic, regardless whether it does exist or not in the dataset.

For example: the rooms were very dirty, cleaning was below standard, room & bathroom were very clean

SimpleX will automatically suggest relevant new answers from the dataset, based on meaning.

It will include answers mentioning the cleanliness but using different words than your samples or even in different languages.

Select those you agree with and save them into the topic.

Of course, you can repeat this process for the following topics you have in mind, in order to build the different topics of your dataset.

Sometimes, you are not familiar with the subject you need to analyse, you don’t have preconceived idea of the topic structure or you just don’t want to be based by your preconceived ideas.

SimpleX can suggest topics based on the meaning of the answers of the text dataset.

This is meant to serve you as as starting point for discovery and further analysis. You can then edit the topic name, move or copy answers from one topic to another, create additional topics or subtopics.

Select "Suggest topics based on meaning"

It will create automatically 10 to 15 topics covering the entire dataset. It will leave it up to you to fine tune its work by editing the name of each topic, adjusting the content of each topic, editing the structure by merging or splitting some topics.

The Suggest topics algorithm is code to build distinct topics, for clarity reasons.

Remember different people are likely to build different topics with the same dataset. It is the same for AI and sometimes SimpleX will suggest topics that will look consistent with your own views… and sometimes not.

The power of SimpleX lies in its flexibility.

It is highly recommended to combine these different approaches.

One typical combination could be:

• Starting with a few ‘obvious’ topics based on keywords
• Once created, use the suggest content to find other answers that would fit but do not contain the keywords
• After a few topics, use the suggest topics on the remaining set of answers, to discover new important themes.
• Review the newly suggest topics, adjust labels and content.