In this section, it is explained how to revise your topics by Explorer according to your assumptions and needs.
Adding and Removing Topics
You might be interested in specific topics which are not frequent enough to appear in the list of topics. You can add these topics to your project manually and Explorer will analyze your data for them with a semantic search.
You can add topics by clicking on the Add Topic button in the top of the working panel. When you click on this button, Explorer considers a new topic area for it at the top of the list. Here you can give a name to your topic, and you can add terms to your topic by writing them in the text box and clicking on Add words. When you are done, click on Remove me after adding words.
Note that the name of the topics are only labels that are assigned to them, and therefore each topic should have at least one term. Terms are the exact words the Explorer actually keeps track of in your data. You cannot add a term which is already included in another topic. The new topics are automatically pinned, so Explorer keeps them in the list regardless of their frequency.
Keep in mind that Explorer does not add these new topics to your analysis until you have pressed the 'Explore' button.
When you add new topics to your project and press Explore, Explorer starts re-analyzing data and it inserts new topics into the list of topics in order of their frequency. Then, you can see the information about related topics, suggestions and statistics for new topics.
Note that Explorer does not remove any topics from the list when you add new ones.
You can also remove a topic from your analysis if you find it unimportant for your project. Hover over that topic in the working panel and click on cross button to remove the topic.
Explorer will ignore the topic when you remove it from the list, however, note that texts including that topic are still available for inclusion in other topics. The terms including in each topic are saved in the the Ignored Terms section, and you can have them back at any point by clicking on them.
Grouping and Merging Topics
When you hover mouse over a topic in the working panel, an upwards arrow appears inside the topic box which is called Select button. You select a topic by clicking on this button. When the topic is selected, the topic blue box turns to orange. You can also deselect the topic by re-clicking on this button.
In Explorer, both merging and grouping should start with selecting a topic. In the following it is explained how we can merge and group topics by Explorer.
If you find some topics to be in the same category you can put them in one group. This will give you a more organized working panel. In addition, Explorer will tell you about the sentiment and statistic of the whole group.
For instance, for the hotel reviews, you might like to put all the words about food (e.g. breakfast, coffee, wine, etc.) in one group called Food. Explorer shows you the number of the respondents that have written about food (frequency of the whole group). It also shows you the aggregate sentiments of this group, so it wil let you know if the respondents are positive or negative about food on the whole.
How To Group
Grouping topics in Explorer starts with selecting one single topic by clicking on Select button. When you click on this button, a down arrow button will appear in the left side of each topic or group. We refer to this button as Group button. You can place the selected topic into a target group by clicking on this button. By default, the name of the group will be the name of the target topic (or group). You can change this name by re-writing it in the working panel.
The new groups are automatically pinned, so Explorer keeps them in the list regardless of their frequency.
After re-exploring data by Explorer, grouped topics are inserted into the list of the topics in order of their aggregate frequency (which is the number of distinct texts that include at least one of the topics divided by the whole number of topics). This value can be found in both working and detail panel under the group name. In the detail panel and under each group name, you can also find the aggregate sentiment of that group. The frequencies, sentiments, related topics and suggestions for single topics can be found under their names like before.
How to Ungroup
You might want to take a topic out of a group. You can click Undo button at the top or click the ungroup button. Note that the topic you ungroup from a group will be automatically pinned. You should update and save after.
In case that you find two topics equivalent you can merge them to one single topic. This is mostly the case if the terms in two topics are synonyms or paradigmatic neighbors. For instance, for two topics income and compensation including the terms income and compensation successively, you might merge them to get a single topic including both the terms income and compensation. Merging topics increases the strength of the resulting topic in terms of the number of texts that are included in the topic and therefore it results in preciser sentiments and more informative related topics.
How to merge
Similar to grouping topics, merging topics starts with selecting a single topic. As before, you select the topic by clicking on the Select button in the topic box. When the topic is selected, an upwards arrow button will appear at the left side of all other topics. We call this button as Merge button. You merge the two topics by clicking on the Merge button for the second topic. By default, the name of the merged topic will be the name of the first selected topic. You can change this name by re-writing it in the topic box.
If you choose to you enable Coverage, you can receive details about how much of your data is covered by the topics in your model. This can be useful to help you understand how complete your analysis is.
When the setting is enabled, an extra topic is featured at the end of your topic list. It is always labelled 'Unclassified'. It shows the number of texts which are not included under your current topic analysis, both as a percentage and as an absolute number. The magnitude of this figure often depends on the characteristics of an individual dataset. It is also possible to get examples of the texts which are unclassified and get a sentiment score for this special topic (in the usual manner). This should aid with an understanding of what data it is which is not currently being covered.