Themes
This version of themes is still fully functional, and tag pools can still be configured to use legacy themes.
Themes with GenAI support all reporting modules and workflows supported by legacy themes. For more information, see Themes with generative AI.
When Topic Finder and Theme Explorer are enabled for the same data, the themes in Topic Finder and Theme Explorer are identical.
Rule-based topics provide you with precise control over the comments the system captures, but when there are millions of comments in the system, it’s difficult to gauge whether your topics are capturing the most important insights.
Themes can supplement your rule-based topics by exposing the issues that you did not know needed topics, allowing you to fill the gaps in your topic list. Also, since themes are generated by the theme discovery engine (rather than from a template or by a person), they represent an unbiased view of the issues customers are discussing.
| Topics | Themes |
|---|---|
|
|
Themes in Topic Finder and Theme Explorer
If you find an interesting theme in Theme Explorer, you can track it by searching for the theme in Topic Finder, then adding it to your topic list. After you add a theme to your topic list, the theme you were interested in appears as a topic in your rule-based Text Analytics reports.
In the image below, the list of top themes in Topic Finder and Theme Explorer are almost identical. The reason the lists are not completely identical is because they are ranked using different criteria. Whereas themes in Theme Explorer are ranked according to the total number of records with captured phrases, themes in Topic Finder are ranked according to the number of untagged phrases they capture.
The theme discovery process
The theme discovery engine uses machine learning techniques (such as unsupervised clustering) to generate themes. During theme discovery, the system performs the following actions:
- Processes text at the phrase level.
The parser analyzes the raw text (e.g. customer comments) at the phrase level to identify parts-of-speech and the root form of each word. For example, each of the phrases in this comment would be analyzed separately:
- Creates concept clusters of words that frequently appear together (words that are conceptually related).
For example, friendly, staff, salesperson, and rude might constitute a concept cluster because they are used similarly in the text.
Although concept clusters are not visible, they form the underlying foundation for the themes the system creates.
Parts-of-speech and synonyms do not influence the process of grouping words into concept clusters. Also, the title of the cluster is typically the most frequently-occurring noun people use when discussing the concept.
In the Phrases panel, Theme Explorer highlights all the words that are part of an underlying concept cluster in bold when you select a theme:
- Creates up to 5,000 themes from the concept clusters.
The top ten themes (the themes that apply to the most records) are displayed in the initial Bubble view. The List view displays the full list of themes.
When you click a theme, the system reveals new level of related themes. In the following image of the List view in Theme Explorer, Year, Channel, Price, Internet, and Ad are the themes with the greatest number of records that are also related to Program, so these themes appear under the Program theme.
Theme Explorer displays up to 10 levels of themes. The number of levels varies depending on the number of records available.
- Adds emerging themes to the system.
At the beginning of each month (typically on day 3), Text Analytics analyzes the records from the previous month and adds any significant new themes. You can have up to 6,000 themes.
