Topics group related insights around a common theme and show how that theme connects to the records in your CRM. When you open a topic, Alchemer Voice displays every CRM record tied to the insights in that topic. From the same screen, you can filter the insights further so that you see only the ones tied to a specific CRM scenario.
Creating topics with a topic model
A topic model groups your insights by how semantically similar they are. When you create a model, Alchemer Voice automatically discovers and generates topics for your insights. You can then edit those topics directly, or tune the model settings and regenerate the topics.
Topic model settings
Two settings control how many topics a model produces and how large each topic is.
| Setting | What it does |
|---|---|
| Minimum Topic Size | Sets the smallest number of insights that can form a topic, which in turn shapes how many topics are generated. A higher value produces fewer, larger topics; a lower value produces more small, micro topics. We recommend increasing this value rather than decreasing it. |
| Maximum number of topics | Sets the largest number of topics the model will keep after training. Topics are merged together to reach this target. Setting the number too low can force unrelated topics to combine. |
How do I reduce outliers?
The number of insights classified as outliers is controlled by the topic density setting. Lowering the density reduces outliers by creating more topics.
Keep in mind that lowering the density can push outliers into topics where they do not truly belong. Aim for a balance between reducing outliers and keeping unrelated insights out of your topics.
I have only a few topics. How do I increase them?
If you have a small number of insights, it can be hard to extract meaningful topics because there is not much data to work with. Filtering to include more insights usually gives better results.
Your Minimum Topic Size may also be too large for the number of insights you have. Lowering the minimum size makes it much more likely that additional topics are generated. Lowering the topic density has a similar effect.
Finally, and less commonly, there simply may not be many distinct topics to find in your insights. A large number of outliers is often a sign of this.
I have too many topics. How do I reduce them?
If you have a large number of insights, the model can generate a very large number of topics. There are a few ways to bring that number down:
- Set the Minimum Topic Size higher relative to the number of insights, so small topics are not generated. This modeling parameter sets the minimum number of insights a topic needs, and larger topics mean fewer topics overall.
- Set the topic density to a higher level. This prevents micro topics, because the model then needs many neighboring insights before it will create a topic.
- Set the maximum number of topics to a value that makes sense for the volume of insights you are working with. Because topics are forced to merge to hit this target, it can lower the quality of the resulting topics.
In practice, we recommend leaving the number of topics on Auto. Auto merges topics that are very similar while keeping dissimilar topics separate.
