Models are the part of Tekst that reads each incoming message and turns it into something your automations can act on. Whenever an email, ticket, or case arrives, your models look at the content and produce a result - a category, a set of extracted fields, or a matched record. This article explains what a model is, the types of models available, and how they fit together so you can decide which one you need.
What a model does
A model takes the text of a message (and the text from any supported attachments) and produces a structured prediction. That prediction is then available to the rest of the platform: it can drive an automation, enrich a record in a connected system, or feed your analytics.
You do not write rules or code to build a model. You describe what you want it to do - the categories it should choose from, the fields it should pull out, or the records it should match against - and Tekst handles the rest. Models are fully managed: Tekst monitors their performance and keeps them improving over time based on your feedback.
For more on what Tekst keeps and how the feedback loop works, see What data does Tekst store and how does the learning process work?.
The three types of models
Tekst offers three types of models. Each one answers a different question about an incoming message. You choose the type when you create a new model.
Classification models
A classification model assigns one or more predefined labels to a message based on its content. This is the right choice for jobs like intent detection ("is this a complaint, a question, or a request?") or topic tagging.
Labels can be organized in a hierarchy, and a model can be set up to pick a single label or several at once. To learn more, see the "What is a classification model?" article.
Extraction models
An extraction model pulls structured data fields (called entities) out of a message or its attachments - for example an invoice number, an order date, or a customer name. The result is a clean, structured object you can pass into another system.
To learn more, see the "What is an extraction model?" article.
Matching models
A matching model compares an incoming message against your own master data records and returns the single best match - for example matching a free-text product description to the correct product code in your catalog.
To learn more, see the "What is a matching model?" article.
How to choose
Use this quick guide to pick the right type:
- Choose classification when you need to sort messages into known categories.
- Choose extraction when you need to pull specific values out of the text.
- Choose matching when you need to connect a message to a record in your own data.
You can run several models at once, and many customers combine them - for example classifying a ticket and extracting key fields from the same message.
Where to find your models
All of your models live in the Models section of the Tekst platform. From there you can open any model to configure it, review its accuracy, and see its status.
How models are kept accurate
Every model is evaluated against a test set of validated examples so you can see how well it is performing, and each type reports accuracy in a way that suits the job it does. Tekst continuously monitors performance and retrains models as needed.
To go deeper on these topics, see the following articles:
0 comments
Please sign in to leave a comment.