How to set up your classification task
- Felix Melchner
- Jiawen Wu
With a proper set of labels describing the classification task, the model can infer by semantic similarity how to categorize a text without having been explicitly trained for that task. For example, a sentiment analysis task could be formatted as follows:
To categorize the sentence The food was delicious! into a positive or a negative label, one would define two corresponding hypotheses such as, This person expresses a positive sentiment and This person expresses a negative sentiment. The model will infer a stronger semantic relationship with the positive tag.
For the ease of use, we can create pattern templates with a curly-bracket placeholder for replacing the corresponding label each time. In the example above, it would have been This person expresses a {} sentiment.
Parameters
Parameter | Description |
|---|---|
model | The language model to use. Call GET /zero-shot/ for the models available. Choose symanto_brain_multilingual only if there is no specific language model for your selected languauge. |
all | If true only returns the most probable label. Otherwise it returns all of them. |
multi_label | If false, returned probability distribution among provided labels for each instance will sum up to 1. Otherwise it will be unnormalized. |
Examples
Please head over to Use Case Examples to view some common use cases and their configurations.
Symanto Brain Knowledge Base 2022