TopicClassifier

[ZanSara]

Classifies the input text/document into one in a list of arbitrary categories. It’s very similar to the LanguageClassifier, but it’s more generic as it can classify by arbitrary categories. It can be implemented with a zero-shot text classifier. Again, we may need a text and a Document version.

Draft I/O for TextTopicClassifier:

@component
class TextTopicClassifier:
    
    def __init__(self, topics: Optional[List[str]] = None):
        self.topics = topics
        component.set_output_types(**{topic: List[str] for topic in topics})

    def run(self, strings: List[str]):
        # classify the strings
        return {"topic_1": strings_1, "topic_2": strings_2, ...}

Draft I/O for DocumentTopicClassifier:

@component
class DocumentTopicClassifier:
    
    def __init__(self, topics: Optional[List[str]] = None):
        self.topics = topics
        component.set_output_types(**{topic: List[str] for topic in topics})

    def run(self, documents: List[Document]):
        # classify the documents
        return {"topic_1": docs_1, "topic_2": docs_2, ...}

Tasks

Beta Give feedback

1. TextTopicClassifier
2. DocumentTopicClassifier

[Timoeller]

We want to turn https://prompthub.deepset.ai/?prompt=deepset%2Ftopic-classification into a more explicit component.