Label identification task.

[pranauv98]

Given a text(page sized texts) and list of labels(20 -30), I want to classify the text into any one of the label based on the content of text. I have more than 1000+ training examples and very challenging texts are expected in the test set. Which is the best suited way of approach for me? Will IReRa method give me good result? I have already tried COPRO method.

class Label(dspy.Signature):
"""Given a text and list of labels, select a label from the given list that matches the content of the text.
The label must be chosen from the given list of labels only. """
text = dspy.InputField(desc="Text data to be labelled")
labelPool = dspy.InputField(desc="List of label names from which a label has to be chosen")
label = dspy.OutputField(desc="Should be chose from the given list of labels.")

class PredictorModule(dspy.Module):
def init(self):
super().init()
self.signature = Label
self.predictor = dspy.ChainOfThought(self.signature)
def forward(self,text,labelPool):
result = self.predictor(text=text,labelPool=labelPool)
if result.label not in labelPool:
result.label="Other"
print(result)
return dspy.Prediction(
label=result.label,
rationale = result.rationale
)