mental_health_model.py

import openai
from rank_bm25 import BM25Okapi

class MentalHealthClassifier:
    def __init__(self, train_data):
        # Tokenize the training data for BM25
        self.tokenized_train = [doc.split() for doc in train_data["text"]]
        self.bm25 = BM25Okapi(self.tokenized_train)
        self.train_data = train_data

    def classify_text(self, api_key, input_text, k=20):
        # Set the OpenAI API key
        openai.api_key = api_key
        if not openai.api_key:
            return "Error: OpenAI API key is not set."

        # Tokenize input text
        tokenized_text = input_text.split()
        # Get top-k similar examples using BM25
        scores = self.bm25.get_scores(tokenized_text)
        top_k_indices = sorted(range(len(scores)), key=lambda i: scores[i], reverse=True)[:k]

        # Build examples for the prompt
        examples = "\n".join(
            f"Example {i+1}:\nText: {self.train_data.iloc[idx]['text']}\nClassification: "
            f"Stress={self.train_data.iloc[idx]['Ground_Truth_Stress']}, "
            f"Anxiety={self.train_data.iloc[idx]['Ground_Truth_Anxiety']}, "
            f"Depression={self.train_data.iloc[idx]['Ground_Truth_Depression']}, "
            f"Other={self.train_data.iloc[idx]['Ground_Truth_Other_binary']}\n"
            for i, idx in enumerate(top_k_indices)
        )

        # Construct OpenAI prompt
        prompt = f"""
        You are a mental health specialist. Analyze the provided text and classify it into one or more of the following categories: Stress, Anxiety, Depression, or Other.

        Respond with a single category that best matches the content: Stress, Anxiety, Depression, or Other.

        Here is the text to classify:
        "{input_text}"

        ### Examples:
        {examples}
        """

        try:
            response = openai.ChatCompletion.create(
                messages=[
                    {"role": "system", "content": "You are a mental health specialist."},
                    {"role": "user", "content": prompt},
                ],
                model="gpt-4",
                temperature=0,
            )
            content = response.choices[0].message.content.strip()
            return content  # Return the label directly
        except Exception as e:
            return f"Error: {e}"