cmbtest/trendtopics

1import gradio as gr2import pandas as pd3import requests4import os5import time6from collections import Counter7from io import StringIO8 9# =====================================================10# CONFIGURAÇÃO11# =====================================================12MODEL_ID = "dslim/bert-base-NER"13API_URL = f"https://router.huggingface.co/{MODEL_ID}"14 15HF_API_TOKEN = os.getenv("HF_API_TOKEN")16if not HF_API_TOKEN:17    raise RuntimeError("HF_API_TOKEN não encontrado. Configure em Settings → Secrets.")18 19HEADERS = {20    "Authorization": f"Bearer {HF_API_TOKEN}",21    "Content-Type": "application/json"22}23 24# =====================================================25# FUNÇÃO PRINCIPAL26# =====================================================27def analyze_abstracts(csv_text):28 29    # ---------- Ler CSV ----------30    try:31        df = pd.read_csv(StringIO(csv_text))32    except Exception as e:33        return pd.DataFrame(34            [["Erro ao ler CSV", str(e), ""]],35            columns=["Term", "Frequency", "Hot"]36        )37 38    if "abstract" not in df.columns:39        return pd.DataFrame(40            [["Erro", "Coluna 'abstract' não encontrada", ""]],41            columns=["Term", "Frequency", "Hot"]42        )43 44    all_terms = []45 46    # ---------- Processar abstracts ----------47    for abstract in df["abstract"].dropna():48 49        payload = {50            "inputs": abstract,51            "parameters": {52                "aggregation_strategy": "simple"53            },54            "options": {55                "wait_for_model": True56            }57        }58 59        try:60            response = requests.post(API_URL, headers=HEADERS, json=payload, timeout=60)61            result = response.json()62 63            # Caso modelo ainda esteja carregando64            if isinstance(result, dict) and "error" in result:65                print("Aviso da API:", result)66                continue67 68            # Caso correto69            if isinstance(result, list):70                for ent in result:71                    term = ent.get("word", "").strip().lower()72                    if len(term) > 2:73                        all_terms.append(term)74 75        except Exception as e:76            print("Erro na requisição:", e)77            continue78 79    if not all_terms:80        return pd.DataFrame(81            [["Nenhum termo médico encontrado (verifique logs da API)", "", ""]],82            columns=["Term", "Frequency", "Hot"]83        )84 85    # ---------- Frequência ----------86    term_counts = Counter(all_terms)87    df_terms = (88        pd.DataFrame(term_counts.items(), columns=["Term", "Frequency"])89        .sort_values(by="Frequency", ascending=False)90        .reset_index(drop=True)91    )92 93    # ---------- Hot terms ----------94    threshold = df_terms["Frequency"].quantile(0.90)95    df_terms["Hot"] = df_terms["Frequency"] >= threshold96 97    return df_terms98 99# =====================================================100# INTERFACE101# =====================================================102interface = gr.Interface(103    fn=analyze_abstracts,104    inputs=gr.Textbox(105        lines=18,106        placeholder="Cole aqui seu CSV com colunas: date,abstract"107    ),108    outputs=gr.Dataframe(),109    title="Medical Hot Terms Analyzer",110    description=(111        "Extrai termos médicos de abstracts usando PubMedBERT (NER) "112        "e identifica hot terms (top 10%)."113    ),114)115 116if __name__ == "__main__":117    interface.launch()