app.py

import streamlit as st
import pandas as pd
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

MODEL_NAME = "valhalla/distilbart-mnli-12-1"
device = 0 if torch.cuda.is_available() else -1

@st.cache_resource
def load_zero_shot_pipeline():
    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
    model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
    return pipeline("zero-shot-classification", model=model, tokenizer=tokenizer, device=device)

zero_shot = load_zero_shot_pipeline()

st.title("ArticleInsight (Demo Offline Pipeline)")

st.markdown(\"\"\"
**Upload a CSV** with an 'Abstract' column. We'll run a simple 8-step analysis:
1. Empirical Study?
2. Identify Construct
3. Sample Details
4. Main Research Question
5. Key Findings
6. Variables (IV/DV)
7. Antecedents, Outcomes
8. Unit of Analysis

**Disclaimer**: This is a *very naive* demonstration using zero-shot classification and simple regex. 
It won't be super accurate, but requires no coding from you!
\"\"\")

uploaded_file = st.file_uploader("Upload CSV with 'Abstract' column")

if uploaded_file:
    df = pd.read_csv(uploaded_file)
    if "Abstract" not in df.columns:
        st.error("CSV must have an 'Abstract' column.")
        st.stop()

    st.success("File uploaded successfully!")
    if st.button("Run Analysis"):
        with st.spinner("Analyzing each abstract..."):
            df["Empirical Study"] = ""
            df["Construct"] = ""
            df["Sample Details"] = ""
            df["Research Question"] = ""
            df["Key Findings"] = ""
            df["Variables"] = ""
            df["Antecedents"] = ""
            df["Outcomes"] = ""
            df["Unit of Analysis"] = ""

            for i, row in df.iterrows():
                abstract = str(row["Abstract"])
                df.at[i, "Empirical Study"] = classify_empirical(abstract)
                if df.at[i, "Empirical Study"] == "Yes":
                    df.at[i, "Construct"] = find_constructs(abstract)
                    df.at[i, "Sample Details"] = extract_sample_details(abstract)
                    df.at[i, "Research Question"] = guess_research_question(abstract)
                    df.at[i, "Key Findings"] = guess_key_findings(abstract)
                    var, ants, outs = identify_variables(abstract)
                    df.at[i, "Variables"] = var
                    df.at[i, "Antecedents"] = ants
                    df.at[i, "Outcomes"] = outs
                    df.at[i, "Unit of Analysis"] = identify_unit_of_analysis(abstract)
                else:
                    for col in ["Construct", "Sample Details", "Research Question", "Key Findings", "Variables", "Antecedents", "Outcomes", "Unit of Analysis"]:
                        df.at[i, col] = "N/A"

        st.success("Done!")
        st.dataframe(df.head(50))
        csv_data = df.to_csv(index=False).encode("utf-8")
        st.download_button("Download Analyzed CSV", data=csv_data, file_name="analysis_output.csv", mime="text/csv")


# === Functions ===

def classify_empirical(text):
    candidate_labels = ["empirical study", "theoretical paper"]
    res = zero_shot(text, candidate_labels)
    top_label = res["labels"][0]
    top_score = res["scores"][0]
    if top_label == "empirical study" and top_score > 0.5:
        return "Yes"
    elif top_label == "theoretical paper" and top_score > 0.5:
        return "No"
    return "Unknown"

def find_constructs(text):
    tokens = text.lower().split()
    freq = {}
    for w in tokens:
        if len(w) > 5 and w.isalpha():
            freq[w] = freq.get(w, 0) + 1
    sorted_freq = sorted(freq.items(), key=lambda x: x[1], reverse=True)
    if not sorted_freq:
        return "Unknown"
    return ", ".join([x[0] for x in sorted_freq[:2]])

def extract_sample_details(text):
    import re
    t = text.lower()
    pattern = r"(n\s*=\s*\d+|sample of \d+|\d+\s+participants|\d+\s+subjects)"
    matches = re.findall(pattern, t)
    info = "; ".join([m[0] if isinstance(m, tuple) else m for m in matches]) if matches else ""
    if "student" in t:
        info += "; students"
    if "employee" in t:
        info += "; employees"
    return info if info else "Unknown"

def guess_research_question(text):
    lower = text.lower()
    if "effect of" in lower:
        idx = lower.index("effect of")
        snippet = text[idx: idx+60]
        return f"Does {snippet}?"
    elif "aim of this study" in lower:
        idx = lower.index("aim of this study")
        snippet = text[idx: idx+60]
        return snippet
    return "Unknown"

def guess_key_findings(text):
    lower = text.lower()
    if "we find that" in lower:
        idx = lower.index("we find that")
        return text[idx: idx+100]
    elif "results show" in lower:
        idx = lower.index("results show")
        return text[idx: idx+100]
    return "Unknown"

def identify_variables(text):
    import re
    t = text.lower()
    pattern = r"(impact|influence|effect) of (\w+) on (\w+)"
    match = re.search(pattern, t)
    if match:
        iv = match.group(2)
        dv = match.group(3)
        return f"IV: {iv}, DV: {dv}", iv, dv
    return "Unknown", "Unknown", "Unknown"

def identify_unit_of_analysis(text):
    lower = text.lower()
    if "team" in lower or "groups" in lower:
        return "Team"
    if "organization" in lower or "firm" in lower:
        return "Organization"
    if any(x in lower for x in ["participant", "individual", "student", "employee"]):
        return "Individual"
    return "Unknown"

import os
os.environ["STREAMLIT_SERVER_HEADLESS"] = "true"
os.environ["STREAMLIT_SERVER_ADDRESS"] = "0.0.0.0"