app.py

"""
🔬 NLP Insight Engine — Text Analytics Dashboard
Sentiment · NER · Keywords · Summarisation · Language Detection · Text Comparison
Built with HuggingFace Transformers + Gradio
"""

import gradio as gr
import json
from utils.pipeline import NLPPipeline
from utils.helpers import EXAMPLE_TEXTS

nlp = NLPPipeline()

ENT_CSS = {
    "PER": ("#dbeafe", "#1e40af"),
    "ORG": ("#dcfce7", "#166534"),
    "LOC": ("#fef3c7", "#92400e"),
    "MISC": ("#f3e8ff", "#6b21a8"),
}

LANG_FLAGS = {
    "English": "🇬🇧", "Spanish": "🇪🇸", "French": "🇫🇷", "German": "🇩🇪",
    "Italian": "🇮🇹", "Portuguese": "🇵🇹", "Dutch": "🇳🇱", "Turkish": "🇹🇷",
    "Arabic": "🇸🇦", "Urdu": "🇵🇰", "Chinese": "🇨🇳", "Japanese": "🇯🇵",
    "Korean": "🇰🇷", "Cyrillic": "🇷🇺", "Devanagari": "🇮🇳",
}


def make_warning_banner(lang):
    """Generate a warning banner for non-English text."""
    return (
        f'<div style="background:#fef3c7;border:1px solid #f59e0b;border-radius:8px;'
        f'padding:0.8rem 1rem;margin-bottom:1rem;display:flex;align-items:center;gap:0.5rem;">'
        f'<span style="font-size:1.2rem;">⚠️</span>'
        f'<span style="color:#92400e;font-size:0.88rem;">'
        f'<strong>Non-English text detected ({lang}).</strong> '
        f'Sentiment and NER results may be inaccurate as the models are trained on English text.'
        f'</span></div>'
    )


# ══════════════════════════════════════════════════════════════════════════════
# SINGLE TEXT ANALYSIS
# ══════════════════════════════════════════════════════════════════════════════

def analyse(text, sentiment_mode, do_sentiment, do_ner, do_keywords, do_summary):
    if not text or len(text.strip()) < 20:
        msg = "<p style='color:#ef4444;'>Please enter at least 20 characters of text.</p>"
        return msg, msg, msg, msg, msg, msg, ""

    text = text.strip()
    word_count = len(text.split())
    sent_count = max(1, text.count(".") + text.count("!") + text.count("?"))
    mode = "financial" if "Financial" in sentiment_mode else "general"

    sent_result = None
    entities = None
    kw = None
    summary_text = None

    # ── Language Detection (always on) ───────────────────────────────────
    lang_result = nlp.detect_language(text)
    lang = lang_result["language"]
    lang_conf = lang_result["confidence"]
    lang_script = lang_result["script"]
    flag = LANG_FLAGS.get(lang, "🌐")
    is_non_english = lang != "English" and lang != "Unknown"

    lang_html = (
        f'<div style="background:linear-gradient(145deg,#f8fafc,#f1f5f9);border:1px solid #e2e8f0;'
        f'border-radius:12px;padding:1.2rem;text-align:center;margin-bottom:1rem;">'
        f'<div style="font-size:2.5rem;margin-bottom:0.3rem;">{flag}</div>'
        f'<div style="font-family:monospace;font-size:1.4rem;font-weight:700;color:#0f172a;">{lang}</div>'
        f'<div style="font-size:0.78rem;color:#64748b;text-transform:uppercase;margin-top:0.2rem;">Confidence: {lang_conf:.1%}</div>'
        f'<div style="font-size:0.75rem;color:#94a3b8;margin-top:0.3rem;">Script: {lang_script}</div>'
        f'</div>'
    )

    warning = make_warning_banner(lang) if is_non_english else ""

    # ── Sentiment ────────────────────────────────────────────────────────
    if do_sentiment:
        sent_result = nlp.analyse_sentiment(text, mode=mode)
        label = sent_result["label"]
        score = sent_result["score"]
        emoji = {"POSITIVE": "😊", "NEGATIVE": "😟", "NEUTRAL": "😐"}.get(label, "🤔")

        if mode == "financial":
            badge = ('<span style="display:inline-block;background:#fef3c7;color:#92400e;'
                     'padding:0.2rem 0.5rem;border-radius:4px;font-size:0.72rem;font-weight:600;'
                     'font-family:monospace;margin-top:0.4rem;">📊 FinBERT — trained on financial text</span>')
        else:
            badge = ('<span style="display:inline-block;background:#dbeafe;color:#1e40af;'
                     'padding:0.2rem 0.5rem;border-radius:4px;font-size:0.72rem;font-weight:600;'
                     'font-family:monospace;margin-top:0.4rem;">💬 DistilBERT — trained on movie reviews</span>')

        bars_html = ""
        if "details" in sent_result:
            for d in sent_result["details"]:
                lbl = d["label"].title()
                s = d["score"]
                color = {"Positive": "#22c55e", "Negative": "#ef4444", "Neutral": "#64748b"}.get(lbl, "#64748b")
                width = max(4, int(s * 100))
                bars_html += (
                    f'<div style="display:flex;align-items:center;margin:0.4rem 0;">'
                    f'<span style="min-width:80px;font-weight:600;color:#334155;">{lbl}</span>'
                    f'<div style="flex:1;background:#f1f5f9;border-radius:6px;height:24px;margin:0 0.8rem;">'
                    f'<div style="width:{width}%;background:{color};height:100%;border-radius:6px;"></div></div>'
                    f'<span style="font-family:monospace;color:#64748b;font-size:0.85rem;">{s:.1%}</span></div>'
                )

        sentiment_html = (
            f'{warning}'
            f'<div style="background:linear-gradient(145deg,#f8fafc,#f1f5f9);border:1px solid #e2e8f0;'
            f'border-radius:12px;padding:1rem;text-align:center;margin-bottom:1rem;">'
            f'<div style="font-family:monospace;font-size:1.6rem;font-weight:700;color:#0f172a;">{emoji} {label.title()}</div>'
            f'<div style="font-size:0.78rem;color:#64748b;text-transform:uppercase;">Confidence: {score:.1%}</div>'
            f'{badge}</div>{bars_html}'
        )
    else:
        sentiment_html = "<p style='color:#94a3b8;'>Sentiment analysis disabled.</p>"

    # ── NER ──────────────────────────────────────────────────────────────
    if do_ner:
        entities = nlp.extract_entities(text)
        ner_warning = warning if is_non_english else ""
        if entities:
            tags = ""
            for ent in entities:
                bg, fg = ENT_CSS.get(ent["entity_group"], ("#f1f5f9", "#334155"))
                tags += (f'<span style="display:inline-block;padding:0.25rem 0.6rem;border-radius:6px;'
                         f'font-size:0.82rem;font-weight:600;margin:0.2rem;font-family:monospace;'
                         f'background:{bg};color:{fg};">{ent["word"]} <small>({ent["entity_group"]})</small></span> ')

            rows = "".join(
                f'<tr style="border-bottom:1px solid #f1f5f9;">'
                f'<td style="padding:0.5rem 0.8rem;">{e["word"]}</td>'
                f'<td style="padding:0.5rem 0.8rem;">{e["entity_group"]}</td>'
                f'<td style="padding:0.5rem 0.8rem;font-family:monospace;">{e["score"]:.2%}</td></tr>'
                for e in entities
            )
            table = (f'<table style="width:100%;border-collapse:collapse;margin-top:1rem;">'
                     f'<thead><tr style="border-bottom:2px solid #e2e8f0;">'
                     f'<th style="text-align:left;padding:0.5rem 0.8rem;color:#64748b;">Entity</th>'
                     f'<th style="text-align:left;padding:0.5rem 0.8rem;color:#64748b;">Type</th>'
                     f'<th style="text-align:left;padding:0.5rem 0.8rem;color:#64748b;">Confidence</th>'
                     f'</tr></thead><tbody>{rows}</tbody></table>')
            ner_html = f'{ner_warning}<div style="margin-bottom:1rem;">{tags}</div>{table}'
        else:
            ner_html = f'{ner_warning}<p style="color:#94a3b8;">No named entities detected.</p>'
    else:
        ner_html = "<p style='color:#94a3b8;'>NER disabled.</p>"

    # ── Keywords ─────────────────────────────────────────────────────────
    if do_keywords:
        kw = nlp.extract_keywords(text)
        if kw:
            max_score = kw[0][1]
            bars = ""
            for word, score in kw[:12]:
                width = max(6, int((score / max_score) * 100))
                bars += (f'<div style="display:flex;align-items:center;margin:0.35rem 0;font-family:monospace;font-size:0.85rem;">'
                         f'<span style="color:#0f172a;min-width:120px;margin-right:0.5rem;">{word}</span>'
                         f'<span style="height:22px;background:linear-gradient(90deg,#0ea5e9,#38bdf8);'
                         f'border-radius:4px;margin-right:0.6rem;min-width:4px;width:{width}%;display:inline-block;"></span>'
                         f'<span style="color:#64748b;font-size:0.78rem;">{score:.4f}</span></div>')
            keywords_html = bars
        else:
            keywords_html = "<p style='color:#94a3b8;'>No keywords extracted.</p>"
    else:
        keywords_html = "<p style='color:#94a3b8;'>Keyword extraction disabled.</p>"

    # ── Summary ──────────────────────────────────────────────────────────
    if do_summary:
        summary_text = nlp.summarise(text)
        summary_len = len(summary_text.split())
        compression = (1 - summary_len / word_count) * 100 if word_count else 0
        summary_html = (
            f'<div style="background:#f0f9ff;border-left:4px solid #0ea5e9;padding:1rem 1.2rem;'
            f'border-radius:0 8px 8px 0;font-size:0.95rem;line-height:1.6;color:#0f172a;">{summary_text}</div>'
            f'<div style="color:#64748b;font-size:0.82rem;margin-top:0.5rem;font-family:monospace;">'
            f'{word_count} → {summary_len} words ({compression:.0f}% reduction)</div>'
        )
    else:
        summary_html = "<p style='color:#94a3b8;'>Summarisation disabled.</p>"

    # ── Metrics ──────────────────────────────────────────────────────────
    sent_label = sent_result["label"].title() if sent_result else "—"
    sent_emoji = {"Positive": "😊", "Negative": "😟", "Neutral": "😐"}.get(sent_label, "—")
    ent_count = len(entities) if entities else 0

    ms = ("background:linear-gradient(145deg,#f8fafc,#f1f5f9);border:1px solid #e2e8f0;"
          "border-radius:12px;padding:1rem;text-align:center;")
    vs = "font-family:monospace;font-size:1.6rem;font-weight:700;color:#0f172a;"
    ls = "font-size:0.78rem;color:#64748b;text-transform:uppercase;"

    metrics_html = (
        f'<div style="display:grid;grid-template-columns:repeat(5,1fr);gap:0.5rem;margin-bottom:0.5rem;">'
        f'<div style="{ms}"><div style="{vs}">{word_count:,}</div><div style="{ls}">Words</div></div>'
        f'<div style="{ms}"><div style="{vs}">{sent_count}</div><div style="{ls}">Sentences</div></div>'
        f'<div style="{ms}"><div style="{vs}">{sent_emoji} {sent_label}</div><div style="{ls}">Sentiment</div></div>'
        f'<div style="{ms}"><div style="{vs}">{ent_count}</div><div style="{ls}">Entities</div></div>'
        f'<div style="{ms}"><div style="{vs}">{flag} {lang}</div><div style="{ls}">Language</div></div>'
        f'</div>'
    )

    # ── JSON export ──────────────────────────────────────────────────────
    export = {"word_count": word_count, "sentence_count": sent_count, "language": lang_result}
    if sent_result:
        export["sentiment"] = sent_result
    if entities:
        export["entities"] = entities
    if kw:
        export["keywords"] = kw
    if summary_text:
        export["summary"] = summary_text
    json_str = json.dumps(export, indent=2, default=str)

    return metrics_html, sentiment_html, ner_html, keywords_html, summary_html, lang_html, json_str


# ══════════════════════════════════════════════════════════════════════════════
# TEXT COMPARISON
# ══════════════════════════════════════════════════════════════════════════════

def compare(text_a, text_b, sentiment_mode):
    if not text_a or len(text_a.strip()) < 20:
        return "<p style='color:#ef4444;'>Text A needs at least 20 characters.</p>"
    if not text_b or len(text_b.strip()) < 20:
        return "<p style='color:#ef4444;'>Text B needs at least 20 characters.</p>"

    mode = "financial" if "Financial" in sentiment_mode else "general"
    result = nlp.compare_texts(text_a.strip(), text_b.strip(), mode=mode)

    a = result["text_a"]
    b = result["text_b"]
    shared = result["shared_keywords"]

    a_label = a["sentiment"]["label"].title()
    b_label = b["sentiment"]["label"].title()
    a_score = a["sentiment"]["score"]
    b_score = b["sentiment"]["score"]
    a_emoji = {"Positive": "😊", "Negative": "😟", "Neutral": "😐"}.get(a_label, "🤔")
    b_emoji = {"Positive": "😊", "Negative": "😟", "Neutral": "😐"}.get(b_label, "🤔")

    a_lang = a["language"]["language"]
    b_lang = b["language"]["language"]
    a_flag = LANG_FLAGS.get(a_lang, "🌐")
    b_flag = LANG_FLAGS.get(b_lang, "🌐")

    a_ents = len(a["entities"])
    b_ents = len(b["entities"])

    a_ent_words = {e["word"].lower() for e in a["entities"]}
    b_ent_words = {e["word"].lower() for e in b["entities"]}
    shared_ents = a_ent_words & b_ent_words

    # Non-English warnings for comparison
    comp_warnings = ""
    non_eng = []
    if a_lang != "English" and a_lang != "Unknown":
        non_eng.append(f"Text A ({a_lang})")
    if b_lang != "English" and b_lang != "Unknown":
        non_eng.append(f"Text B ({b_lang})")
    if non_eng:
        comp_warnings = (
            f'<div style="background:#fef3c7;border:1px solid #f59e0b;border-radius:8px;'
            f'padding:0.8rem 1rem;margin-bottom:1rem;display:flex;align-items:center;gap:0.5rem;">'
            f'<span style="font-size:1.2rem;">⚠️</span>'
            f'<span style="color:#92400e;font-size:0.88rem;">'
            f'<strong>Non-English text detected in {" and ".join(non_eng)}.</strong> '
            f'Sentiment and NER results may be inaccurate as the models are trained on English text.'
            f'</span></div>'
        )

    cs = ("background:linear-gradient(145deg,#f8fafc,#f1f5f9);border:1px solid #e2e8f0;"
          "border-radius:12px;padding:1rem;text-align:center;")
    hdr = "font-size:0.72rem;color:#94a3b8;text-transform:uppercase;letter-spacing:0.05em;margin-bottom:0.5rem;"

    html = f"""
    {comp_warnings}
    <div style="display:grid;grid-template-columns:1fr auto 1fr;gap:1rem;align-items:start;">
        <div>
            <div style="{hdr}">Text A</div>
            <div style="{cs}margin-bottom:0.8rem;">
                <div style="font-family:monospace;font-size:1.3rem;font-weight:700;">{a_emoji} {a_label}</div>
                <div style="font-size:0.75rem;color:#64748b;">Confidence: {a_score:.1%}</div>
            </div>
            <div style="{cs}margin-bottom:0.8rem;">
                <div style="font-family:monospace;font-size:1.1rem;font-weight:700;">{a_flag} {a_lang}</div>
            </div>
            <div style="{cs}margin-bottom:0.8rem;">
                <div style="font-family:monospace;font-size:1.3rem;font-weight:700;">{a["word_count"]}</div>
                <div style="font-size:0.75rem;color:#64748b;">Words</div>
            </div>
            <div style="{cs}">
                <div style="font-family:monospace;font-size:1.3rem;font-weight:700;">{a_ents}</div>
                <div style="font-size:0.75rem;color:#64748b;">Entities</div>
            </div>
        </div>
        <div style="display:flex;align-items:center;justify-content:center;height:100%;">
            <div style="font-family:monospace;font-size:1.5rem;font-weight:700;color:#cbd5e1;
                        background:#f8fafc;border-radius:50%;width:50px;height:50px;
                        display:flex;align-items:center;justify-content:center;border:2px solid #e2e8f0;">VS</div>
        </div>
        <div>
            <div style="{hdr}">Text B</div>
            <div style="{cs}margin-bottom:0.8rem;">
                <div style="font-family:monospace;font-size:1.3rem;font-weight:700;">{b_emoji} {b_label}</div>
                <div style="font-size:0.75rem;color:#64748b;">Confidence: {b_score:.1%}</div>
            </div>
            <div style="{cs}margin-bottom:0.8rem;">
                <div style="font-family:monospace;font-size:1.1rem;font-weight:700;">{b_flag} {b_lang}</div>
            </div>
            <div style="{cs}margin-bottom:0.8rem;">
                <div style="font-family:monospace;font-size:1.3rem;font-weight:700;">{b["word_count"]}</div>
                <div style="font-size:0.75rem;color:#64748b;">Words</div>
            </div>
            <div style="{cs}">
                <div style="font-family:monospace;font-size:1.3rem;font-weight:700;">{b_ents}</div>
                <div style="font-size:0.75rem;color:#64748b;">Entities</div>
            </div>
        </div>
    </div>
    <div style="margin-top:1.5rem;padding:1rem;background:#f0fdf4;border:1px solid #bbf7d0;border-radius:12px;">
        <div style="font-weight:700;color:#166534;margin-bottom:0.5rem;">🔗 Shared Keywords ({len(shared)})</div>
        <div>{"".join(f'<span style="display:inline-block;background:#dcfce7;color:#166534;padding:0.2rem 0.5rem;border-radius:4px;font-family:monospace;font-size:0.82rem;margin:0.2rem;">{kw}</span>' for kw in shared) if shared else '<span style="color:#94a3b8;">No shared keywords found</span>'}</div>
    </div>
    <div style="margin-top:0.8rem;padding:1rem;background:#eff6ff;border:1px solid #bfdbfe;border-radius:12px;">
        <div style="font-weight:700;color:#1e40af;margin-bottom:0.5rem;">🏷️ Shared Entities ({len(shared_ents)})</div>
        <div>{"".join(f'<span style="display:inline-block;background:#dbeafe;color:#1e40af;padding:0.2rem 0.5rem;border-radius:4px;font-family:monospace;font-size:0.82rem;margin:0.2rem;">{e}</span>' for e in shared_ents) if shared_ents else '<span style="color:#94a3b8;">No shared entities found</span>'}</div>
    </div>
    """
    return html


# ══════════════════════════════════════════════════════════════════════════════
# GRADIO UI
# ══════════════════════════════════════════════════════════════════════════════

with gr.Blocks(title="NLP Insight Engine") as demo:

    gr.Markdown("# 🔬 NLP Insight Engine")
    gr.Markdown("*Sentiment · Entities · Keywords · Summarisation · Language Detection · Text Comparison*")

    with gr.Tabs():

        with gr.Tab("📝 Analyse Text"):
            with gr.Row():
                with gr.Column(scale=3):
                    text_input = gr.Textbox(
                        label="Enter text to analyse",
                        placeholder="Paste an article, review, report, or any text here…",
                        lines=8, max_lines=20,
                    )
                    examples = gr.Examples(
                        examples=[[v] for v in EXAMPLE_TEXTS.values()],
                        inputs=[text_input],
                        label="Try an example",
                    )
                with gr.Column(scale=1):
                    gr.Markdown("### ⚙️ Options")
                    sentiment_mode = gr.Radio(
                        choices=["💬 General (movie reviews)", "📊 Financial (FinBERT)"],
                        value="💬 General (movie reviews)",
                        label="Sentiment Model",
                    )
                    do_sentiment = gr.Checkbox(label="Sentiment Analysis", value=True)
                    do_ner = gr.Checkbox(label="Named Entity Recognition", value=True)
                    do_keywords = gr.Checkbox(label="Keyword Extraction", value=True)
                    do_summary = gr.Checkbox(label="Extractive Summary", value=True)

            analyse_btn = gr.Button("🚀 Analyse", variant="primary", size="lg")
            metrics_output = gr.HTML(label="Quick Metrics")

            with gr.Tabs():
                with gr.Tab("💬 Sentiment"):
                    sentiment_output = gr.HTML()
                with gr.Tab("🏷️ Entities"):
                    ner_output = gr.HTML()
                with gr.Tab("🔑 Keywords"):
                    keywords_output = gr.HTML()
                with gr.Tab("📝 Summary"):
                    summary_output = gr.HTML()
                with gr.Tab("🌍 Language"):
                    lang_output = gr.HTML()
                with gr.Tab("📥 Export JSON"):
                    json_output = gr.Textbox(label="Analysis Results (JSON)", lines=15)

            analyse_btn.click(
                fn=analyse,
                inputs=[text_input, sentiment_mode, do_sentiment, do_ner, do_keywords, do_summary],
                outputs=[metrics_output, sentiment_output, ner_output, keywords_output, summary_output, lang_output, json_output],
            )

        with gr.Tab("⚖️ Compare Texts"):
            gr.Markdown("### Compare two texts side-by-side")
            gr.Markdown("*Analyse sentiment, entities, keywords, and language across two texts simultaneously.*")

            with gr.Row():
                text_a = gr.Textbox(label="Text A", placeholder="Paste first text here…", lines=6)
                text_b = gr.Textbox(label="Text B", placeholder="Paste second text here…", lines=6)

            compare_mode = gr.Radio(
                choices=["💬 General (movie reviews)", "📊 Financial (FinBERT)"],
                value="💬 General (movie reviews)",
                label="Sentiment Model",
            )
            compare_btn = gr.Button("⚖️ Compare", variant="primary", size="lg")
            compare_output = gr.HTML(label="Comparison Results")

            compare_btn.click(
                fn=compare,
                inputs=[text_a, text_b, compare_mode],
                outputs=[compare_output],
            )

            gr.Markdown("**💡 Try comparing:** A positive product review vs a negative one, "
                        "or a bullish financial article vs a bearish one.")

    gr.Markdown(
        "<center style='color:#94a3b8;font-size:0.78rem;'>"
        "NLP Insight Engine · Built by Wasif · "
        "Models: distilbert-sst2 · ProsusAI/finbert · dslim/bert-base-NER · "
        "100% free, no API keys"
        "</center>"
    )

if __name__ == "__main__":
    demo.launch(ssr_mode=False)