Redesign merchant tools and exports

app.py

@@ -8,6 +8,7 @@ import re
 import traceback
 from functools import lru_cache
 from pathlib import Path
+import tempfile
 from typing import Any, Dict, List, Tuple
 from urllib.parse import quote
 
@@ -267,6 +268,15 @@ def _report_image(name: str):
     return str(path) if path.exists() else None
 
 
+def _write_csv_download(name: str, headers: List[str], rows: List[List[Any]]):
+    path = Path(tempfile.gettempdir()) / name
+    with path.open("w", encoding="utf-8-sig", newline="") as fh:
+        writer = csv.writer(fh)
+        writer.writerow(headers)
+        writer.writerows(rows or [])
+    return str(path)
+
+
 @lru_cache(maxsize=1)
 def _predictor() -> AspectPredictor:
     return AspectPredictor(checkpoint_dir=CHECKPOINT_DIR)
@@ -662,6 +672,63 @@ def merchant_product_scores(metric: str) -> List[List[Any]]:
     return rows or [["No rows", "Try Refresh", metric, "-", 0.0, 0.0, 0.0]]
 
 
+def merchant_score_filter(aspect: str, prediction: str, category: str) -> List[List[Any]]:
+    rows = []
+    for product in PRODUCTS:
+        if category != "All" and product.get("category") != category:
+            continue
+        try:
+            result = _predict_product(product["name"])
+        except Exception:
+            continue
+        d = result.get("overall", {}) if aspect == "Overall" else result.get("aspect_details", {}).get(aspect, {})
+        label = d.get("label", "Unknown")
+        if prediction != "Any" and label != prediction:
+            continue
+        rows.append([
+            product["name"], product["category"], aspect, label,
+            round(_safe_float(d.get("confidence")), 4),
+            product["price"], product["average_rating"],
+        ])
+    return rows or [["No matching products", category, aspect, prediction, 0.0, 0.0, 0.0]]
+
+
+def export_merchant_scores(rows):
+    return _write_csv_download(
+        "merchant_product_scores.csv",
+        ["Product", "Category", "Metric", "Prediction", "Confidence", "Price", "Rating"],
+        rows,
+    )
+
+
+def negative_product_spotlight(limit: int = 6) -> str:
+    items = []
+    for product in PRODUCTS:
+        try:
+            result = _predict_product(product["name"])
+        except Exception:
+            continue
+        for aspect in ASPECTS:
+            d = result.get("aspect_details", {}).get(aspect, {})
+            if d.get("label") != "Negative":
+                continue
+            hits = _keyword_hits(product.get("review", ""), aspect)
+            reason = ", ".join(hits) if hits else _short_text(product.get("review") or product.get("features"), 90)
+            items.append((_safe_float(d.get("confidence")), aspect, product, reason))
+    items.sort(key=lambda x: x[0], reverse=True)
+    cards = []
+    for conf, aspect, product, reason in items[:limit]:
+        img = f'<img class="mini-product-img" src="{_esc(_product_image_url(product))}" alt="{_esc(product["name"])}">'
+        cards.append(
+            f'<div class="negative-card">{img}<div><b>{_esc(product["name"])}</b>'
+            f'<div class="small-label">{_esc(product["category"])} · {aspect} Negative · conf {conf:.2f}</div>'
+            f'<p>{_esc(reason)}</p></div></div>'
+        )
+    if not cards:
+        return '<div class="note-card">No highly negative product found in the current catalog sample.</div>'
+    return '<div class="negative-grid">' + "".join(cards) + '</div>'
+
+
 def _metadata_risks(features: str, categories: str, price: Any, rating: Any, count: Any) -> Dict[str, str]:
     text = f"{features} {categories}".lower()
     risks = {}
@@ -702,6 +769,36 @@ def screen_new_product(features: str, categories: str, price: Any, rating: Any,
     return f'<div class="note-card"><b>New product risk focus:</b> {_esc(summary)}<br><span class="muted">This is a metadata screening tool, not a replacement for real review evaluation.</span></div>', rows
 
 
+def import_new_product_payload(file_obj):
+    if not file_obj:
+        return gr.update(), gr.update(), gr.update(), gr.update(), gr.update(), gr.update()
+    path = Path(getattr(file_obj, "name", file_obj))
+    try:
+        if path.suffix.lower() == ".json":
+            data = json.loads(path.read_text(encoding="utf-8"))
+        else:
+            with path.open("r", encoding="utf-8-sig", newline="") as fh:
+                data = next(csv.DictReader(fh), {})
+    except Exception:
+        data = {}
+    return (
+        data.get("features") or data.get("features_text") or "",
+        data.get("categories") or data.get("categories_text") or "",
+        _safe_float(data.get("price"), 0.0),
+        _safe_float(data.get("average_rating") or data.get("rating"), 4.0),
+        _safe_float(data.get("rating_number") or data.get("rating_count"), 0.0),
+        data.get("focus_aspect") or data.get("focus") or "All",
+    )
+
+
+def export_risk_rows(rows):
+    return _write_csv_download(
+        "new_product_metadata_risk.csv",
+        ["Aspect", "Risk Level", "Model Signal", "Confidence", "Reason"],
+        rows,
+    )
+
+
 def external_review_predict(review: str, features: str, categories: str, price: Any, rating: Any, count: Any, selected_aspect: str) -> Tuple[str, str, List[List[Any]]]:
     try:
         result = _predict_custom(review, features, categories, price, rating, count)
@@ -710,6 +807,37 @@ def external_review_predict(review: str, features: str, categories: str, price:
     return _overall_html(result), _aspect_cards(result, review or "", selected_aspect), _aspect_rows(result, review or "")
 
 
+def import_external_review_payload(file_obj):
+    if not file_obj:
+        return gr.update(), gr.update(), gr.update(), gr.update(), gr.update(), gr.update(), gr.update()
+    path = Path(getattr(file_obj, "name", file_obj))
+    try:
+        if path.suffix.lower() == ".json":
+            data = json.loads(path.read_text(encoding="utf-8"))
+        else:
+            with path.open("r", encoding="utf-8-sig", newline="") as fh:
+                data = next(csv.DictReader(fh), {})
+    except Exception:
+        data = {}
+    return (
+        data.get("review") or data.get("review_text") or "",
+        data.get("features") or data.get("features_text") or "",
+        data.get("categories") or data.get("categories_text") or "",
+        _safe_float(data.get("price"), 0.0),
+        _safe_float(data.get("average_rating") or data.get("rating"), 4.0),
+        _safe_float(data.get("rating_number") or data.get("rating_count"), 0.0),
+        data.get("highlight_aspect") or data.get("aspect") or "All",
+    )
+
+
+def export_external_rows(rows):
+    return _write_csv_download(
+        "external_review_prediction.csv",
+        ["Aspect", "Prediction", "Confidence", "Key review evidence"],
+        rows,
+    )
+
+
 def _status_html() -> str:
     missing = []
     if not CHECKPOINT_PATH.exists() or CHECKPOINT_PATH.stat().st_size < 1024 * 1024:
@@ -776,7 +904,15 @@ mark { background:#fde68a; color:#111827; border-radius:4px; padding:1px 3px; }
 .kv-table td { border-bottom:1px solid #e2e8f0; padding:9px 12px; }
 .kv-table td:first-child { width:220px; color:#334155; font-weight:700; background:#f8fafc; }
 .compact-note { color:#475569; font-size:13px; margin:4px 0 10px; }
-@media (max-width:860px) { .aspect-grid, .metric-grid, .product-layout, .decision-layout { grid-template-columns:1fr; } .product-img { width:100%; height:180px; } }
+.module-head { display:flex; justify-content:space-between; align-items:center; gap:12px; margin:0 0 10px; }
+.info-tip { position:relative; display:inline-flex; align-items:center; justify-content:center; width:24px; height:24px; border-radius:999px; border:1px solid #bfdbfe; background:#eff6ff; color:#1e40af; font-weight:800; cursor:help; }
+.info-tip .tip-content { display:none; position:absolute; right:0; top:30px; z-index:20; width:420px; max-width:80vw; background:white; border:1px solid var(--line); border-radius:8px; padding:10px; box-shadow:0 12px 30px rgba(15,23,42,.16); }
+.info-tip:hover .tip-content { display:block; }
+.negative-grid { display:grid; grid-template-columns:repeat(3, minmax(0, 1fr)); gap:10px; margin:8px 0 14px; }
+.negative-card { display:grid; grid-template-columns:64px 1fr; gap:10px; border:1px solid #fecaca; border-left:4px solid #ef4444; border-radius:8px; padding:10px; background:#fffafa; }
+.negative-card p { margin:5px 0 0; color:#334155; font-size:13px; }
+.mini-product-img { width:64px; height:76px; object-fit:cover; border-radius:6px; border:1px solid var(--line); background:#f8fafc; }
+@media (max-width:860px) { .aspect-grid, .metric-grid, .product-layout, .decision-layout, .negative-grid { grid-template-columns:1fr; } .product-img { width:100%; height:180px; } }
 """
 
 
@@ -818,45 +954,64 @@ def build_app() -> gr.Blocks:
                 filter_btn.click(filter_products, [filter_aspect, filter_sentiment, filter_category, filter_tags, min_rating], [filter_table, filter_summary])
 
             with gr.Tab("Merchant Interface"):
-                with gr.Row():
-                    with gr.Column(scale=4):
-                        gr.Markdown("### Model Basic Information")
-                        gr.HTML(model_info_html())
-                    with gr.Column(scale=6):
-                        gr.Markdown("### Product Score Monitor")
-                        merchant_metric = gr.Dropdown(["Overall"] + ASPECTS, value="Overall", label="Choose overall or aspect")
-                        merchant_scores = gr.Dataframe(headers=["Product", "Category", "Metric", "Prediction", "Confidence", "Price", "Rating"], value=[["Click Refresh Product Scores", "", "", "", 0.0, 0.0, 0.0]], datatype=["str", "str", "str", "str", "number", "number", "number"], interactive=False)
-                        refresh_scores = gr.Button("Refresh Product Scores", variant="primary")
-                with gr.Accordion("New Product Metadata Risk Screening", open=False):
+                gr.HTML('<div class="module-head"><h3>Product Score Monitor</h3><span class="info-tip">i<span class="tip-content">' + model_info_html() + '</span></span></div>')
+                gr.HTML('<div class="small-label">Most negative products across the six aspects</div>')
+                negative_spotlight = gr.HTML('<div class="note-card">Loading most negative products...</div>')
+                with gr.Accordion("Score filters and export", open=True):
                     with gr.Row():
-                        new_features = gr.Textbox("Cotton Polyester Blend, Slim Fit, Graphic Print, Machine Wash", label="New product features")
-                        new_categories = gr.Textbox("Clothing > Women > Tops > T-Shirts", label="New product categories")
+                        merchant_metric = gr.Dropdown(["Overall"] + ASPECTS, value="Overall", label="Aspect")
+                        merchant_prediction = gr.Dropdown(["Any", "Positive", "Negative", "Not_Mentioned", "Neutral"], value="Any", label="Prediction")
+                        merchant_category = gr.Dropdown(categories, value="All", label="Category")
+                    merchant_scores = gr.Dataframe(headers=["Product", "Category", "Metric", "Prediction", "Confidence", "Price", "Rating"], value=[["Click Apply Score Filter", "", "", "", 0.0, 0.0, 0.0]], datatype=["str", "str", "str", "str", "number", "number", "number"], interactive=False)
                     with gr.Row():
-                        new_price = gr.Number(29.99, label="Price")
-                        new_rating = gr.Number(4.1, label="Expected or early average rating")
-                        new_count = gr.Number(35, label="Expected or early rating count")
-                        new_focus = gr.Dropdown(["All"] + ASPECTS, value="All", label="Focus aspect")
-                    screen_btn = gr.Button("Predict Metadata Risk", variant="primary")
-                    risk_summary = gr.HTML()
-                    risk_table = gr.Dataframe(headers=["Aspect", "Risk Level", "Model Signal", "Confidence", "Reason"], value=[["Click Predict Metadata Risk", "", "", 0.0, ""]], datatype=["str", "str", "str", "number", "str"], interactive=False)
-                with gr.Accordion("External Review Prediction", open=False):
-                    external_review = gr.Textbox("The fabric is soft and the color looks good, but it runs small and the zipper feels weak.", label="External customer review", lines=4)
+                        refresh_scores = gr.Button("Apply Score Filter", variant="primary")
+                        export_scores = gr.Button("Export Score List")
+                    merchant_scores_file = gr.File(label="Downloaded score CSV", interactive=False)
+                with gr.Accordion("New Product Metadata Risk Screening", open=False):
                     with gr.Row():
-                        ext_features = gr.Textbox("Cotton Blend, Slim Fit, Zipper Closure", label="Product features")
-                        ext_categories = gr.Textbox("Clothing > Women > Jackets", label="Product categories")
+                        with gr.Column(scale=1):
+                            new_import = gr.File(label="Import product metadata JSON/CSV")
+                            new_features = gr.Textbox("Cotton Polyester Blend, Slim Fit, Graphic Print, Machine Wash", label="New product features", lines=4)
+                            new_categories = gr.Textbox("Clothing > Women > Tops > T-Shirts", label="New product categories")
+                            new_price = gr.Number(29.99, label="Price")
+                            new_rating = gr.Number(4.1, label="Expected or early average rating")
+                            new_count = gr.Number(35, label="Expected or early rating count")
+                            new_focus = gr.Dropdown(["All"] + ASPECTS, value="All", label="Focus aspect")
+                            screen_btn = gr.Button("Predict Metadata Risk", variant="primary")
+                        with gr.Column(scale=1):
+                            risk_summary = gr.HTML()
+                            risk_table = gr.Dataframe(headers=["Aspect", "Risk Level", "Model Signal", "Confidence", "Reason"], value=[["Click Predict Metadata Risk", "", "", 0.0, ""]], datatype=["str", "str", "str", "number", "str"], interactive=False)
+                            export_risk = gr.Button("Export Risk Result")
+                            risk_file = gr.File(label="Downloaded risk CSV", interactive=False)
+                with gr.Accordion("External Review Prediction", open=False):
                     with gr.Row():
-                        ext_price = gr.Number(39.99, label="Price")
-                        ext_rating = gr.Number(4.2, label="Average rating")
-                        ext_count = gr.Number(312, label="Rating count")
-                        ext_aspect = gr.Dropdown(["All"] + ASPECTS, value="All", label="Highlight aspect")
-                    external_btn = gr.Button("Analyze External Review", variant="primary")
-                    ext_overall = gr.HTML()
-                    ext_aspects = gr.HTML()
-                    ext_table = gr.Dataframe(headers=["Aspect", "Prediction", "Confidence", "Key review evidence"], value=[["Click Analyze External Review", "", 0.0, ""]], datatype=["str", "str", "number", "str"], interactive=False)
-                refresh_scores.click(merchant_product_scores, merchant_metric, merchant_scores)
-                merchant_metric.change(merchant_product_scores, merchant_metric, merchant_scores)
+                        with gr.Column(scale=1):
+                            ext_import = gr.File(label="Import review metadata JSON/CSV")
+                            external_review = gr.Textbox("The fabric is soft and the color looks good, but it runs small and the zipper feels weak.", label="External customer review", lines=5)
+                            ext_features = gr.Textbox("Cotton Blend, Slim Fit, Zipper Closure", label="Product features", lines=3)
+                            ext_categories = gr.Textbox("Clothing > Women > Jackets", label="Product categories")
+                            ext_price = gr.Number(39.99, label="Price")
+                            ext_rating = gr.Number(4.2, label="Average rating")
+                            ext_count = gr.Number(312, label="Rating count")
+                            ext_aspect = gr.Dropdown(["All"] + ASPECTS, value="All", label="Highlight aspect")
+                            external_btn = gr.Button("Analyze External Review", variant="primary")
+                        with gr.Column(scale=1):
+                            ext_overall = gr.HTML()
+                            ext_aspects = gr.HTML()
+                            ext_table = gr.Dataframe(headers=["Aspect", "Prediction", "Confidence", "Key review evidence"], value=[["Click Analyze External Review", "", 0.0, ""]], datatype=["str", "str", "number", "str"], interactive=False)
+                            export_ext = gr.Button("Export Review Result")
+                            ext_file = gr.File(label="Downloaded review CSV", interactive=False)
+                refresh_scores.click(merchant_score_filter, [merchant_metric, merchant_prediction, merchant_category], merchant_scores)
+                export_scores.click(export_merchant_scores, merchant_scores, merchant_scores_file)
+                merchant_metric.change(merchant_score_filter, [merchant_metric, merchant_prediction, merchant_category], merchant_scores)
+                merchant_prediction.change(merchant_score_filter, [merchant_metric, merchant_prediction, merchant_category], merchant_scores)
+                merchant_category.change(merchant_score_filter, [merchant_metric, merchant_prediction, merchant_category], merchant_scores)
+                new_import.change(import_new_product_payload, new_import, [new_features, new_categories, new_price, new_rating, new_count, new_focus])
                 screen_btn.click(screen_new_product, [new_features, new_categories, new_price, new_rating, new_count, new_focus], [risk_summary, risk_table])
+                export_risk.click(export_risk_rows, risk_table, risk_file)
+                ext_import.change(import_external_review_payload, ext_import, [external_review, ext_features, ext_categories, ext_price, ext_rating, ext_count, ext_aspect])
                 external_btn.click(external_review_predict, [external_review, ext_features, ext_categories, ext_price, ext_rating, ext_count, ext_aspect], [ext_overall, ext_aspects, ext_table])
+                export_ext.click(export_external_rows, ext_table, ext_file)
 
             with gr.Tab("Research Metrics"):
                 gr.Markdown("Metrics are loaded from the updated 10W experiment reports. Tables use CSV outputs when available, with JSON fallback.")
@@ -879,7 +1034,8 @@ def build_app() -> gr.Blocks:
                 refresh_research = gr.Button("Refresh Research Metrics", variant="primary")
                 refresh_research.click(refresh_research_outputs, outputs=[research_cards, overall_table, aspect_table, ablation_table, meta_source_table])
         demo.load(consumer_product_view, [product_select, consumer_aspect], [product_detail, aspect_html, evidence_html, consumer_table])
-        demo.load(merchant_product_scores, merchant_metric, merchant_scores)
+        demo.load(negative_product_spotlight, outputs=negative_spotlight)
+        demo.load(merchant_score_filter, [merchant_metric, merchant_prediction, merchant_category], merchant_scores)
         demo.load(filter_products, [filter_aspect, filter_sentiment, filter_category, filter_tags, min_rating], [filter_table, filter_summary])
     return demo