Skip to content

skill-harvest-gradio-feedback-panel.md

Latest commit

 

History

History
98 lines (72 loc) · 3.48 KB

skill-harvest-gradio-feedback-panel.md

File metadata and controls

98 lines (72 loc) · 3.48 KB
Raw
domain archive
title Gradio Feedback Panel
verification metadata-normalized
source skill-pipeline
status draft
created 2026-05-09

背景

（此 lesson 从 skill gradio-feedback-panel 自动提取，待补全）

根因

（待补充）

修复

Gradio Feedback Panel for RAG Web UI

Add a complete user feedback management system to a Gradio-based RAG web application. Collects thumbs-up/down, tracks category errors, displays statistics, and lists problematic queries — all from a single SQLite database.

Architecture

rag_core.py                  rag_web.py
 ┌─────────────────┐         ┌─────────────────────┐
 │ log_feedback()   │◄───────│ submit_feedback_up()│
 │ get_feedback_stats()─────►│ submit_feedback_down│
 │ get_feedback_list()       │ show_feedback_panel │
 └─────────────────┘         └─────────────────────┘
        │                          │
        ▼                          ▼
  SQLite (feedback table)    Gradio Tab "反馈管理"

Procedure

1. Add SQLite feedback queries in the core module

In rag_core.py (or equivalent core RAG module), add two functions:

get_feedback_stats(days=7) — returns dict with:

  • total, good, bad, wrong_category, satisfaction_rate
  • by_type: list of {feedback_type, count}
  • by_day: daily trend {day, good, bad, wrong_cat, total}
  • bad_list: recent bad/wrong_category feedback with comments

get_feedback_list(limit, offset, min_date, max_date, filter_type) — paginated feedback list with optional filters.

SQL queries pattern:

# By feedback_type aggregation
conn.execute("SELECT feedback_type, COUNT(*) c FROM feedback WHERE ts >= ? GROUP BY feedback_type", (cutoff,))

# Bad queries detail (for review panel)
conn.execute("""SELECT fb.id, fb.ts, fb.query, fb.feedback_type, fb.feedback_text, fb.category, fb.sender
   FROM feedback fb WHERE fb.ts >= ? AND fb.feedback_type IN ('bad','wrong_category')
   ORDER BY fb.id DESC LIMIT 50""", (cutoff,))

# Daily trend
conn.execute("""SELECT substr(ts,1,10) AS day,
    SUM(CASE WHEN feedback_type IN ('good','up') THEN 1 ELSE 0 END) good,
    SUM(CASE WHEN feedback_type IN ('bad','down') THEN 1 ELSE 0 END) bad, ...
   FROM feedback WHERE ts >= ? GROUP BY day ORDER BY day""", (cutoff,))

Make sure the feedback table exists (created in _init_log_db()):

CREATE TABLE IF NOT EXISTS feedback (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    ts TEXT NOT NULL DEFAULT (strftime('%Y-%m-%d %H:%M:%S', 'now', 'localtime')),
    query_id INTEGER REFERENCES query_log(id),
    query TEXT NOT NULL,
    feedback_type TEXT NOT NULL,   -- 'good'|'bad'|'wrong_category'|'other' (also 'up'/'down' for legacy)
    feedback_text TEXT,
    category TEXT,
    sender TEXT
);

2. Add feedback UI handlers in web UI module

In rag_web.py:

submit_feedback_up(query_id) — 👍 handler. Calls log_feedback(int(query_id), "good", "", "") from the core module (not kb_learning.json).

submit_feedback_down(query_id, comment) — 👎 handler. Calls log_feedback(int(query_id), "bad", comment or "").

show_feedback_panel(days) — generates Markdown for the feedback tab:

  • If no data: show instruction message
  • If data: stats table (total, good, bad, wrong_category, satisfa

验证

（待补充）