@cablate/banini-tracker 2.0.2 → 2.0.4

package/README.md

@@ -56,6 +56,10 @@ LLM_API_KEY=...
 LLM_MODEL=MiniMaxAI/MiniMax-M2.5
 TG_BOT_TOKEN=...
 TG_CHANNEL_ID=-100...
+
+# 影片轉錄（選填，啟用後自動轉錄影片貼文）
+TRANSCRIBER=groq
+GROQ_API_KEY=gsk_...
 ```
 
 ## CLI 工具模式
@@ -123,12 +127,30 @@ npx @cablate/banini-tracker push -m "分析結果..."
 |------|---------|------|--------|
 | Facebook 抓取（Apify） | ~$0.02 | 盤中 ~198 次 + 盤後 30 次 | ~$4.56 |
 | LLM 分析（常駐模式） | 依模型而定 | 同上 | 依模型定價 |
+| 影片轉錄（Groq Whisper） | ~$0.006/分鐘 | 視影片數量 | 極低 |
 | Telegram 推送 | 免費 | — | $0 |
 
 > 盤中：週一~五 09:00-13:30 每 30 分鐘（~9 次/日 × 22 工作日）
 > 盤後：每天 23:00（30 次/月）
 > CLI 模式搭配 Claude Code 使用則不需 LLM 費用，Claude 自己分析
 
+## 為什麼只用 Facebook？
+
+早期版本同時支援 Threads 和 Facebook 爬取，後來基於兩個原因移除了 Threads：
+
+1. **費用差距大**：Threads 每次抓取 ~$0.15（Pay-per-event），Facebook 只要 ~$0.02（CU 計費），差 7 倍以上
+2. **FB 參考價值更高**：巴逆逆的投資相關貼文（持倉截圖、操作心得）主要發在 Facebook 粉專，Threads 多為生活日常，反指標參考價值較低
+
+## Star History
+
+<a href="https://star-history.com/#cablate/banini-tracker&Date">
+ <picture>
+   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=cablate/banini-tracker&type=Date&theme=dark" />
+   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=cablate/banini-tracker&type=Date" />
+   <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=cablate/banini-tracker&type=Date" />
+ </picture>
+</a>
+
 ## 免責聲明
 
 本專案僅供娛樂參考，不構成任何投資建議。

package/dist/facebook.js

@@ -34,7 +34,9 @@ export async function fetchFacebookPosts(pageUrl, token, maxPosts = 3) {
             shareCount: item.shares ?? 0,
             url: item.url ?? '',
             mediaType: media?.__typename?.toLowerCase() ?? 'text',
-            mediaUrl: media?.thumbnail ?? media?.photo_image?.uri ?? '',
+            mediaUrl: media?.video_url ?? media?.playable_url
+                ?? (media?.__typename?.toLowerCase() === 'video' ? media?.url : null)
+                ?? media?.thumbnail ?? media?.photo_image?.uri ?? '',
         };
     });
 }

package/dist/index.js

@@ -16,6 +16,7 @@ import { analyzePosts } from './analyze.js';
 import { sendTelegramMessageWithConfig, formatReport, formatFallbackReport } from './telegram.js';
 import { filterNewPosts as filterNew, markPostsSeen } from './seen.js';
 import { withRetry } from './retry.js';
+import { createTranscriber, transcribeVideoPosts } from './transcribe.js';
 // ── Config ──────────────────────────────────────────────────
 const FB_PAGE_URL = 'https://www.facebook.com/DieWithoutBang/';
 const DATA_DIR = join(process.cwd(), 'data');
@@ -32,6 +33,7 @@ function fromFacebook(p) {
         source: 'facebook',
         text: p.text,
         ocrText: p.ocrText,
+        transcriptText: '',
         timestamp: p.timestamp,
         likeCount: p.likeCount,
         replyCount: p.commentCount,
@@ -78,6 +80,17 @@ async function runInner(opts) {
         console.log('沒有新貼文，結束');
         return;
     }
+    // 2.5. 影片轉錄
+    const transcriberType = (process.env.TRANSCRIBER ?? 'noop');
+    const transcriber = createTranscriber(transcriberType);
+    if (transcriber.name !== 'noop') {
+        const transcripts = await transcribeVideoPosts(newPosts, transcriber);
+        for (const p of newPosts) {
+            const result = transcripts.get(p.id);
+            if (result)
+                p.transcriptText = result.text;
+        }
+    }
     // 按時間從新到舊排序
     newPosts.sort((a, b) => new Date(b.timestamp).getTime() - new Date(a.timestamp).getTime());
     // 標記當天貼文
@@ -97,6 +110,8 @@ async function runInner(opts) {
         const localTime = new Date(p.timestamp).toLocaleString('zh-TW', { timeZone: 'Asia/Taipei' });
         console.log(`--- [${tag}]${todayTag} ${localTime} [${p.mediaType}] ---`);
         console.log(p.text || '（無文字，可能是純圖片）');
+        if (p.transcriptText)
+            console.log(`[影片轉錄] ${p.transcriptText}`);
         if (p.mediaUrl)
             console.log(`媒體: ${p.mediaUrl}`);
         console.log(`讚: ${p.likeCount} | 回覆: ${p.replyCount} | ${p.url}\n`);
@@ -107,13 +122,15 @@ async function runInner(opts) {
     }
     // 5. AI 分析
     const textsForAnalysis = newPosts
-        .filter((p) => p.text.trim().length > 0 || p.ocrText.trim().length > 0)
+        .filter((p) => p.text.trim().length > 0 || p.ocrText.trim().length > 0 || p.transcriptText.trim().length > 0)
         .map((p) => {
         const tag = 'Facebook';
         const localTime = new Date(p.timestamp).toLocaleString('zh-TW', { timeZone: 'Asia/Taipei' });
         let content = `[${tag}] ${p.text}`;
         if (p.ocrText)
             content += `\n[圖片 OCR] ${p.ocrText}`;
+        if (p.transcriptText)
+            content += `\n[影片轉錄] ${p.transcriptText}`;
         return { text: content, timestamp: localTime, isToday: isToday(p.timestamp) };
     });
     if (textsForAnalysis.length === 0) {

package/dist/transcribe.d.ts

@@ -0,0 +1,30 @@
+export interface TranscribeResult {
+    text: string;
+    durationSec?: number;
+}
+export interface Transcriber {
+    readonly name: string;
+    transcribe(videoUrl: string): Promise<TranscribeResult>;
+}
+export declare class NoopTranscriber implements Transcriber {
+    readonly name = "noop";
+    transcribe(_videoUrl: string): Promise<TranscribeResult>;
+}
+export declare class GroqTranscriber implements Transcriber {
+    readonly name = "groq";
+    private client;
+    private model;
+    constructor(apiKey: string, model?: string);
+    transcribe(videoUrl: string): Promise<TranscribeResult>;
+    private transcribeViaUrl;
+    private transcribeViaDownload;
+}
+export type TranscriberType = 'noop' | 'groq';
+export declare function createTranscriber(type?: TranscriberType): Transcriber;
+export declare function isVideoPost(mediaType: string): boolean;
+export interface TranscribablePost {
+    id: string;
+    mediaType: string;
+    mediaUrl: string;
+}
+export declare function transcribeVideoPosts<T extends TranscribablePost>(posts: T[], transcriber: Transcriber): Promise<Map<string, TranscribeResult>>;

package/dist/transcribe.js

@@ -0,0 +1,133 @@
+// ── 影片轉錄抽象層 ──────────────────────────────────────────
+// 策略模式：定義轉錄介面，具體實作由外部決定。
+// 新增轉錄服務時只需實作 Transcriber 介面並在 createTranscriber() 加入。
+import { execFile } from 'child_process';
+import { createReadStream, unlinkSync, mkdirSync } from 'fs';
+import { join } from 'path';
+import { tmpdir } from 'os';
+import { promisify } from 'util';
+import Groq from 'groq-sdk';
+const execFileAsync = promisify(execFile);
+// ── Noop（預設：不轉錄）────────────────────────────────────
+export class NoopTranscriber {
+    name = 'noop';
+    async transcribe(_videoUrl) {
+        return { text: '' };
+    }
+}
+// ── Groq Whisper ────────────────────────────────────────────
+// Facebook 影片 URL 通常是頁面連結（reel/watch），Groq 無法直接存取。
+// 流程：yt-dlp 下載音訊 → 傳檔案給 Groq Whisper → 清理暫存檔。
+function needsDownload(url) {
+    return /facebook\.com\/(reel|watch|video)/i.test(url);
+}
+async function downloadAudio(videoUrl) {
+    const tmpDir = join(tmpdir(), 'banini-tracker');
+    mkdirSync(tmpDir, { recursive: true });
+    const outTemplate = join(tmpDir, `audio-${Date.now}.%(ext)s`);
+    const outFile = join(tmpDir, `audio-${Date.now()}.m4a`);
+    await execFileAsync('yt-dlp', [
+        '-f', 'ba',
+        '--extract-audio',
+        '--audio-format', 'm4a',
+        '-o', outFile.replace('.m4a', '.%(ext)s'),
+        '--no-playlist',
+        '--quiet',
+        videoUrl,
+    ], { timeout: 60_000 });
+    return outFile;
+}
+export class GroqTranscriber {
+    name = 'groq';
+    client;
+    model;
+    constructor(apiKey, model = 'whisper-large-v3') {
+        this.client = new Groq({ apiKey });
+        this.model = model;
+    }
+    async transcribe(videoUrl) {
+        if (needsDownload(videoUrl)) {
+            return this.transcribeViaDownload(videoUrl);
+        }
+        return this.transcribeViaUrl(videoUrl);
+    }
+    async transcribeViaUrl(url) {
+        const result = await this.client.audio.transcriptions.create({
+            url,
+            model: this.model,
+            language: 'zh',
+            temperature: 0,
+            response_format: 'verbose_json',
+        });
+        const raw = result;
+        return {
+            text: result.text ?? '',
+            durationSec: raw.duration ? Math.round(raw.duration) : undefined,
+        };
+    }
+    async transcribeViaDownload(videoUrl) {
+        console.log(`[轉錄] 下載音訊: ${videoUrl.slice(0, 60)}...`);
+        const audioFile = await downloadAudio(videoUrl);
+        try {
+            const result = await this.client.audio.transcriptions.create({
+                file: createReadStream(audioFile),
+                model: this.model,
+                language: 'zh',
+                temperature: 0,
+                response_format: 'verbose_json',
+            });
+            const raw = result;
+            return {
+                text: result.text ?? '',
+                durationSec: raw.duration ? Math.round(raw.duration) : undefined,
+            };
+        }
+        finally {
+            try {
+                unlinkSync(audioFile);
+            }
+            catch { }
+        }
+    }
+}
+export function createTranscriber(type = 'noop') {
+    switch (type) {
+        case 'noop':
+            return new NoopTranscriber();
+        case 'groq': {
+            const apiKey = process.env.GROQ_API_KEY;
+            if (!apiKey)
+                throw new Error('GROQ_API_KEY 環境變數未設定');
+            return new GroqTranscriber(apiKey, process.env.GROQ_WHISPER_MODEL ?? 'whisper-large-v3');
+        }
+        default:
+            throw new Error(`不支援的轉錄器類型: ${type}`);
+    }
+}
+// ── 輔助：判斷貼文是否為影片 ────────────────────────────────
+export function isVideoPost(mediaType) {
+    const videoTypes = ['video', 'native_video', 'live_video', 'reel'];
+    return videoTypes.includes(mediaType.toLowerCase());
+}
+export async function transcribeVideoPosts(posts, transcriber) {
+    const results = new Map();
+    for (const post of posts) {
+        if (!isVideoPost(post.mediaType) || !post.mediaUrl)
+            continue;
+        try {
+            console.log(`[轉錄][${transcriber.name}] 處理影片: ${post.id}`);
+            const result = await transcriber.transcribe(post.mediaUrl);
+            if (result.text.trim().length > 0) {
+                results.set(post.id, result);
+                console.log(`[轉錄] ${post.id}: ${result.text.slice(0, 50)}...（${result.durationSec ?? '?'}s）`);
+            }
+            else {
+                console.log(`[轉錄] ${post.id}: 無可辨識內容`);
+            }
+        }
+        catch (err) {
+            console.error(`[轉錄] ${post.id} 失敗: ${err instanceof Error ? err.message : err}`);
+        }
+    }
+    return results;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cablate/banini-tracker",
-  "version": "2.0.2",
+  "version": "2.0.4",
   "description": "巴逆逆反指標追蹤器 — 常駐排程 + CLI 雙模式",
   "type": "module",
   "bin": {
@@ -9,7 +9,7 @@
   "scripts": {
     "dev": "tsx src/index.ts",
     "dry": "tsx src/index.ts --dry",
-    "market": "tsx src/index.ts --fb-only --max-posts=1",
+    "market": "tsx src/index.ts --max-posts=1",
     "evening": "tsx src/index.ts --max-posts=3",
     "cron": "tsx src/index.ts --cron",
     "cli": "tsx src/cli.ts",
@@ -20,7 +20,6 @@
   "keywords": [
     "banini",
     "reverse-indicator",
-    "threads",
     "facebook",
     "apify",
     "telegram",
@@ -31,6 +30,7 @@
   "dependencies": {
     "commander": "^13.0.0",
     "dotenv": "^16.4.0",
+    "groq-sdk": "^1.1.2",
     "node-cron": "^4.2.1",
     "openai": "^4.0.0"
   },