docs: start milestone v1.6 语义分段改进
This commit is contained in:
+19
-12
@@ -1,15 +1,23 @@
|
||||
# kb-skills 工作流重构
|
||||
|
||||
## Current State: v1.5 B站字幕简化处理 (Shipped 2026-03-21)
|
||||
## Current Milestone: v1.6 语义分段改进
|
||||
|
||||
**Delivered:** B站视频简化处理流程 — 基于 SRT 字幕生成易读的 Markdown 文档
|
||||
**Goal:** 使用 LLM 语义分析实现智能分段，解决当前基于时间+长度规则的问题
|
||||
|
||||
**Key Features:**
|
||||
- SRT 字幕解析与时间戳去除
|
||||
- 简单语义分段（基于时间间隔和长度）
|
||||
- LLM 自动纠正语音识别错误（可选）
|
||||
- Pipeline 平台路由（B站视频使用 SRTProcessor）
|
||||
- 文件命名包含视频标题
|
||||
**目标特性:**
|
||||
- LLM 语义分析 — 识别视频类型并适配分段策略
|
||||
- 主题/章节识别 — 自动识别并添加标题
|
||||
- 论点/观点分段 — 保持内容完整性
|
||||
- 内容变化边界 — 自然分段位置
|
||||
- 通用自适应 — 支持各种类型视频
|
||||
|
||||
### Active
|
||||
|
||||
- [ ] **语义分段改进** — 使用 LLM 替代简单规则，实现智能分段
|
||||
- [ ] **视频类型识别** — 自动判断视频类型（知识类、故事类、对话类等）
|
||||
- [ ] **主题识别与标题** — 识别主题章节并生成标题
|
||||
- [ ] **内容完整性** — 保持话题/论点的完整性，避免错误截断
|
||||
- [ ] **自然分段边界** — 在句子/段落边界处分段
|
||||
|
||||
## Previous Milestones
|
||||
|
||||
@@ -125,7 +133,6 @@ python3 kb-workflow/kb_workflow.py process "https://www.bilibili.com/video/BVxxx
|
||||
|
||||
### Deferred / Backlog
|
||||
|
||||
- [ ] **语义分段改进** — 当前基于简单规则（时间+长度），无法识别话题转换（已记录到 `.planning/backlog/semantic-segmentation.md`）
|
||||
- [ ] **ADV-01**: 进度可见性 — 实时显示工作流执行进度
|
||||
- [ ] **ADV-02**: 并行步骤执行 — 支持独立步骤的并行执行
|
||||
- [ ] **ADV-03**: 说话者分离默认启用 — HF_TOKEN 配置后自动使用
|
||||
@@ -164,7 +171,7 @@ python3 kb-workflow/kb_workflow.py process "https://www.bilibili.com/video/BVxxx
|
||||
### 技术债务
|
||||
- 环境变量配置分散 (~/.kb_env, HF_TOKEN, ZHIHU_COOKIE)
|
||||
- SKILL.md 需要更新为简化的 CLI 使用方式
|
||||
- 语义分段需要改进（基于简单规则，无法识别话题转换）
|
||||
- 语义分段 v1.5 使用简单规则（v1.6 中改进）
|
||||
|
||||
## Constraints
|
||||
|
||||
@@ -189,10 +196,10 @@ python3 kb-workflow/kb_workflow.py process "https://www.bilibili.com/video/BVxxx
|
||||
| playwright-stealth | 使用外部库实现指纹伪装 | ✓ Good — 成熟方案，活跃维护 |
|
||||
| stealth 默认启用 | stealth_config=None 时默认启用 | ✓ Good — 向后兼容，无需修改现有代码 |
|
||||
| SRT 独立模块 | SRT 处理独立于 DeepAnalyzer，位于 kb_workflow/srt/ | ✓ Good — 职责分离 |
|
||||
| 简单分段规则 | 语义分段基于时间间隔+长度，不依赖 LLM | ✓ Good — 快速，低成本 |
|
||||
| LLM 语义分段 | 使用 LLM 分析内容语义，实现智能分段 | — Pending — v1.6 目标 |
|
||||
| ASR 纠错可选 | AI 不可用时返回原文 | ✓ Good — 降级策略 |
|
||||
| 平台路由 | B站视频使用 SRTProcessor，其他使用 AnalyzeProcessor | ✓ Good — 灵活性 |
|
||||
| disable_reasoning 参数 | ASR 纠错禁用思考模式获得直接结果 | ✓ Good — 避免 AI 分析过程输出 |
|
||||
|
||||
---
|
||||
*Last updated: 2026-03-21 after completing v1.5 milestone*
|
||||
*Last updated: 2026-03-21 after starting v1.6 milestone*
|
||||
|
||||
+29
-34
@@ -1,60 +1,55 @@
|
||||
---
|
||||
gsd_state_version: 1.0
|
||||
milestone: v1.4
|
||||
milestone_name: Zhihu Anti-Crawler
|
||||
status: completed
|
||||
stopped_at: Completed 12-02 Pipeline Integration
|
||||
last_updated: "2026-03-21T15:24:05.123Z"
|
||||
last_activity: 2026-03-19 — Completed 12-02 Pipeline Integration
|
||||
milestone: v1.6
|
||||
milestone_name: 语义分段改进
|
||||
status: in_progress
|
||||
stopped_at: Defining requirements
|
||||
last_updated: "2026-03-21T16:00:00.000Z"
|
||||
last_activity: 2026-03-21 — Milestone v1.6 started
|
||||
progress:
|
||||
total_phases: 12
|
||||
completed_phases: 10
|
||||
total_plans: 31
|
||||
completed_plans: 30
|
||||
percent: 100
|
||||
total_phases: 0
|
||||
completed_phases: 0
|
||||
total_plans: 0
|
||||
completed_plans: 0
|
||||
percent: 0
|
||||
---
|
||||
|
||||
# Project State
|
||||
|
||||
## Project Reference
|
||||
|
||||
See: .planning/PROJECT.md (updated 2026-03-19)
|
||||
See: .planning/PROJECT.md (updated 2026-03-21)
|
||||
|
||||
**Core value:** 可靠、可维护的处理管道 - 每个步骤独立、可验证、可恢复
|
||||
**Current focus:** Phase 12 - SRT 简化处理器 (完成)
|
||||
**Current focus:** 语义分段改进 — 使用 LLM 实现智能分段
|
||||
|
||||
## Current Position
|
||||
|
||||
Phase: 12-srt-simplifier (2/2 plans complete)
|
||||
Plan: 12-02 (完成)
|
||||
Status: Phase 12 Complete
|
||||
Last activity: 2026-03-19 — Completed 12-02 Pipeline Integration
|
||||
Phase: Not started (defining requirements)
|
||||
Plan: —
|
||||
Status: Defining requirements
|
||||
Last activity: 2026-03-21 — Milestone v1.6 started
|
||||
|
||||
Progress: [██████████] 100%
|
||||
Progress: [░░░░░░░░░░] 0%
|
||||
|
||||
## Accumulated Context
|
||||
|
||||
### Decisions
|
||||
|
||||
Key decisions from v1.0-v1.4 (preserved):
|
||||
Key decisions from v1.0-v1.5 (preserved):
|
||||
- 管道模型 — BaseProcessor + Pipeline 架构
|
||||
- 步骤级状态 — WorkflowStateManager
|
||||
- Schema 验证 — Pydantic v2
|
||||
- 顺序执行 — 简单性和可调试性
|
||||
- CLI subparsers — 可扩展命令结构
|
||||
- SRT 独立模块 — kb_workflow/srt/ 职责分离
|
||||
- ASR 纠错可选 — AI 不可用时返回原文
|
||||
|
||||
v1.5 scope:
|
||||
- 仅针对 B 站视频处理
|
||||
- 简化流程：SRT → 去时间戳 → 分段 → 纠正 → Markdown
|
||||
- 移除深度观点分析
|
||||
|
||||
v1.5 Phase 12 decisions:
|
||||
- TDD approach with RED-GREEN cycles for all modules
|
||||
- Lazy import for ASRCorrector to avoid dependency issues
|
||||
- Simple rule-based segmentation (no LLM) for speed
|
||||
- ASR correction is optional, fallback to original text
|
||||
- B站视频使用 SRTProcessor，其他平台保持 AnalyzeProcessor
|
||||
- Pipeline 平台路由：检测 platform=bilibili + content_type=video
|
||||
v1.6 scope (待验证):
|
||||
- LLM 语义分析 — 替代简单规则分段
|
||||
- 通用自适应 — 自动识别视频类型
|
||||
- 主题识别与标题 — 主题章节生成标题
|
||||
- 内容完整性 — 避免错误截断话题
|
||||
|
||||
### Blockers/Concerns
|
||||
|
||||
@@ -62,8 +57,8 @@ None.
|
||||
|
||||
## Session Continuity
|
||||
|
||||
Last session: 2026-03-19
|
||||
Stopped at: Completed 12-02 Pipeline Integration
|
||||
Last session: 2026-03-21
|
||||
Stopped at: Milestone v1.6 started
|
||||
Resume file: None
|
||||
|
||||
**Next step:** Phase 12 完成，可以继续下一里程碑任务
|
||||
**Next step:** Define requirements for semantic segmentation
|
||||
|
||||
Reference in New Issue
Block a user