diff --git a/frontend/src/i18n/locales/en-US.ts b/frontend/src/i18n/locales/en-US.ts index b4d1ce2c..c3f34ba7 100755 --- a/frontend/src/i18n/locales/en-US.ts +++ b/frontend/src/i18n/locales/en-US.ts @@ -2760,6 +2760,8 @@ export default { embeddingDesc: 'Embedding model used for text vectorization', embeddingPlaceholder: 'Select an embedding model', embeddingLocked: 'Knowledge base already has files. Embedding model cannot be modified', + embeddingOptional: '(Optional)', + embeddingWikiOptionalDesc: 'Optional. When set, it powers similarity matching for Wiki directory classification so new pages reuse existing folders better; otherwise the full directory tree is used.', rerankLabel: 'ReRank Model', rerankDesc: 'Model for re-ranking search results (optional)', rerankPlaceholder: 'Select a ReRank model (optional)' diff --git a/frontend/src/i18n/locales/ko-KR.ts b/frontend/src/i18n/locales/ko-KR.ts index 15cd55c3..1e1ea8d9 100755 --- a/frontend/src/i18n/locales/ko-KR.ts +++ b/frontend/src/i18n/locales/ko-KR.ts @@ -3631,6 +3631,8 @@ export default { embeddingDesc: "텍스트 벡터화를 위한 임베딩 모델", embeddingPlaceholder: "Embedding 모델을 선택해주세요", embeddingLocked: "지식베이스에 이미 파일이 있어 Embedding 모델을 변경할 수 없습니다", + embeddingOptional: "(선택)", + embeddingWikiOptionalDesc: "선택 사항입니다. 설정하면 Wiki 디렉터리 분류의 유사도 매칭에 사용되어 새 페이지가 기존 폴더를 더 잘 재사용합니다. 설정하지 않으면 전체 디렉터리 트리를 기준으로 분류합니다.", rerankLabel: "ReRank 재정렬 모델", rerankDesc: "검색 결과 재정렬을 위한 모델 (선택)", rerankPlaceholder: "ReRank 모델을 선택해주세요 (선택)", diff --git a/frontend/src/i18n/locales/ru-RU.ts b/frontend/src/i18n/locales/ru-RU.ts index 300ae3d6..5ad93d41 100755 --- a/frontend/src/i18n/locales/ru-RU.ts +++ b/frontend/src/i18n/locales/ru-RU.ts @@ -3117,6 +3117,8 @@ export default { embeddingDesc: 'Модель встраивания для векторизации текста', embeddingPlaceholder: 'Выберите модель встраивания', embeddingLocked: 'В базе знаний уже есть файлы. Модель встраивания нельзя изменить', + embeddingOptional: '(необязательно)', + embeddingWikiOptionalDesc: 'Необязательно. Если задано, используется для сопоставления по сходству при классификации каталога Wiki, чтобы новые страницы лучше повторно использовали существующие папки; иначе классификация выполняется по полному дереву каталога.', rerankLabel: 'Модель ReRank', rerankDesc: 'Модель для переранжирования результатов поиска (необязательно)', rerankPlaceholder: 'Выберите модель ReRank (необязательно)' diff --git a/frontend/src/i18n/locales/zh-CN.ts b/frontend/src/i18n/locales/zh-CN.ts index cfd1caf1..67c3f0ee 100755 --- a/frontend/src/i18n/locales/zh-CN.ts +++ b/frontend/src/i18n/locales/zh-CN.ts @@ -3624,6 +3624,8 @@ export default { embeddingDesc: "用于文本向量化的嵌入模型", embeddingPlaceholder: "请选择 Embedding 模型", embeddingLocked: "知识库中已有文件，无法修改 Embedding 模型", + embeddingOptional: "（可选）", + embeddingWikiOptionalDesc: "可选。配置后用于 Wiki 目录归类的相似度匹配，让新页面更好地复用已有目录；不配置则按完整目录树归类。", rerankLabel: "ReRank 重排序模型", rerankDesc: "用于搜索结果重排序的模型（可选）", rerankPlaceholder: "请选择 ReRank 模型（可选）", diff --git a/frontend/src/views/knowledge/settings/KBModelConfig.vue b/frontend/src/views/knowledge/settings/KBModelConfig.vue index 4207da04..1e6a440b 100644 --- a/frontend/src/views/knowledge/settings/KBModelConfig.vue +++ b/frontend/src/views/knowledge/settings/KBModelConfig.vue @@ -25,13 +25,21 @@ - -
+ +
- -

{{ $t('knowledgeEditor.models.embeddingDesc') }}

+ +

+ {{ (wikiEnabled && ragEnabled === false) + ? $t('knowledgeEditor.models.embeddingWikiOptionalDesc') + : $t('knowledgeEditor.models.embeddingDesc') }} +

{ color: var(--td-error-color); margin-left: 2px; } + + .optional { + color: var(--td-text-color-placeholder); + font-size: 12px; + font-weight: 400; + margin-left: 4px; + } } .desc { diff --git a/internal/agent/prompts_wiki.go b/internal/agent/prompts_wiki.go index 6fa0c05d..c375240f 100644 --- a/internal/agent/prompts_wiki.go +++ b/internal/agent/prompts_wiki.go @@ -7,7 +7,7 @@ package agent // batch, when the KB still has no folders to anchor on). The result is applied // in reduce only to pages that don't already have a category, so user edits and // previously-filed pages are never churned. -const WikiTaxonomyPlanPrompt = `You are organizing a wiki knowledge base into a navigation directory. Assign each item below to a directory path (category) so the whole set lands on ONE coherent tree that reuses existing folders. +const WikiTaxonomyPlanPrompt = `You are organizing a wiki knowledge base into a navigation directory. Assign each item below to a directory path (category) so the whole set lands on ONE coherent tree. {{.ExistingTaxonomy}} @@ -20,12 +20,16 @@ const WikiTaxonomyPlanPrompt = `You are organizing a wiki knowledge base into a For every item, output a category path: an array of folder labels from broad to narrow (at most 2 levels). The category classifies WHAT the item fundamentally IS (the stable library "shelf" it always sits on), never the role it plays in one document. -Rules: -- Reuse the EXACT folder labels listed in (character-for-character) whenever an item belongs under the same domain. Do NOT invent synonym folders (e.g. do NOT create "春节习俗" when "春节 / 传统习俗" already fits). +How to choose a path for each item: +1. If an existing folder in fits, REUSE its EXACT label (character-for-character). Do NOT invent a synonym folder (e.g. do NOT create "春节习俗" when "春节 / 传统习俗" already fits). +2. If NO existing folder fits, CREATE a new, broad, durable folder for it (e.g. an organization → "组织", a legal idea → "法律概念", a place → "地点"). The directory does not have to stay small — most items DO have a natural home, so coin a sensible top-level folder rather than leaving them unfiled. Group items of the SAME kind under the SAME new folder so the tree stays coherent. +3. Only give an empty path [] when an item genuinely belongs to NO durable subject at all. This must be RARE. The absence of a matching existing folder is NOT a reason for []; create a folder instead. + +Other rules: - Group items of the SAME kind under the SAME folder at the SAME depth. Do not file one equivalent item a level deeper than its siblings (e.g. avoid "地点 / 地址 / Address1" next to "地点 / Address2" — pick one consistent depth for equivalent items). - Prefer a single broad top-level folder; add a second level only for a genuinely durable sub-domain shared by several items. - Do NOT use the item type ("entity"/"concept") as a folder. Do NOT put slashes inside a single label. -- Every item slug in MUST appear exactly once in the output. If an item truly cannot be classified under any durable folder, give it an empty path []. +- Every item slug in MUST appear exactly once in the output. - Write ALL folder labels in {{.Language}}. ### JSON Formatting Rules diff --git a/internal/application/service/wiki_ingest.go b/internal/application/service/wiki_ingest.go index afb340e2..591d291b 100644 --- a/internal/application/service/wiki_ingest.go +++ b/internal/application/service/wiki_ingest.go @@ -1536,50 +1536,6 @@ func splitSummaryLine(raw string) (summary string, content string) { return "", raw } -// splitCategoryLine extracts a leading "CATEGORY: a / b" line from content -// (the page body AFTER the SUMMARY line has been stripped) and returns the -// parsed folder labels plus the remaining content with that line removed. -// -// Category assignment lives in the page-modify output (reduce time) rather -// than per-document extraction so it reflects WHAT the whole entity IS, not -// the role it played in one source document. An empty/absent CATEGORY line -// yields nil so callers keep the page's existing category instead of wiping it. -func splitCategoryLine(content string) (category []string, rest string) { - trimmed := strings.TrimLeft(content, " \t\r\n") - if !strings.HasPrefix(trimmed, "CATEGORY:") && !strings.HasPrefix(trimmed, "CATEGORY：") { - return nil, content - } - var line, remainder string - if idx := strings.IndexByte(trimmed, '\n'); idx < 0 { - line = trimmed - } else { - line = trimmed[:idx] - remainder = trimmed[idx+1:] - } - line = strings.TrimPrefix(line, "CATEGORY:") - line = strings.TrimPrefix(line, "CATEGORY：") - return parseCategoryPath(line), strings.TrimSpace(remainder) -} - -// parseCategoryPath splits a "a / b" breadcrumb string into clean folder -// labels. Separator variants and wrapping quotes/brackets are normalized; the -// deeper structural cleaning (type-label stripping, depth cap, wiki_path) is -// left to normalizeWikiHierarchy on the page. -func parseCategoryPath(line string) []string { - line = strings.NewReplacer("／", "/", "｜", "/", "|", "/", "›", "/", "»", "/", ">", "/").Replace(line) - parts := strings.Split(line, "/") - out := make([]string, 0, len(parts)) - for _, p := range parts { - label := strings.TrimSpace(p) - label = strings.Trim(label, `"'“”‘’[]（）()`) - label = strings.TrimSpace(label) - if label != "" { - out = append(out, label) - } - } - return out -} - // buildLogEntry builds a WikiLogEntry struct for the current batch. It is // pure (no DB access) so callers can accumulate entries cheaply under their // lock and flush them in a single AppendBatch call at the end of the batch. diff --git a/internal/application/service/wiki_ingest_batch.go b/internal/application/service/wiki_ingest_batch.go index 74454d82..4be3ebdb 100644 --- a/internal/application/service/wiki_ingest_batch.go +++ b/internal/application/service/wiki_ingest_batch.go @@ -1610,11 +1610,6 @@ func (s *wikiIngestService) reduceSlugUpdates( if err == nil && updatedContent != "" { updatedSummary, updatedBody := splitSummaryLine(updatedContent) - // Category now comes from the batch taxonomy plan, not this prompt. - // Defensively strip any stray leading CATEGORY line the model may - // still emit so it never leaks into the page body; discard the value. - _, bodyAfterCategory := splitCategoryLine(updatedBody) - updatedBody = bodyAfterCategory if updatedBody != "" { page.Content = updatedBody } else { diff --git a/internal/application/service/wiki_ingest_taxonomy.go b/internal/application/service/wiki_ingest_taxonomy.go index edb34745..4b3c19b9 100644 --- a/internal/application/service/wiki_ingest_taxonomy.go +++ b/internal/application/service/wiki_ingest_taxonomy.go @@ -132,7 +132,10 @@ func (s *wikiIngestService) selectRelevantFolders( } } - if !kb.NeedsEmbeddingModel() || strings.TrimSpace(kb.EmbeddingModelID) == "" || len(deeper) == 0 { + // Gate purely on whether an embedding model is configured — NOT on + // NeedsEmbeddingModel(), which is false for wiki-only KBs that may still + // opt into an embedding model purely for directory/taxonomy similarity. + if strings.TrimSpace(kb.EmbeddingModelID) == "" || len(deeper) == 0 { return capFolders(pool, wikiTaxonomyPromptMaxPaths) } diff --git a/internal/application/service/wiki_ingest_taxonomy_test.go b/internal/application/service/wiki_ingest_taxonomy_test.go index 85440cdb..21465dc3 100644 --- a/internal/application/service/wiki_ingest_taxonomy_test.go +++ b/internal/application/service/wiki_ingest_taxonomy_test.go @@ -7,58 +7,6 @@ import ( "github.com/Tencent/WeKnora/internal/types" ) -func TestSplitCategoryLine(t *testing.T) { - tests := []struct { - name string - content string - wantCat []string - wantRest string - }{ - { - name: "two level category", - content: "CATEGORY: 组织 / 企业\n# 标题\n正文", - wantCat: []string{"组织", "企业"}, - wantRest: "# 标题\n正文", - }, - { - name: "single level category", - content: "CATEGORY: 人物\n# 陈洋", - wantCat: []string{"人物"}, - wantRest: "# 陈洋", - }, - { - name: "empty category keeps existing (nil) and strips line", - content: "CATEGORY:\n# 标题", - wantCat: nil, - wantRest: "# 标题", - }, - { - name: "no category line left untouched", - content: "# 标题\n正文", - wantCat: nil, - wantRest: "# 标题\n正文", - }, - { - name: "fullwidth colon and separators", - content: "CATEGORY：组织｜企业\n正文", - wantCat: []string{"组织", "企业"}, - wantRest: "正文", - }, - } - - for _, tt := range tests { - t.Run(tt.name, func(t *testing.T) { - cat, rest := splitCategoryLine(tt.content) - if strings.Join(cat, "/") != strings.Join(tt.wantCat, "/") { - t.Fatalf("splitCategoryLine() cat = %v, want %v", cat, tt.wantCat) - } - if rest != tt.wantRest { - t.Fatalf("splitCategoryLine() rest = %q, want %q", rest, tt.wantRest) - } - }) - } -} - func TestFormatExistingTaxonomyForPrompt(t *testing.T) { got := formatExistingTaxonomyForPrompt([][]string{ {"春节", "传统习俗"},