536 lines
16 KiB
Go
536 lines
16 KiB
Go
package usecase
|
||
|
||
import (
|
||
"encoding/json"
|
||
"fmt"
|
||
"regexp"
|
||
"strings"
|
||
"unicode/utf8"
|
||
|
||
"apps/backend/internal/module/studio/domain"
|
||
)
|
||
|
||
// 8D keys — 與前端 DIM_ORDER 對齊
|
||
var dimKeys = []string{
|
||
"d1Tone", "d2Structure", "d3Interaction", "d4Topics",
|
||
"d5Rhythm", "d6Visual", "d7Conversion", "d8Risk",
|
||
}
|
||
|
||
func splitSamples(raw string, max int) []string {
|
||
parts := regexp.MustCompile(`\n\s*---\s*\n|\n{3,}`).Split(raw, -1)
|
||
out := make([]string, 0, max)
|
||
for _, p := range parts {
|
||
p = strings.TrimSpace(p)
|
||
if utf8.RuneCountInString(p) < 8 {
|
||
continue
|
||
}
|
||
out = append(out, p)
|
||
if max > 0 && len(out) >= max {
|
||
break
|
||
}
|
||
}
|
||
if len(out) == 0 && strings.TrimSpace(raw) != "" {
|
||
out = append(out, strings.TrimSpace(raw))
|
||
}
|
||
return out
|
||
}
|
||
|
||
func pickTone(blob string) string {
|
||
switch {
|
||
case regexp.MustCompile(`哈哈|笑死|哭|靠|真的假的`).MatchString(blob):
|
||
return "輕鬆吐槽、情緒外露"
|
||
case regexp.MustCompile(`建議|成分|數據|對比|規格`).MatchString(blob):
|
||
return "務實、有依據、少廢話"
|
||
case regexp.MustCompile(`懂|抱抱|辛苦|没关系|沒關係|一起`).MatchString(blob):
|
||
return "共感、溫柔、像朋友"
|
||
default:
|
||
return "口語親近、不端著"
|
||
}
|
||
}
|
||
|
||
func pickHooks(blob string) string {
|
||
switch {
|
||
case strings.ContainsAny(blob, "？?"):
|
||
return "用真心疑問開場，邀請別人補經驗"
|
||
case regexp.MustCompile(`有人|大家|求`).MatchString(blob):
|
||
return "先丟痛點再求推坑／經驗"
|
||
default:
|
||
return "先講自己卡關的情境，再拋問題"
|
||
}
|
||
}
|
||
|
||
func pickEvidence(samples []string, max int) []string {
|
||
out := make([]string, 0, max)
|
||
for _, s := range samples {
|
||
r := []rune(s)
|
||
if len(r) > 36 {
|
||
s = string(r[:36]) + "…"
|
||
}
|
||
out = append(out, s)
|
||
if len(out) >= max {
|
||
break
|
||
}
|
||
}
|
||
return out
|
||
}
|
||
|
||
// serializeDraftText — 與前端 fingerprint 格式一致（【標籤】）
|
||
func serializeDraftText(d domain.PersonaDraft) string {
|
||
var lines []string
|
||
push := func(label, value string) {
|
||
v := strings.TrimSpace(value)
|
||
if v == "" {
|
||
return
|
||
}
|
||
lines = append(lines, "【"+label+"】\n"+v)
|
||
}
|
||
push("我是誰", d.Identity)
|
||
push("語氣", d.Tone)
|
||
push("對誰說", d.Audience)
|
||
push("開場鉤子", d.Hooks)
|
||
push("語言指紋", d.LanguageFingerprint)
|
||
push("節奏", d.Rhythm)
|
||
push("標點", d.Punctuation)
|
||
push("內容套路", d.ContentPatterns)
|
||
push("知識轉譯", d.KnowledgeTranslation)
|
||
push("CTA", d.CtaStyle)
|
||
push("像他會說的話", d.Examples)
|
||
push("絕不怎麼說", d.Avoid)
|
||
return strings.Join(lines, "\n\n")
|
||
}
|
||
|
||
func buildDimensions(draft domain.PersonaDraft, samples []string, dims map[string]domain.StyleDimension) map[string]domain.StyleDimension {
|
||
ev := pickEvidence(samples, 2)
|
||
out := make(map[string]domain.StyleDimension, 8)
|
||
// 規則底稿
|
||
defaults := map[string]string{
|
||
"d1Tone": draft.Tone,
|
||
"d2Structure": draft.ContentPatterns,
|
||
"d3Interaction": joinNonEmpty("；", draft.Hooks, draft.CtaStyle),
|
||
"d4Topics": joinNonEmpty("；", draft.Audience, "常談生活／實用經驗"),
|
||
"d5Rhythm": draft.Rhythm,
|
||
"d6Visual": draft.Punctuation,
|
||
"d7Conversion": draft.CtaStyle,
|
||
"d8Risk": draft.Avoid,
|
||
}
|
||
for _, k := range dimKeys {
|
||
sum := ""
|
||
var evidence []string
|
||
if dims != nil {
|
||
if d, ok := dims[k]; ok {
|
||
sum = strings.TrimSpace(d.Summary)
|
||
evidence = d.Evidence
|
||
}
|
||
}
|
||
if sum == "" {
|
||
sum = strings.TrimSpace(defaults[k])
|
||
}
|
||
if sum == "" {
|
||
sum = "（待補）"
|
||
}
|
||
if len(evidence) == 0 {
|
||
evidence = append([]string(nil), ev...)
|
||
}
|
||
out[k] = domain.StyleDimension{Summary: sum, Evidence: evidence}
|
||
}
|
||
return out
|
||
}
|
||
|
||
func joinNonEmpty(sep string, parts ...string) string {
|
||
var ok []string
|
||
for _, p := range parts {
|
||
p = strings.TrimSpace(p)
|
||
if p != "" {
|
||
ok = append(ok, p)
|
||
}
|
||
}
|
||
return strings.Join(ok, sep)
|
||
}
|
||
|
||
func splitAvoidList(avoid string) []string {
|
||
var out []string
|
||
for _, a := range regexp.MustCompile(`[、,，;；/｜|]`).Split(avoid, -1) {
|
||
a = strings.TrimSpace(a)
|
||
if a != "" {
|
||
out = append(out, a)
|
||
}
|
||
}
|
||
if len(out) == 0 {
|
||
return []string{"硬銷", "客服腔", "條列說教", "AI 腔"}
|
||
}
|
||
return out
|
||
}
|
||
|
||
func isPlaceholderName(name string) bool {
|
||
n := strings.TrimSpace(name)
|
||
if n == "" {
|
||
return true
|
||
}
|
||
// 前端新建預設／測試名
|
||
switch n {
|
||
case "新人設", "新身份", "未命名人設", "未命名", "分析", "對標", "新", "New persona", "Untitled":
|
||
return true
|
||
}
|
||
return strings.HasPrefix(n, "人設 ") || strings.HasPrefix(n, "persona")
|
||
}
|
||
|
||
func shortNameFromIdentity(identity, fallback string) string {
|
||
id := strings.TrimSpace(identity)
|
||
if id == "" {
|
||
return fallback
|
||
}
|
||
// 取第一句或前 16 字
|
||
if i := strings.IndexAny(id, "，,。；;"); i > 0 && i <= 24 {
|
||
id = id[:i]
|
||
}
|
||
r := []rune(id)
|
||
if len(r) > 16 {
|
||
id = string(r[:16])
|
||
}
|
||
id = strings.TrimSpace(id)
|
||
if id == "" {
|
||
return fallback
|
||
}
|
||
return id
|
||
}
|
||
|
||
// finalizePersonaAnalysis 寫入概要 + 指紋 + 8D（分析完成後一律覆寫可分析欄位）
|
||
func finalizePersonaAnalysis(
|
||
p *domain.Persona,
|
||
draft domain.PersonaDraft,
|
||
samples []string,
|
||
source, username, sourceLabel, brief, nameHint string,
|
||
dims map[string]domain.StyleDimension,
|
||
) {
|
||
now := domain.NowNano()
|
||
p.Status = domain.PersonaReady
|
||
p.Style.Source = source
|
||
p.Style.BenchmarkUsername = username
|
||
p.Style.SourceLabel = sourceLabel
|
||
p.Style.SampleCount = len(samples)
|
||
p.Style.AnalyzedAt = now
|
||
p.Style.SamplePreviews = make([]string, 0, min(5, len(samples)))
|
||
for i, s := range samples {
|
||
if i >= 5 {
|
||
break
|
||
}
|
||
r := []rune(s)
|
||
if len(r) > 100 {
|
||
s = string(r[:100]) + "…"
|
||
}
|
||
p.Style.SamplePreviews = append(p.Style.SamplePreviews, s)
|
||
}
|
||
p.Style.Draft = draft
|
||
p.Style.DraftText = serializeDraftText(draft)
|
||
p.Style.Dimensions = buildDimensions(draft, samples, dims)
|
||
p.Voice = draft.Tone
|
||
|
||
// 概要 brief：分析後一律填（LLM brief 優先）
|
||
brief = strings.TrimSpace(brief)
|
||
if brief == "" {
|
||
brief = joinNonEmpty("。", draft.Identity, draft.Tone+"；對「"+draft.Audience+"」說話")
|
||
}
|
||
p.Brief = brief
|
||
|
||
// 名稱：仍是佔位名時用 identity / hint 補上
|
||
if isPlaceholderName(p.Name) {
|
||
hint := strings.TrimSpace(nameHint)
|
||
if hint == "" {
|
||
hint = shortNameFromIdentity(draft.Identity, p.Name)
|
||
}
|
||
if hint != "" && !isPlaceholderName(hint) {
|
||
p.Name = hint
|
||
} else if draft.Identity != "" {
|
||
p.Name = shortNameFromIdentity(draft.Identity, "寫作人設")
|
||
}
|
||
}
|
||
|
||
// 護欄：分析後覆寫 avoid / 預設字數 / 禁 AI 腔
|
||
p.Guard.BanAiTone = true
|
||
if p.Guard.MaxChars == 0 {
|
||
p.Guard.MaxChars = 280
|
||
}
|
||
if draft.Avoid != "" {
|
||
p.Guard.Avoid = splitAvoidList(draft.Avoid)
|
||
} else if len(p.Guard.Avoid) == 0 {
|
||
p.Guard.Avoid = []string{"硬銷", "客服腔", "條列說教", "AI 腔"}
|
||
}
|
||
|
||
// notes 放分析來源摘要（不覆寫使用者手動長註？分析後可寫來源）
|
||
if source == "benchmark" && username != "" {
|
||
p.Notes = fmt.Sprintf("對標 @%s · 樣本 %d 則 · 已寫入概要／8D／指紋", username, len(samples))
|
||
} else if sourceLabel != "" {
|
||
p.Notes = fmt.Sprintf("來源：%s · 樣本 %d 則 · 已寫入概要／8D／指紋", sourceLabel, len(samples))
|
||
}
|
||
|
||
p.UpdatedAt = now
|
||
}
|
||
|
||
// applyStyleFromSamples fills persona style from real post samples (中文啟發式，可接 AI 再精修).
|
||
func applyStyleFromSamples(p *domain.Persona, samples []string, source string, username, sourceLabel string) {
|
||
blob := strings.Join(samples, "\n")
|
||
tone := pickTone(blob)
|
||
hooks := pickHooks(blob)
|
||
|
||
identity := p.Style.Draft.Identity
|
||
if identity == "" {
|
||
identity = p.Name
|
||
}
|
||
if isPlaceholderName(identity) {
|
||
identity = "生活觀察者"
|
||
}
|
||
if source == "benchmark" && username != "" {
|
||
identity = fmt.Sprintf("像 @%s 的發文者：%s", username, tone)
|
||
}
|
||
audience := p.Style.Draft.Audience
|
||
if audience == "" {
|
||
audience = "同溫層、會在 Threads 求經驗的人"
|
||
}
|
||
langFP := "短句、口語、少成語"
|
||
if regexp.MustCompile(`有時候|後來|其實|講真的`).MatchString(blob) {
|
||
langFP = "常用「有時候／後來／其實」當轉折"
|
||
}
|
||
rhythm := "一段講完再補一句問句"
|
||
if strings.Contains(blob, "\n") {
|
||
rhythm = "2～4 個短段落，段間空行"
|
||
}
|
||
punct := "逗號與問號為主，少驚嘆"
|
||
if strings.Contains(blob, "…") || strings.Contains(blob, "...") {
|
||
punct = "愛用省略號與問號"
|
||
}
|
||
avoid := "硬銷、條列教學、客服腔、假裝中立實則業配、AI 腔"
|
||
ex := ""
|
||
if len(samples) > 0 {
|
||
r := []rune(samples[0])
|
||
if len(r) > 80 {
|
||
ex = string(r[:80])
|
||
} else {
|
||
ex = samples[0]
|
||
}
|
||
}
|
||
|
||
draft := domain.PersonaDraft{
|
||
Identity: identity,
|
||
Tone: tone,
|
||
Audience: audience,
|
||
Hooks: hooks,
|
||
LanguageFingerprint: langFP,
|
||
Rhythm: rhythm,
|
||
Punctuation: punct,
|
||
ContentPatterns: "情境 → 感受或觀察 → 輕問一句",
|
||
KnowledgeTranslation: "先講使用情境，再講重點，不丟術語牆",
|
||
CtaStyle: "自然邀留言（「你們呢」「有人也…嗎」），不命令",
|
||
Examples: ex,
|
||
Avoid: avoid,
|
||
}
|
||
brief := fmt.Sprintf("%s。語氣%s，主要對「%s」說話。", identity, tone, audience)
|
||
finalizePersonaAnalysis(p, draft, samples, source, username, sourceLabel, brief, "", nil)
|
||
}
|
||
|
||
func min(a, b int) int {
|
||
if a < b {
|
||
return a
|
||
}
|
||
return b
|
||
}
|
||
|
||
type aiDimJSON struct {
|
||
Summary string `json:"summary"`
|
||
Evidence []string `json:"evidence"`
|
||
}
|
||
|
||
type aiStyleJSON struct {
|
||
Identity string `json:"identity"`
|
||
Tone string `json:"tone"`
|
||
Audience string `json:"audience"`
|
||
Hooks string `json:"hooks"`
|
||
LanguageFingerprint string `json:"language_fingerprint"`
|
||
Rhythm string `json:"rhythm"`
|
||
Punctuation string `json:"punctuation"`
|
||
ContentPatterns string `json:"content_patterns"`
|
||
KnowledgeTranslation string `json:"knowledge_translation"`
|
||
CtaStyle string `json:"cta_style"`
|
||
Examples string `json:"examples"`
|
||
Avoid string `json:"avoid"`
|
||
Brief string `json:"brief"`
|
||
Name string `json:"name"`
|
||
Dimensions map[string]aiDimJSON `json:"dimensions"`
|
||
// 扁平 8D 後備
|
||
D1Tone aiDimJSON `json:"d1Tone"`
|
||
D2Structure aiDimJSON `json:"d2Structure"`
|
||
D3Interaction aiDimJSON `json:"d3Interaction"`
|
||
D4Topics aiDimJSON `json:"d4Topics"`
|
||
D5Rhythm aiDimJSON `json:"d5Rhythm"`
|
||
D6Visual aiDimJSON `json:"d6Visual"`
|
||
D7Conversion aiDimJSON `json:"d7Conversion"`
|
||
D8Risk aiDimJSON `json:"d8Risk"`
|
||
}
|
||
|
||
func buildStyleAnalysisPrompt(samples []string, username, sourceLabel string) string {
|
||
var b strings.Builder
|
||
b.WriteString("你是 Threads 寫作風格分析師。根據下列貼文樣本，用繁體中文（台灣用語）完整分析作者的寫作人設。\n")
|
||
if username != "" {
|
||
b.WriteString("對標帳號：@")
|
||
b.WriteString(username)
|
||
b.WriteString("\n")
|
||
}
|
||
if sourceLabel != "" {
|
||
b.WriteString("來源標籤：")
|
||
b.WriteString(sourceLabel)
|
||
b.WriteString("\n")
|
||
}
|
||
b.WriteString("\n【貼文樣本】\n")
|
||
for i, s := range samples {
|
||
b.WriteString(fmt.Sprintf("--- 樣本 %d ---\n%s\n", i+1, s))
|
||
}
|
||
b.WriteString(`
|
||
【輸出要求】
|
||
只輸出一個 JSON 物件（不要 markdown 代碼塊、不要前言）。必須填滿所有欄位，不可留空字串。
|
||
{
|
||
"name": "簡短人設名稱（4～12 字，不要帳號）",
|
||
"brief": "兩到三句定位：是誰、對誰說、核心訊息",
|
||
"identity": "發文身份／角色（一句完整描述，不要只寫帳號名）",
|
||
"tone": "語氣（具體、可模仿）",
|
||
"audience": "主要對誰說話",
|
||
"hooks": "常見開場鉤子手法",
|
||
"language_fingerprint": "用詞／口頭禪／轉折特徵",
|
||
"rhythm": "段落節奏、長短句",
|
||
"punctuation": "標點與 emoji 習慣",
|
||
"content_patterns": "內容結構模式",
|
||
"knowledge_translation": "如何把知識／經驗講成人話",
|
||
"cta_style": "如何邀互動／收尾",
|
||
"examples": "摘一句最像他的原句或改寫示範",
|
||
"avoid": "模仿時要避開的寫法（可用頓號分隔多項）",
|
||
"dimensions": {
|
||
"d1Tone": { "summary": "D1 語氣人格一句話", "evidence": ["樣本短摘"] },
|
||
"d2Structure": { "summary": "D2 結構模板", "evidence": [] },
|
||
"d3Interaction": { "summary": "D3 互動方式", "evidence": [] },
|
||
"d4Topics": { "summary": "D4 主題分布", "evidence": [] },
|
||
"d5Rhythm": { "summary": "D5 發文節奏", "evidence": [] },
|
||
"d6Visual": { "summary": "D6 視覺語法（換行／emoji）", "evidence": [] },
|
||
"d7Conversion": { "summary": "D7 轉換／邀約方式", "evidence": [] },
|
||
"d8Risk": { "summary": "D8 風險紅線", "evidence": [] }
|
||
}
|
||
}
|
||
`)
|
||
return b.String()
|
||
}
|
||
|
||
func collectDimsFromAI(parsed aiStyleJSON) map[string]domain.StyleDimension {
|
||
out := make(map[string]domain.StyleDimension)
|
||
put := func(k string, d aiDimJSON) {
|
||
sum := strings.TrimSpace(d.Summary)
|
||
if sum == "" {
|
||
return
|
||
}
|
||
out[k] = domain.StyleDimension{Summary: sum, Evidence: d.Evidence}
|
||
}
|
||
if parsed.Dimensions != nil {
|
||
for _, k := range dimKeys {
|
||
if d, ok := parsed.Dimensions[k]; ok {
|
||
put(k, d)
|
||
}
|
||
}
|
||
}
|
||
// 扁平後備
|
||
put("d1Tone", parsed.D1Tone)
|
||
put("d2Structure", parsed.D2Structure)
|
||
put("d3Interaction", parsed.D3Interaction)
|
||
put("d4Topics", parsed.D4Topics)
|
||
put("d5Rhythm", parsed.D5Rhythm)
|
||
put("d6Visual", parsed.D6Visual)
|
||
put("d7Conversion", parsed.D7Conversion)
|
||
put("d8Risk", parsed.D8Risk)
|
||
return out
|
||
}
|
||
|
||
// applyAIStyleJSON parses LLM JSON into persona style. Returns false if unusable.
|
||
func applyAIStyleJSON(p *domain.Persona, raw string, samples []string, source, username, sourceLabel string) bool {
|
||
raw = strings.TrimSpace(raw)
|
||
if raw == "" {
|
||
return false
|
||
}
|
||
// strip ```json fences if any
|
||
if i := strings.Index(raw, "{"); i >= 0 {
|
||
if j := strings.LastIndex(raw, "}"); j > i {
|
||
raw = raw[i : j+1]
|
||
}
|
||
}
|
||
var parsed aiStyleJSON
|
||
if err := json.Unmarshal([]byte(raw), &parsed); err != nil {
|
||
return false
|
||
}
|
||
// need at least tone or identity
|
||
if strings.TrimSpace(parsed.Tone) == "" && strings.TrimSpace(parsed.Identity) == "" {
|
||
return false
|
||
}
|
||
coalesce := func(v, fallback string) string {
|
||
v = strings.TrimSpace(v)
|
||
if v == "" {
|
||
return fallback
|
||
}
|
||
return v
|
||
}
|
||
base := p.Style.Draft
|
||
draft := domain.PersonaDraft{
|
||
Identity: coalesce(parsed.Identity, base.Identity),
|
||
Tone: coalesce(parsed.Tone, base.Tone),
|
||
Audience: coalesce(parsed.Audience, base.Audience),
|
||
Hooks: coalesce(parsed.Hooks, base.Hooks),
|
||
LanguageFingerprint: coalesce(parsed.LanguageFingerprint, base.LanguageFingerprint),
|
||
Rhythm: coalesce(parsed.Rhythm, base.Rhythm),
|
||
Punctuation: coalesce(parsed.Punctuation, base.Punctuation),
|
||
ContentPatterns: coalesce(parsed.ContentPatterns, base.ContentPatterns),
|
||
KnowledgeTranslation: coalesce(parsed.KnowledgeTranslation, base.KnowledgeTranslation),
|
||
CtaStyle: coalesce(parsed.CtaStyle, base.CtaStyle),
|
||
Examples: coalesce(parsed.Examples, base.Examples),
|
||
Avoid: coalesce(parsed.Avoid, base.Avoid),
|
||
}
|
||
// reject empty-shell identity like "有趣人設 · 對標 @x"
|
||
if strings.Contains(draft.Identity, "· 對標 @") || strings.Contains(draft.Identity, "· 对标 @") {
|
||
if parsed.Tone != "" {
|
||
draft.Identity = "像 @" + username + " 的發文者：" + parsed.Tone
|
||
}
|
||
}
|
||
// 補齊仍空的 draft 欄位（規則底稿）
|
||
if draft.Audience == "" {
|
||
draft.Audience = "同溫層、會在 Threads 求經驗的人"
|
||
}
|
||
if draft.Hooks == "" {
|
||
draft.Hooks = pickHooks(strings.Join(samples, "\n"))
|
||
}
|
||
if draft.LanguageFingerprint == "" {
|
||
draft.LanguageFingerprint = "短句、口語、少成語"
|
||
}
|
||
if draft.Rhythm == "" {
|
||
draft.Rhythm = "2～4 個短段落，段間空行"
|
||
}
|
||
if draft.Punctuation == "" {
|
||
draft.Punctuation = "逗號與問號為主"
|
||
}
|
||
if draft.ContentPatterns == "" {
|
||
draft.ContentPatterns = "情境 → 感受或觀察 → 輕問一句"
|
||
}
|
||
if draft.KnowledgeTranslation == "" {
|
||
draft.KnowledgeTranslation = "先講使用情境，再講重點"
|
||
}
|
||
if draft.CtaStyle == "" {
|
||
draft.CtaStyle = "自然邀留言，不命令"
|
||
}
|
||
if draft.Avoid == "" {
|
||
draft.Avoid = "硬銷、客服腔、AI 腔"
|
||
}
|
||
if draft.Examples == "" && len(samples) > 0 {
|
||
r := []rune(samples[0])
|
||
if len(r) > 80 {
|
||
draft.Examples = string(r[:80])
|
||
} else {
|
||
draft.Examples = samples[0]
|
||
}
|
||
}
|
||
|
||
dims := collectDimsFromAI(parsed)
|
||
finalizePersonaAnalysis(p, draft, samples, source, username, sourceLabel, parsed.Brief, parsed.Name, dims)
|
||
return true
|
||
}
|