## 文档重构(5 个 md + 1 个 html) - README.md: '尚数网本体管理平台' → '本体元数据平台 · Ontology Platform' - docs/README.md: 通用化定位,补全 P0/P1/P2 (Palantir Ontology 对标) + Materialization/BackingSource/Suggestions 全清单 - docs/PLAN.md: 通用化设计目标,改为 'Palantir Ontology 对标' 路线图 - docs/USAGE.md: 用通用占位符(项目/空间/组织)替换尚数网具体业务举例 - docs/IMPORT.md: 把 shangshu 当成 '一个采用方示例' 展示通用模式 - docs/product-intro.html: title/brand/hero/场景/footer 全改通用,沙数只作典型场景之一 ## 代码字符串同步去 shangshu - server/main.go: '[boot] 尚数网本体管理平台监听' → '[boot] 本体平台监听' - server/handlers_connector.go: manifest 'server' 字段 → 'ontology-platform' - server/handlers_mcp.go: tool description + serverInfo → 通用 - server/handlers_functions.go: builtin interface description → 通用 - server/handlers_backing_source.go: 注释去 shangshuwang.cn 具体网址 - server/config.go + server/db.go: 注释/seed 数据改占位 (demo_self / demo_partner_a/b) - server/ai/anthropic.go: fallback systemPrompt → '通用本体平台工程师' - server/ai/heuristic.go: 未识别 summary → '本体概念' - server/ai/prompt.go: 静态 systemPrompt 头部 + JSON 示例去 shangshu ## 前端文案同步 - web/src/App.vue: title 改 '本体元数据平台' - web/src/views/LoginView.vue: 登录页 h1 改通用 - web/src/views/OntologyView.vue: backing source placeholder 改 example.com - web/src/components/ReviewWorkbench.vue + useReviewWorkbench.js: '尚数网' → '本体平台' ## 验证 - go build ./server OK - 剩余 '尚数网' / 'shangshu-ontology' 引用: - 文档 README/README/USAGE/PLAN/IMPORT 仅在 '采用方示例' 上下文出现 - 代码注释 sswOntologySystemPrompt 等 identifier 保持不变(避免改动 API 表面) - skill 名 'ssw-ontology-reviewer' 保持不变(已发布技能) Refs: docs/README.md (1.1 元数据组件表)、docs/PLAN.md (4 当前已完成能力)
316 lines
9.7 KiB
Go
316 lines
9.7 KiB
Go
package ai
|
||
|
||
import (
|
||
"context"
|
||
"fmt"
|
||
"hash/fnv"
|
||
"regexp"
|
||
"strings"
|
||
"unicode/utf8"
|
||
)
|
||
|
||
// heuristicProvider 不依赖大模型的内置启发式抽取器,保证无 Key 时平台也可运行。
|
||
// 它扫描资料中的「实体名」并结合前后文关键词,将已知概念映射为对象实例,并尽量保持幂等。
|
||
type heuristicProvider struct{}
|
||
|
||
func (h *heuristicProvider) Name() string { return "heuristic" }
|
||
|
||
// Complete 启发式提供者不支持 AI 补全. 返回空字符串 + nil (调用方据此走启发式兜底).
|
||
func (h *heuristicProvider) Complete(ctx context.Context, prompt string) (string, error) {
|
||
return "", nil
|
||
}
|
||
|
||
// CompleteWithSystem 启发式同样返回空字符串.
|
||
func (h *heuristicProvider) CompleteWithSystem(ctx context.Context, system, prompt string) (string, error) {
|
||
return "", nil
|
||
}
|
||
|
||
// 关键词 -> 对象类型(api_name) + 中文名
|
||
var keywordMap = []struct {
|
||
kw string
|
||
apiName string
|
||
display string
|
||
}{
|
||
// 业务对象
|
||
{"数据资源", "DataResource", "数据资源"},
|
||
{"资源", "DataResource", "数据资源"},
|
||
{"数据产品", "DataProduct", "数据产品"},
|
||
{"产品", "DataProduct", "数据产品"},
|
||
{"可信数据空间", "TrustedDataSpace", "可信数据空间"},
|
||
{"数据空间", "TrustedDataSpace", "可信数据空间"},
|
||
{"数据交易所", "DataExchange", "数据交易所"},
|
||
{"交易所", "DataExchange", "数据交易所"},
|
||
{"应用场景", "AppScenario", "应用场景"},
|
||
{"场景", "AppScenario", "应用场景"},
|
||
{"分支机构", "Branch", "分支机构"},
|
||
{"分公司", "Branch", "分支机构"},
|
||
{"总部", "Branch", "分支机构"},
|
||
{"会员单位", "MemberOrg", "会员单位"},
|
||
{"会员", "MemberOrg", "会员单位"},
|
||
// 硬件 / 一体机
|
||
{"密算一体机", "Appliance", "密算一体机"},
|
||
{"一体机", "Appliance", "密算一体机"},
|
||
// 服务
|
||
{"核心服务", "Service", "核心服务"},
|
||
{"服务", "Service", "核心服务"},
|
||
// 知识产权
|
||
{"专利", "IPAsset", "知识产权"},
|
||
{"软著", "IPAsset", "知识产权"},
|
||
{"商标", "IPAsset", "知识产权"},
|
||
{"标准", "IPAsset", "知识产权"},
|
||
{"认证", "IPAsset", "知识产权"},
|
||
}
|
||
|
||
// bracketRe 匹配 「x」 “x” ‘x’ (x) (x) 中的实体名
|
||
var bracketRe = regexp.MustCompile(`[「“'((]([^」”'))]+)[」”'))]`)
|
||
|
||
func (h *heuristicProvider) Suggest(ctx context.Context, text, directive, existingContext string) (*Suggestion, error) {
|
||
s := &Suggestion{}
|
||
seen := map[string]bool{}
|
||
|
||
directiveType := ""
|
||
for _, km := range keywordMap {
|
||
if strings.Contains(directive, km.kw) {
|
||
directiveType = km.apiName
|
||
break
|
||
}
|
||
}
|
||
|
||
// 1) 提取括号/书名号包裹的实体名,并在其“前文小窗口”中判定类型。
|
||
// 业务文本中类型关键词通常紧邻实体名之前(如“会员单位新增「X」”),故只取实体名前 N 个字符,
|
||
// 并用 LastIndex 取离名字最近的关键词,避免远处“数据产品”等泛化词干扰。
|
||
const beforeCtx = 16
|
||
matches := bracketRe.FindAllStringSubmatchIndex(text, -1)
|
||
for _, m := range matches {
|
||
name := strings.TrimSpace(text[m[2]:m[3]])
|
||
if len([]rune(name)) > 40 || name == "" {
|
||
continue
|
||
}
|
||
nameRuneInText := utf8.RuneCountInString(text[:m[2]])
|
||
runes := []rune(text)
|
||
bs := nameRuneInText - beforeCtx
|
||
if bs < 0 {
|
||
bs = 0
|
||
}
|
||
before := string(runes[bs:nameRuneInText])
|
||
|
||
typ := detectType(before+name, directiveType)
|
||
if typ == "" {
|
||
continue
|
||
}
|
||
key := typ + "/" + name
|
||
if seen[key] {
|
||
continue
|
||
}
|
||
seen[key] = true
|
||
s.Objects = append(s.Objects, SuggestedObject{
|
||
Type: typ,
|
||
PK: objPK(typ, name),
|
||
Properties: map[string]interface{}{"name": name},
|
||
})
|
||
}
|
||
|
||
// 2) 兜底:未匹配到括号时,按 "关键词:名称" 行式结构抽取
|
||
if len(s.Objects) == 0 {
|
||
for _, line := range strings.Split(text, "\n") {
|
||
line = strings.TrimSpace(line)
|
||
typ := detectType(line, directiveType)
|
||
if typ == "" {
|
||
continue
|
||
}
|
||
name := extractName(line)
|
||
if name == "" {
|
||
continue
|
||
}
|
||
key := typ + "/" + name
|
||
if seen[key] {
|
||
continue
|
||
}
|
||
seen[key] = true
|
||
s.Objects = append(s.Objects, SuggestedObject{
|
||
Type: typ,
|
||
PK: objPK(typ, name),
|
||
Properties: map[string]interface{}{"name": name},
|
||
})
|
||
}
|
||
}
|
||
|
||
// 3) 兜底:在整段文本中寻找首个类型关键词 + 紧邻的实体名(如“密算一体机 A5 卡”)。
|
||
// 仅在前两步无任何命中时启用,避免与目录化文本重复。
|
||
if len(s.Objects) == 0 {
|
||
// 按 keywordMap 顺序遍历,每个 kw 仅取首次命中(保证多样性)
|
||
usedKinds := map[string]bool{}
|
||
for _, km := range keywordMap {
|
||
if directiveType != "" && km.apiName != directiveType {
|
||
continue
|
||
}
|
||
if usedKinds[km.apiName] {
|
||
continue
|
||
}
|
||
idx := strings.Index(text, km.kw)
|
||
if idx < 0 {
|
||
continue
|
||
}
|
||
rest := text[idx+len(km.kw):]
|
||
name := pickFirstName(rest)
|
||
if name == "" {
|
||
continue
|
||
}
|
||
usedKinds[km.apiName] = true
|
||
key := km.apiName + "/" + name
|
||
if seen[key] {
|
||
continue
|
||
}
|
||
seen[key] = true
|
||
s.Objects = append(s.Objects, SuggestedObject{
|
||
Type: km.apiName,
|
||
PK: objPK(km.apiName, name),
|
||
Properties: map[string]interface{}{"name": name},
|
||
})
|
||
}
|
||
}
|
||
|
||
// 4) 兜底:资料整体命中多种类型关键词时,每类至少落 1 条带原始资料的实例(用于让用户看到提取痕迹)。
|
||
if len(s.Objects) == 0 {
|
||
fallbackTypes := map[string]string{
|
||
"密算一体机": "Appliance",
|
||
"可信数据空间": "TrustedDataSpace",
|
||
"数据资源": "DataResource",
|
||
"数据产品": "DataProduct",
|
||
"应用场景": "AppScenario",
|
||
"核心服务": "Service",
|
||
"知识产权": "IPAsset",
|
||
}
|
||
for kw, api := range fallbackTypes {
|
||
if directiveType != "" && api != directiveType {
|
||
continue
|
||
}
|
||
if !strings.Contains(text, kw) {
|
||
continue
|
||
}
|
||
name := kw + "(未识别具体实例)"
|
||
key := api + "/" + name
|
||
if seen[key] {
|
||
continue
|
||
}
|
||
seen[key] = true
|
||
s.Objects = append(s.Objects, SuggestedObject{
|
||
Type: api,
|
||
PK: objPK(api, name),
|
||
Properties: map[string]interface{}{"name": name, "raw_excerpt": truncate(text, 120)},
|
||
})
|
||
}
|
||
}
|
||
|
||
if len(s.Objects) == 0 {
|
||
s.Summary = "启发式抽取未从资料中发现可识别的本体概念;建议配置大模型(AI_ENABLED=true)以获得更精确的抽取与关系识别。"
|
||
} else {
|
||
s.Summary = fmt.Sprintf("启发式抽取识别到 %d 个候选实例(基于关键词+实体名映射)。如需更精准的结构化抽取与关系识别,请配置 AI_ENABLED=true 并提供大模型 Key。", len(s.Objects))
|
||
}
|
||
return s, nil
|
||
}
|
||
|
||
func truncate(s string, n int) string {
|
||
if utf8.RuneCountInString(s) <= n {
|
||
return s
|
||
}
|
||
r := []rune(s)
|
||
return string(r[:n]) + "..."
|
||
}
|
||
|
||
// detectType 在给定文本窗口中查找首个匹配的关键词,返回对应对象类型。
|
||
// detectType 在给定的前文窗口中查找关键词,返回离窗口末尾(即离实体名最近)的那个关键词对应的对象类型。
|
||
func detectType(window string, directiveType string) string {
|
||
best := ""
|
||
bestPos := -1
|
||
for _, km := range keywordMap {
|
||
if directiveType != "" && km.apiName != directiveType {
|
||
continue
|
||
}
|
||
// LastIndex 取最靠近窗口末尾(=最靠近实体名)的一次出现
|
||
idx := strings.LastIndex(window, km.kw)
|
||
if idx < 0 {
|
||
continue
|
||
}
|
||
if idx > bestPos {
|
||
bestPos = idx
|
||
best = km.apiName
|
||
}
|
||
}
|
||
return best
|
||
}
|
||
|
||
// extractName 从一行文本中提取实体名称:取冒号/空格后、括号前的内容。
|
||
func extractName(line string) string {
|
||
line = strings.TrimLeft(line, "-*•· \t#")
|
||
line = strings.TrimSpace(line)
|
||
re := regexp.MustCompile(`[::]`)
|
||
parts := re.Split(line, 2)
|
||
body := parts[0]
|
||
if len(parts) > 1 && strings.TrimSpace(parts[1]) != "" && len([]rune(parts[1])) < len([]rune(parts[0])) {
|
||
body = parts[1]
|
||
}
|
||
body = strings.TrimSpace(body)
|
||
if idx := strings.IndexAny(body, "(("); idx >= 0 {
|
||
body = strings.TrimSpace(body[:idx])
|
||
}
|
||
body = strings.TrimRight(body, "。,,;;、")
|
||
body = strings.TrimSpace(body)
|
||
if len([]rune(body)) > 40 {
|
||
return ""
|
||
}
|
||
return body
|
||
}
|
||
|
||
// hasTypeKeyword 在文本中查找任意 keywordMap 关键词命中,返回 apiName。
|
||
// 若同时指定 directiveType(不为空)则只匹配该 apiName。
|
||
func hasTypeKeyword(text, directiveType string) string {
|
||
for _, km := range keywordMap {
|
||
if directiveType != "" && km.apiName != directiveType {
|
||
continue
|
||
}
|
||
if strings.Contains(text, km.kw) {
|
||
return km.apiName
|
||
}
|
||
}
|
||
return ""
|
||
}
|
||
|
||
// pickFirstName 在 text 中提取第一个像实体名的词(≥2 字汉字或纯中文短语),用于无括号、无换行的整段文本兜底。
|
||
func pickFirstName(text string) string {
|
||
rest := text
|
||
if len([]rune(rest)) > 24 {
|
||
rest = string([]rune(rest)[:24])
|
||
}
|
||
re := regexp.MustCompile(`[\p{Han}]{2,40}`)
|
||
m := re.FindString(rest)
|
||
if m == "" {
|
||
return ""
|
||
}
|
||
return m
|
||
}
|
||
|
||
// objPK 生成对象实例主键:类型前缀 + 名称的 FNV-64 哈希,保证
|
||
// (a) 仅含 ASCII、URL 安全;(b) 同一名称稳定且跨次摄取幂等;
|
||
// (c) 同类型不同名称不会像单纯 slug 那样塌缩成同一个主键。
|
||
func objPK(typ, name string) string {
|
||
h := fnv.New64a()
|
||
h.Write([]byte(name))
|
||
return slug(typ) + "_" + fmt.Sprintf("%016x", h.Sum64())
|
||
}
|
||
|
||
// slug 生成英文 snake_case 主键,保证主键合法且尽量唯一。
|
||
func slug(s string) string {
|
||
s = strings.TrimSpace(s)
|
||
s = strings.ToLower(s)
|
||
repl := strings.NewReplacer(" ", "_", "(", "_", ")", "_", "(", "_", ")", "_", "/", "_", "-", "_", ":", "_", ":", "_", "「", "", "」", "", "“", "", "”", "", "'", "", "(", "", ")", "")
|
||
s = repl.Replace(s)
|
||
re := regexp.MustCompile(`[^a-z0-9_]+`)
|
||
s = re.ReplaceAllString(s, "_")
|
||
s = strings.Trim(s, "_")
|
||
if s == "" {
|
||
s = "item"
|
||
}
|
||
return s
|
||
}
|