package extract
import (
"data_ai/clean"
"data_ai/prompt"
"data_ai/ul"
log "github.com/donnie4w/go-logger/logger"
qu "jygit.jydev.jianyu360.cn/data_processing/common_utils"
"strings"
"sync"
"unicode/utf8"
)
// 识别结构化字段
func ExtractFieldInfo(sid string, eid string) {
q := map[string]interface{}{
"_id": map[string]interface{}{
"$gt": ul.StringTOBsonId(sid),
"$lte": ul.StringTOBsonId(eid),
},
}
//先查询抽取表-确定大模型需要识别到范围
dict := ConfrimExtractInfo(q)
log.Debug("查询语句...", q, "~", len(dict))
if len(dict) >= ul.MaxUdp { //根据数量限制使用具体模型
log.Debug("数量超过限制临时使用:glm-4-flashx")
ul.FlashModel = "glm-4-flashx"
} else {
ul.FlashModel = "glm-4-flash"
}
pool_mgo := make(chan bool, ul.Reading)
wg_mgo := &sync.WaitGroup{}
sess := ul.BidMgo.GetMgoConn()
defer ul.BidMgo.DestoryMongoConn(sess)
total, isok := 0, 0
it := sess.DB(ul.BidMgo.DbName).C(ul.Bid_Name).Find(&q).Sort("_id").Iter()
for tmp := make(map[string]interface{}); it.Next(&tmp); total++ {
if total%200 == 0 {
log.Debug("cur ai index ", total)
}
tmpid := ul.BsonTOStringId(tmp["_id"])
infoformat := qu.IntAll(tmp["infoformat"])
if infoformat > 1 || dict[tmpid] == nil {
tmp = make(map[string]interface{})
continue
}
isok++
pool_mgo <- true
wg_mgo.Add(1)
go func(tmp map[string]interface{}) {
defer func() {
<-pool_mgo
wg_mgo.Done()
}()
u_id := ul.BsonTOStringId(tmp["_id"])
data := ResolveInfo(tmp)
if len(data) > 0 || u_id == "" {
ul.SourceMgo.UpdateById(ul.Ext_Name, u_id, map[string]interface{}{
"$set": map[string]interface{}{"ai_zhipu": data},
})
}
}(tmp)
tmp = make(map[string]interface{})
}
wg_mgo.Wait()
log.Debug("ai is over ...", sid, "~", eid)
}
// 获取处理数据...
func ResolveInfo(v map[string]interface{}) map[string]interface{} {
tmpid := ul.BsonTOStringId(v["_id"])
title := qu.ObjToString(v["title"])
old_detail := getDetailText(v, tmpid) //获取正文文本
isTable := false //是否表格
if strings.Contains(old_detail, "
") {
isTable = true
}
if NotInProgressInfo(title, old_detail, v) { //过滤信息
return map[string]interface{}{}
}
//识别结构,短文本结构
f_data, shorText := map[string]interface{}{}, false
if utf8.RuneCountInString(old_detail) < 100 {
shorText = true
}
//文本格式转换
new_detail := ul.HttpConvertToMarkdown(old_detail)
//短文本判断是否有效性
if shorText {
if info := prompt.AcquireJudgeShortInfo(new_detail); info["结果"] != "是" {
return map[string]interface{}{}
}
}
//获取外围字段数据-拆分合并字段
f_info_1 := prompt.AcquireExtractFieldInfoFirst(new_detail)
f_info_2 := prompt.AcquireExtractFieldInfoSecond(new_detail)
f_info_3 := prompt.AcquireExtractFieldInfoThird(new_detail)
f_info := MergeInfo([]map[string]interface{}{f_info_1, f_info_2, f_info_3})
//非短文本以下识别-纯测试
if !shorText {
//获取分包信息
if pkg := prompt.AcquireNewMultiplePackageInfo(new_detail, isTable); len(pkg) > 0 {
f_info["s_pkg"] = pkg
}
//获取分类字段数据
s_toptype, s_subtype := prompt.AcquireClassInfo(new_detail, title, qu.ObjToString(v["toptype"]))
f_info["s_toptype"] = s_toptype
f_info["s_subtype"] = s_subtype
//调用标的物识别
if !ul.IsTool {
if s_purchasinglist := getPurList(v, old_detail, f_info); len(s_purchasinglist) > 0 {
f_info["s_purchasinglist"] = s_purchasinglist
}
}
}
//字段清洗
fns := getpnsinfo(v) //获取附件名字
f_data = clean.CleanFieldInfo(f_info, fns, isTable)
//采购单位二级校验
CheckOutBuyerInfo(f_data)
//强制逻辑判断-
ForcedLogicDecideInfo(f_data)
//返回数据
return f_data
}
// 暂时不启用...无限重试
func RunResetUpdateFieldInfo(arr []string, name string, s_name string) {
//log.Debug("开始重置更新...", len(arr))
//reset := []string{}
//for k, v := range arr {
// log.Debug("...", k, "...", v)
// data := ul.SourceMgo.FindById(name, v)
// content := PromptFieldText(qu.ObjToString(data["detail"]))
// zp, ok := map[string]interface{}{}, 0
// for {
// ok++
// if zp = ai.PostZhiPuAI(content); len(zp) > 0 {
// break
// }
// if ok >= 5 {
// log.Debug("请求数据失败...", v)
// reset = append(reset, v)
// break
// }
// }
// ul.SourceMgo.UpdateById(s_name, v, map[string]interface{}{
// "$set": map[string]interface{}{
// "zhipu": zp,
// },
// })
//}
//if len(reset) > 0 { //无限尝试
// RunResetUpdateFieldInfo(reset, name, s_name)
//}
//log.Debug("本轮重置更新结束......")
}