text-classifier

2016/12/05之前有数据：

db.msg_overdue.count({text: {$exists: true, $ne: ''}, max_overdue_days: {$exists: true, $gt: 2}})
db.msg_overdue.count({text: {$exists: true, $ne: ''}, max_overdue_days: {$exists: true, $lt: 3}})

pos: 11049
neg: 43681
rate: 0.20188196601498265

---

12/03 - 12/05: 用作测试

db.msg_overdue.count({text: {$exists: true, $ne: ''}, max_overdue_days: {$exists: true, $gt: 2}, creates_time: {$gte: 1480694400}})
db.msg_overdue.count({text: {$exists: true, $ne: ''}, max_overdue_days: {$exists: true, $lt: 3}, creates_time: {$gte: 1480694400}})

pos: 763
neg: 2983
rate: 0.20368392952482647

算法策略

先用special word hit，然后jieba分词，算互信息，然后train