Simhash的算法简单的来说就是,从海量文本中快速搜索和已知simhash相差小于k位的simhash集合,这里每个文本都可以用一个simhash值来代表,一个simhash有64bit,相似的文本,64bit也相似,论文中k的经验值为3。该方法的缺点如优点一样明显,主要有两点,对于短文本,k值很敏感;另一个是由于算法是以空间换时间,系统内存吃不消。
demo:
package main import (
"strings"
"math/big"
"fmt"
) type SimHash struct {
IntSimHash *big.Int
HashBits int
} func main() {
str := "有理数是“数与代数”领域中的重要内容之一,是数、代数式、方程、不等式、直角坐标系、函数、统计等数学内容以及相关学科知识的基础"
str2 := "有理数是“数与代数”领域中的重要内容之一,是继续学习实数、代数式、方程、不等式、直角坐标系、函数、统计等数学内容以及相关学科知识的基础"
//str3:="nai nai ge xiong cao"
s := params()
//str hash 值
hash := s.Simhash(str)
fmt.Println(hash)
////str2 距离
hash2 := s.Simhash(str2)
fmt.Println(hash2) ////距离
ts := s.HammingDistance(hash, hash2)
fmt.Println(ts)
////计算相似度
sm := s.Similarity(hash, hash2)
fmt.Println(sm) } /**
距离 补偿
*/
func (s *SimHash) HammingDistance(hash, other *big.Int) *big.Int {
hase_v := new(big.Int)
c_w := hase_v.Xor(hash, other)
fbIng := big.NewInt(1) fbIng.Lsh(fbIng, uint(s.HashBits))
bit_big := new(big.Int)
t_mak := bit_big.Sub(fbIng, big.NewInt(1))
c_result := new(big.Int)
c_result.And(c_w, t_mak)
//fmt.Println(c_result) tot := big.NewInt(0)
for c_result.Cmp(big.NewInt(0)) > 0 {
tot.Add(tot, big.NewInt(1))
ts := new(big.Int)
ts.Sub(c_result, big.NewInt(1))
c_result.And(c_result, ts)
}
return tot
} /**
相似度
*/
func (s *SimHash) Similarity(hash, other *big.Int) float64 {
a := new(big.Rat)
a.SetInt(hash)
b := new(big.Rat)
b.SetInt(other)
val := new(big.Rat)
if a.Cmp(b) > 0 {
val.Quo(b, a)
f, _ := val.Float64()
return f
}
val.Quo(a, b)
f, _ := val.Float64()
return f } /**
海明距离hash
*/
func (s *SimHash) Simhash(str string) *big.Int {
m := strings.Split(str, ",") token_int := make([]int, s.HashBits)
for i := 0; i < len(m); i++ {
temp := m[i]
t := s.Hash(temp)
//fmt.Println(t)
for j := 0; j < s.HashBits; j++ {
fbIng := big.NewInt(1)
bitMask := fbIng.Lsh(fbIng, uint(j))
statusBig := new(big.Int)
statusBig.And(t, bitMask)
if statusBig.Cmp(big.NewInt(0)) != 0 {
token_int[j] += 1
} else {
token_int[j] -= 1
}
} }
fingerprint := big.NewInt(0)
for i := 0; i < s.HashBits; i++ {
if token_int[i] >= 0 {
oneBig := big.NewInt(1)
tbig := big.NewInt(0)
fingerprint.Add(fingerprint, tbig.Lsh(oneBig, uint(i)))
}
}
return fingerprint
} /**
初始化
*/
func params() (s *SimHash) {
s = &SimHash{}
s.HashBits = 128
return s
} /**
hash 值
*/
func (s *SimHash) Hash(token string) *big.Int {
if token == "" {
return big.NewInt(0)
} else {
//fmt.Println(token)
bigIntToken := big.NewInt(int64(token[0]))
funit := new(big.Int)
x := funit.Lsh(bigIntToken, 7)
m := big.NewInt(1000003)
mslB := big.NewInt(1)
mask := mslB.Lsh(mslB, uint(s.HashBits))
tsk_b := mask.Sub(mask, big.NewInt(1))
for i := 0; i < len(token); i++ {
tokens := big.NewInt(int64(token[i]))
x.Mul(x, m)
x.Xor(x, tokens)
x.And(x, tsk_b)
}
x = x.Xor(x, big.NewInt(int64(len(token))))
if x == big.NewInt(-1) {
x = big.NewInt(-2)
}
return x
}
}