1
0
mirror of https://github.com/fumiama/jieba.git synced 2026-06-05 00:32:51 +08:00

added simple implemention of textrank, corresponding to jieba commit #4030d8ed86dd3ff54e215ebe88c141b2a8345eda

This commit is contained in:
Wang Bin
2015-02-04 18:23:47 +08:00
parent 70a8da2687
commit 74dbb7d525
3 changed files with 166 additions and 4 deletions

View File

@@ -1,7 +1,7 @@
package analyse
import (
// "fmt"
"fmt"
"github.com/wangbin/jiebago"
"sort"
"strings"
@@ -13,9 +13,9 @@ type TfIdf struct {
Freq float64
}
//func (t TfIdf) String() string {
// return fmt.Sprintf("{%s: %f}", t.Word, t.Freq)
//}
func (t TfIdf) String() string {
return fmt.Sprintf("{%s: %f}", t.Word, t.Freq)
}
type TfIdfs []TfIdf

126
analyse/textrank.go Normal file
View File

@@ -0,0 +1,126 @@
package analyse
import (
mapset "github.com/deckarep/golang-set"
"github.com/wangbin/jiebago/posseg"
"math"
"sort"
)
const (
DampingFactor = 0.85
)
var (
defaultAllowPOS = []string{"ns", "n", "vn", "v"}
)
type edge struct {
start string
end string
weight float64
}
type undirectWeightedGraph struct {
graph map[string][]edge
}
func (u *undirectWeightedGraph) addEdge(start, end string, weight float64) {
if _, ok := u.graph[start]; !ok {
u.graph[start] = []edge{edge{start: start, end: end, weight: weight}}
} else {
u.graph[start] = append(u.graph[start], edge{start: start, end: end, weight: weight})
}
if _, ok := u.graph[end]; !ok {
u.graph[start] = []edge{edge{start: end, end: start, weight: weight}}
} else {
u.graph[start] = append(u.graph[start], edge{start: end, end: start, weight: weight})
}
}
func (u *undirectWeightedGraph) rank() TfIdfs {
ws := make(map[string]float64)
outSum := make(map[string]float64)
wsdef := 1.0
if len(u.graph) > 0 {
wsdef /= float64(len(u.graph))
}
for n, out := range u.graph {
ws[n] = wsdef
sum := 0.0
for _, e := range out {
sum += e.weight
}
outSum[n] = sum
}
for x := 0; x < 10; x++ {
for n, inedges := range u.graph {
s := 0.0
for _, e := range inedges {
s += e.weight / outSum[e.end] * ws[e.start]
}
ws[n] = (1 - DampingFactor) + DampingFactor*s
}
}
minRank := math.MaxFloat64
maxRank := math.SmallestNonzeroFloat64
for _, w := range ws {
if w < minRank {
minRank = w
} else if w > maxRank {
maxRank = w
}
}
result := make(TfIdfs, 0)
for n, w := range ws {
result = append(result, TfIdf{Word: n, Freq: (w - minRank/10.0) / (maxRank - minRank/10.0)})
}
sort.Sort(sort.Reverse(result))
return result
}
func TextRankWithPOS(sentence string, topK int, allowPOS []string) TfIdfs {
posFilt := mapset.NewSet()
for _, pos := range allowPOS {
posFilt.Add(pos)
}
g := new(undirectWeightedGraph)
cm := make(map[[2]string]float64)
span := 5
wordTags := posseg.Cut(sentence, true)
for i := range wordTags {
if posFilt.Contains(wordTags[i].Tag) {
for j := i + 1; j < i+span; i++ {
if j > len(wordTags) {
break
}
if !posFilt.Contains(wordTags[j].Tag) {
continue
}
if _, ok := cm[[2]string{wordTags[i].Word, wordTags[j].Word}]; !ok {
cm[[2]string{wordTags[i].Word, wordTags[j].Word}] = 1.0
} else {
cm[[2]string{wordTags[i].Word, wordTags[j].Word}] += 1.0
}
}
}
}
for startEnd, weight := range cm {
g.addEdge(startEnd[0], startEnd[1], weight)
}
tags := g.rank()
if topK > 0 && len(tags) > topK {
tags = tags[:topK]
}
return tags
}
func TextRank(sentence string, topK int) TfIdfs {
return TextRankWithPOS(sentence, topK, defaultAllowPOS)
}

36
analyse/textrank_test.go Normal file
View File

@@ -0,0 +1,36 @@
package analyse
import (
"github.com/wangbin/jiebago"
"math"
"testing"
)
var (
sentence = "此外公司拟对全资子公司吉林欧亚置业有限公司增资4.3亿元增资后吉林欧亚置业注册资本由7000万元增加到5亿元。吉林欧亚置业主要经营范围为房地产开发及百货零售等业务。目前在建吉林欧亚城市商业综合体项目。2013年实现营业收入0万元实现净利润-139.13万元。"
tagRanks = TfIdfs{
TfIdf{Word: "吉林", Freq: 1.000000},
TfIdf{Word: "欧亚", Freq: 0.864834},
TfIdf{Word: "置业", Freq: 0.553466},
TfIdf{Word: "实现", Freq: 0.520661},
TfIdf{Word: "收入", Freq: 0.379700},
TfIdf{Word: "增资", Freq: 0.355086},
TfIdf{Word: "子公司", Freq: 0.349758},
TfIdf{Word: "全资", Freq: 0.308537},
TfIdf{Word: "城市", Freq: 0.306104},
TfIdf{Word: "商业", Freq: 0.304837},
}
)
func TesTextRank(t *testing.T) {
jiebago.SetDictionary("../dict.txt")
SetIdf("idf.txt")
results := TextRank(sentence, 10)
for index, tw := range results {
if tw.Word != tagRanks[index].Word || math.Abs(tw.Freq-tagRanks[index].Freq) > 1e-6 {
t.Errorf("%v != %v", tw, tagRanks[index])
}
}
}