182 lines
4.7 KiB
Go
182 lines
4.7 KiB
Go
package extractor
|
||
|
||
import (
|
||
"unicode"
|
||
"unicode/utf8"
|
||
)
|
||
|
||
type WalkFn func(start, end int) bool
|
||
|
||
type KeywordProcessor struct {
|
||
root *Node
|
||
whiteSpaceChars []string // 不知道有啥用,反正python也定义了,就跟着定义一个吧
|
||
nonWordBoundaries string
|
||
caseSensitive bool // 匹配是否区分大小写
|
||
}
|
||
|
||
func NewKeywordProcessor(caseSensitive bool) *KeywordProcessor {
|
||
return &KeywordProcessor{
|
||
root: newNode(),
|
||
caseSensitive: caseSensitive,
|
||
whiteSpaceChars: []string{".", "\t", "\n", "\a", " ", ","}, // python 就是这样定义的,而且它也没有用到,所以先留着这里吧
|
||
}
|
||
}
|
||
|
||
func (kp *KeywordProcessor) setItem(keyword string) {
|
||
if len(keyword) == 0 {
|
||
return
|
||
}
|
||
|
||
node := kp.root
|
||
for _, char := range keyword {
|
||
if !kp.caseSensitive {
|
||
char = unicode.ToLower(char)
|
||
}
|
||
if _, ok := node.children[char]; !ok {
|
||
// 子节点已存在 将子节点设为当前节点
|
||
node.children[char] = newNode()
|
||
}
|
||
node = node.children[char]
|
||
}
|
||
// 记录当前匹配词的长度
|
||
node.exist[len(keyword)] = struct{}{}
|
||
}
|
||
|
||
func (kp *KeywordProcessor) Build() {
|
||
// 创建一个空队列,用于层级遍历 Trie 树节点
|
||
queue := make([]*Node, 0)
|
||
// 将根节点作为初始节点
|
||
queue = append(queue, kp.root)
|
||
|
||
for len(queue) > 0 {
|
||
// 获取队列中的第一个节点
|
||
currentNode := queue[0]
|
||
// 弹出队列中的第一个节点
|
||
queue = queue[1:]
|
||
|
||
// 遍历当前节点的子节点
|
||
for char, childNode := range currentNode.children {
|
||
// 将子节点添加到队列中
|
||
queue = append(queue, childNode)
|
||
// 当前节点父节点的失败指针
|
||
faFail := currentNode.failure
|
||
|
||
// 递归获取父失败指针 直到获取到根节点为止
|
||
for faFail != nil && faFail.children[char] == nil {
|
||
faFail = faFail.failure
|
||
}
|
||
childNode.failure = kp.root
|
||
if faFail != nil {
|
||
childNode.failure = faFail.children[char]
|
||
}
|
||
for key := range childNode.failure.exist {
|
||
childNode.exist[key] = struct{}{}
|
||
}
|
||
}
|
||
}
|
||
}
|
||
func (kp *KeywordProcessor) AddKeyWord(keyword string) *KeywordProcessor {
|
||
kp.setItem(keyword)
|
||
return kp
|
||
}
|
||
|
||
func (kp *KeywordProcessor) AddKeywordsFromList(keywords []string) *KeywordProcessor {
|
||
for _, keyword := range keywords {
|
||
kp.setItem(keyword)
|
||
}
|
||
return kp
|
||
}
|
||
|
||
func (kp *KeywordProcessor) walk(sentence string, wf WalkFn) {
|
||
// 从根节点开始查找
|
||
currentNode := kp.root
|
||
// 遍历文本 sentence 的每个字符,并记录当前字符的索引为 idx,当前字符为 r
|
||
idx := 0
|
||
for len(sentence) > 0 {
|
||
r, l := utf8.DecodeRuneInString(sentence)
|
||
idx += l
|
||
sentence = sentence[l:]
|
||
if !kp.caseSensitive {
|
||
r = unicode.ToLower(r)
|
||
}
|
||
// 在循环中 判断是否有当前字符子节点
|
||
//如果不存在,则说明匹配失败,需要通过失败路径回溯到前一个节点,直到找到一个匹配的子节点或回溯到根节点。
|
||
for currentNode.children[r] == nil && currentNode.failure != nil {
|
||
currentNode = currentNode.failure
|
||
}
|
||
if currentNode.children[r] == nil {
|
||
continue
|
||
}
|
||
currentNode = currentNode.children[r]
|
||
for length := range currentNode.exist {
|
||
if !wf(idx-length, idx) {
|
||
return
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
func (kp *KeywordProcessor) walkByte(sentence []byte, wf WalkFn) {
|
||
// 从根节点开始查找
|
||
currentNode := kp.root
|
||
// 遍历文本 sentence 的每个字符,并记录当前字符的索引为 idx,当前字符为 r
|
||
idx := 0
|
||
for len(sentence) > 0 {
|
||
r, l := utf8.DecodeRune(sentence)
|
||
idx += l
|
||
sentence = sentence[l:]
|
||
if !kp.caseSensitive {
|
||
r = unicode.ToLower(r)
|
||
}
|
||
// 在循环中 判断是否有当前字符子节点
|
||
//如果不存在,则说明匹配失败,需要通过失败路径回溯到前一个节点,直到找到一个匹配的子节点或回溯到根节点。
|
||
for currentNode.children[r] == nil && currentNode.failure != nil {
|
||
currentNode = currentNode.failure
|
||
}
|
||
if currentNode.children[r] == nil {
|
||
continue
|
||
}
|
||
currentNode = currentNode.children[r]
|
||
for length := range currentNode.exist {
|
||
if !wf(idx-length, idx) {
|
||
return
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
// ExtractKeywords 匹配关键词
|
||
func (kp *KeywordProcessor) ExtractKeywords(sentence string) []Match {
|
||
var matches []Match
|
||
if len(sentence) == 0 {
|
||
return matches
|
||
}
|
||
|
||
kp.walk(sentence, func(start, end int) bool {
|
||
matches = append(matches, Match{
|
||
start: start,
|
||
end: end,
|
||
match: sentence[start:end],
|
||
})
|
||
return true
|
||
})
|
||
return matches
|
||
}
|
||
|
||
func (kp *KeywordProcessor) ExtractKeywordsFromBytes(sentence []byte) []Match {
|
||
var matches []Match
|
||
if len(sentence) == 0 {
|
||
return matches
|
||
}
|
||
|
||
kp.walkByte(sentence, func(start, end int) bool {
|
||
matches = append(matches, Match{
|
||
start: start,
|
||
end: end,
|
||
match: string(sentence[start:end]),
|
||
})
|
||
return true
|
||
})
|
||
return matches
|
||
}
|