gse

Go efficient multilingual NLP and text segmentation; support English, Chinese, Japanese and others. And supports with riot, zedsearch, zincsearch, elasticsearch and bleve.

简体中文 | 日本語

Gse is implements jieba by golang, and added NLP support and more features

I build Codg now, Easy code and work AI agent system: auto, asynchronous, concurrency, efficiently and High accuracy

Feature:

• Support common, search engine, full mode, precise mode and HMM mode multiple word segmentation modes;
• Support user and embed dictionary, Part-of-speech/POS tagging, analyze segment info, stop and trim words
• Support multilingual: English, Chinese, Japanese and others
• Support Traditional Chinese (Use single dictionary)
• Support HMM cut text use Viterbi algorithm
• Support NLP by Pytorch (in work)
• Named Entity Recognition (in work)
• Supports riot and zedsearch, and elasticsearch, bleve and others
• run JSON RPC service.

Algorithm:

• Dictionary with the double array trie (Double-Array Trie, cedar) by default, or a rune-level hash trie with -tags hash
• Segmenter algorithm is the shortest path (based on word frequency and dynamic programming), and DAG and HMM algorithm word segmentation.

Text Segmentation speed:

• single thread 9.2MB/s
• goroutines concurrent 26.8MB/s.
• HMM text segmentation single thread 3.2MB/s. (2core 4threads Macbook Pro).

Binding:

gse-bind, binding JavaScript and other, support more language.

Install / update

With Go module support (Go 1.11+), just import:

import "github.com/go-ego/gse"

Otherwise, to install the gse package, run the command:

go get -u github.com/go-ego/gse

Build tags

• -tags hash: use a rune-level hash trie instead of the default cedar double array trie; ~8 MiB (~40%) less trie memory for the zh dict, but ~10% slower segmentation.
• -tags ne: do not embed the default dictionaries into the binary (smaller size); load them with LoadDict("your dict path") instead of LoadDictEmbed().

go build -tags hash
go build -tags ne

Low memory devices

The zh dictionary takes about 132 MiB of heap by default. For low memory devices, combine:

• -tags hash: the hash trie, about 8 MiB less.
• seg.SkipSubSeg = true before loading: skips the precomputed sub-words of the search mode (Slice, String, Pos with searchMode), about 50 MiB less. CutSearch(text) uses them too, CutSearch(text, true) and CutSearch(text, false) still work.
• seg.LoadDict("zh_s"): only the simplified Chinese dictionary.

// go build -tags hash
var seg gse.Segmenter
seg.SkipSubSeg = true
err := seg.LoadDict("zh_s") // about 53 MiB

Lint

The code is checked by golangci-lint v2 in CI, with the config in .golangci.yml:

go install github.com/golangci/golangci-lint/v2/cmd/golangci-lint@latest
golangci-lint run ./...
golangci-lint run --build-tags hash .

Use

package main

import (
	_ "embed"
	"fmt"

	"github.com/go-ego/gse"
)

//go:embed testdata/test_en2.txt
var testDict string

//go:embed testdata/test_en.txt
var testEn string

var (
	text  = "To be or not to be, that's the question!"
	test1 = "Hiworld, Helloworld!"
)

func main() {
	// keep the origin capital letter
	// gse.ToLower = false

	var seg1 gse.Segmenter
	seg1.DictSep = ","
	err := seg1.LoadDict("./testdata/test_en.txt")
	if err != nil {
		fmt.Println("Load dictionary error: ", err)
	}

	s1 := seg1.Cut(text)
	fmt.Println("seg1 Cut: ", s1)
	// seg1 Cut:  [to be   or   not to be ,   that's the question!]

	var seg2 gse.Segmenter
	seg2.AlphaNum = true
	seg2.LoadDict("./testdata/test_en_dict3.txt")

	s2 := seg2.Cut(test1)
	fmt.Println("seg2 Cut: ", s2)
	// seg2 Cut:  [hi world ,   hello world !]

	var seg3 gse.Segmenter
	seg3.AlphaNum = true
	seg3.DictSep = ","
	err = seg3.LoadDictEmbed(testDict + "\n" + testEn)
	if err != nil {
		fmt.Println("loadDictEmbed error: ", err)
	}
	s3 := seg3.Cut(text + test1)
	fmt.Println("seg3 Cut: ", s3)
	// seg3 Cut:  [to be   or   not to be ,   that's the question! hi world ,   hello world !]

	// example2()
}

Example2:

package main

import (
	"fmt"
	"regexp"

	"github.com/go-ego/gse"
	"github.com/go-ego/gse/hmm/pos"
)

var (
	text = "Hello world, Helloworld. Winter is coming! こんにちは世界, 你好世界."

	new, _ = gse.New("zh,testdata/test_en_dict3.txt", "alpha")

	seg gse.Segmenter
	posSeg pos.Segmenter
)

func main() {
	// Loading the default dictionary
	seg.LoadDict()
	// Loading the default dictionary with embed
	// seg.LoadDictEmbed()
	//
	// Loading the Simplified Chinese dictionary
	// seg.LoadDict("zh_s")
	// seg.LoadDictEmbed("zh_s")
	//
	// Loading the Traditional Chinese dictionary
	// seg.LoadDict("zh_t")
	//
	// Loading the Japanese dictionary
	// seg.LoadDict("jp")
	//
	// Load the dictionary
	// seg.LoadDict("your gopath"+"/src/github.com/go-ego/gse/data/dict/dictionary.txt")

	cut()

	segCut()
}

func cut() {
	hmm := new.Cut(text, true)
	fmt.Println("cut use hmm: ", hmm)

	hmm = new.CutSearch(text, true)
	fmt.Println("cut search use hmm: ", hmm)
	fmt.Println("analyze: ", new.Analyze(hmm, text))

	hmm = new.CutAll(text)
	fmt.Println("cut all: ", hmm)

	reg := regexp.MustCompile(`(\d+年|\d+月|\d+日|[\p{Latin}]+|[\p{Hangul}]+|\d+\.\d+|[a-zA-Z0-9]+)`)
	text1 := `헬로월드 헬로 서울, 2021年09月10日, 3.14`
	hmm = seg.CutDAG(text1, reg)
	fmt.Println("Cut with hmm and regexp: ", hmm, hmm[0], hmm[6])
}

func analyzeAndTrim(cut []string) {
	a := seg.Analyze(cut, "")
	fmt.Println("analyze the segment: ", a)

	cut = seg.Trim(cut)
	fmt.Println("cut all: ", cut)

	fmt.Println(seg.String(text, true))
	fmt.Println(seg.Slice(text, true))
}

func cutPos() {
	po := seg.Pos(text, true)
	fmt.Println("pos: ", po)
	po = seg.TrimPos(po)
	fmt.Println("trim pos: ", po)

	pos.WithGse(seg)
	po = posSeg.Cut(text, true)
	fmt.Println("pos: ", po)

	po = posSeg.TrimWithPos(po, "zg")
	fmt.Println("trim pos: ", po)
}

func segCut() {
	// Text Segmentation
	tb := []byte(text)
	fmt.Println(seg.String(text, true))

	segments := seg.Segment(tb)
	// Handle word segmentation results, search mode
	fmt.Println(gse.ToString(segments, true))
}

Look at an custom dictionary example

package main

import (
	"fmt"
	_ "embed"

	"github.com/go-ego/gse"
)

//go:embed test_en_dict3.txt
var testDict string

func main() {
	// var seg gse.Segmenter
	// seg.LoadDict("zh, testdata/zh/test_dict.txt, testdata/zh/test_dict1.txt")
	// seg.LoadStop()
	seg, err := gse.NewEmbed("zh, word 20 n"+testDict, "en")
	// seg.LoadDictEmbed()
	seg.LoadStopEmbed()

	text1 := "Hello world, こんにちは世界, 你好世界!"
	s1 := seg.Cut(text1, true)
	fmt.Println(s1)
	fmt.Println("trim: ", seg.Trim(s1))
	fmt.Println("stop: ", seg.Stop(s1))
	fmt.Println(seg.String(text1, true))

	segments := seg.Segment([]byte(text1))
	fmt.Println(gse.ToString(segments))
}

Look at an Chinese example

Look at an Japanese example

Elasticsearch

How to use it with elasticsearch?

go-gse-elastic

Authors

• Maintainers
• Contributors

License

Gse is primarily distributed under the terms of "the Apache License (Version 2.0)". See LICENSE-APACHE, LICENSE.

Thanks for jieba(jiebago) and sego.