Natural Language Processing (NLP) Tokenization Libary designed for English. Fast, Lean, Customizable. Tokenizes text, replaces abbreviations, replaces contractions, lowercases words, optionally you can remove stop words as well (must specify, see usage). This library is a work in progress but all features mentioned in README should be working as advertised.
Tokenizing text is one of the first steps in NLP, this preformant library should help Go users get started with their NLP tasks.
- Convert Text to Tokens
- Replace Contractions
- Replace Abbreviations
- Remove Stop Words - defaults to keeping them
- Blazingly Fast
- Low Allocation
- Practically zero dependency (only dependency is testify for unit testing)
Coming soon:
- Streamed Reading
- Sentence Tokenization
go get -u github.com/rekram1-node/tokenizerFor detailed examples see examples
package main
import (
"fmt"
"strings"
"github.com/rekram1-node/tokenizer/tokenizer"
)
func main() {
myStr := "This is my long string! I can replace contractions like can't or they've! I can replace abbreviations such as: demonstr. or jan."
t := tokenizer.New()
tokens := t.TokenizeString(myStr)
fmt.Println(tokens)
// Output: [this is my long string i can replace contractions like cannot or they have i can replace abbreviations such as demonstration or jan.]
/*
Note: you can remove stop words too!!!
*/
t.SetStopWordRemoval(true)
myOtherStr := "This is another string to demonstrate stop words removal, words like: and or but the are are all stop word examples"
tokens = t.TokenizeString(myOtherStr)
fmt.Println(tokens)
// Output: [string demonstrate stop words removal words stop word examples]
}package main
import (
"fmt"
"log"
"github.com/rekram1-node/tokenizer/languages"
"github.com/rekram1-node/tokenizer/tokenizer"
)
func main() {
// add a string containing all the "separators" you want
// important note: including the "." would degrade the ability to replace abbreviations
customSeparators := "\t\n\r ,:?\"!;()"
// specify your settings here
settings := &tokenizer.Settings{
KeepSeparators: false,
RemoveStopWords: true,
// you can have your own language configuration, see the language struct
/*
type Lanuage struct {
StopWords map[string]uint8
Contractions map[string]string
Abbreviations map[string]string
}
you can create your own using languages.NewLanguage(yourStopWords, yourContractions, yourAbbreviations)
*/
Lanuage: languages.English,
}
// custom settings return an error incase of a misconfigured/missing setting
t, err := settings.Custom(customSeparators)
if err != nil {
log.Fatal(err)
}
myStr := "This is my long string! I can replace contractions like can't or they've! I can replace abbreviations such as: demonstr. or jan. This is another string to demonstrate stop words removal, words like: and or but the are are all stop word examples"
tokens := t.TokenizeString(myStr)
fmt.Println(tokens)
// Output: [long string replace contractions replace abbreviations demonstration january string demonstrate stop words removal words stop word examples]
}See benchmark test
Using the benchmark test, you can see that even with 1 million words we still only have a meager 39 allocations per operation. I believe this number can be beat as well but I am still looking into the feasibility of some operations
task: [bench] go test -bench=. ./tokenizer -run=^# -count=10 -benchmem | tee preformance.txt
goos: darwin
goarch: arm64
pkg: github.com/rekram1-node/tokenizer/tokenizer
BenchmarkTokenize/word_count:_10-8 893482 1302 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 909124 1309 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 908605 1298 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 905206 1270 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 900795 1254 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 912558 1267 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 928588 1255 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 885170 1264 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 894373 1253 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_10-8 917389 1258 ns/op 496 B/op 5 allocs/op
BenchmarkTokenize/word_count:_100-8 81775 13954 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 85268 13914 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 85263 13912 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 85602 13973 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 85400 13908 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 85245 13941 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 85260 13896 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 85585 14014 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 85687 13925 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_100-8 84966 13921 ns/op 4080 B/op 8 allocs/op
BenchmarkTokenize/word_count:_1000-8 8143 143447 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 7996 144030 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 8072 144614 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 8059 143350 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 8186 143046 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 8046 144263 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 8026 143982 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 8038 144174 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 8079 144182 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_1000-8 8030 142584 ns/op 50416 B/op 12 allocs/op
BenchmarkTokenize/word_count:_10000-8 795 1447904 ns/op 685299 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 784 1460662 ns/op 685299 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 824 1459105 ns/op 685300 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 818 1462055 ns/op 685299 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 805 1454443 ns/op 685299 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 818 1458763 ns/op 685300 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 819 1453174 ns/op 685299 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 811 1452126 ns/op 685298 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 823 1450348 ns/op 685299 B/op 19 allocs/op
BenchmarkTokenize/word_count:_10000-8 817 1457028 ns/op 685298 B/op 19 allocs/op
BenchmarkTokenize/word_count:_100000-8 70 15327946 ns/op 8942875 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 75 15442076 ns/op 8942871 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 76 15366360 ns/op 8942882 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 75 15329089 ns/op 8942887 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 74 15337678 ns/op 8942874 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 74 15398484 ns/op 8942891 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 74 15302757 ns/op 8942877 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 75 15362398 ns/op 8942871 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 74 15387052 ns/op 8942882 B/op 29 allocs/op
BenchmarkTokenize/word_count:_100000-8 75 15310445 ns/op 8942881 B/op 29 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 171073882 ns/op 88036674 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 169618688 ns/op 88036672 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 169880493 ns/op 88036592 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 169492944 ns/op 88036672 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 170651521 ns/op 88036640 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 169312208 ns/op 88036624 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 169648021 ns/op 88036640 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 170238188 ns/op 88036624 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 170449750 ns/op 88036656 B/op 39 allocs/op
BenchmarkTokenize/word_count:_1000000-8 6 169968132 ns/op 88036592 B/op 39 allocs/op
PASS
ok github.com/rekram1-node/tokenizer/tokenizer 73.772s