This package is a golang version of the original python library flashtext, based on the FlashText algorithm which is a special version of the Aho-Corasick algorithm.
The utility of the package is focused on keywords extraction and replacement with fixed strings at scale, the time complexity of the algorithm doesn't depend on the number of keys being searched or replaced. For a document of size N (characters) and a dictionary of M keys to search/replace, the time complexity is O(N).
Flashtext doesn't do regular expression and it's not a replacement of regex
$ go get github.com/ayoyu/flashtext
package main
import (
"fmt"
"github.com/ayoyu/flashtext/flash"
)
func main(){
flashKeys := flash.NewFlashKeywords(false)
// add to dictionnary the key `Apple` with its cleanWord `Fruit`.
flashKeys.AddKeyWord("Apple", "Fruit")
// add to dictionnary the key `Apple` with its cleanWord `Company`.
flashKeys.AddKeyWord("FootBall", "Sport")
// add to dictionnary the key without a cleanWord.
flashKeys.Add("Banana")
// It depends on the context of the use case, the cleanWord can be
// seen as a synonym to its key or a label/entity to describe its key,...etc.
// (Similar to the Elasticsearch `Synonym token filter` functionality)
// The search will extract the keys that exists in the flash keyword dictionnary
// with their `cleanWords`
res := flashKeys.Search("I played football, while eating my apple")
for _, item := range res {
fmt.Println(item)
}
}- Output
{football false sport 9 16}
{apple false fruit 35 39}From the previous example with the same added keys:
flashKeys := flash.NewFlashKeywords(true)
// ...
res := flashKeys.Search("I played football, while eating my Apple")
for _, item := range res {
fmt.Println(item)
}- Ouput
{Apple false Fruit 35 39}
The structure of the resulting output is the following:
type Result struct {
Key string
IsPrefix bool
CleanWord string
Start int
End int
}- `Key`: the string keyword found in the search text
- `IsPrefix`: Indicates if the key A is a prefix of another key B
where A and B are both in the dictionary of the flash keywords
- `CleanWord`: The string with which the found key will be replaced in the text.
It depends on the context of the use case, the cleanWord can be
seen as a synonym to its key or a label/entity to describe its key,...etc (Similar to the Elasticsearch `Synonym token filter` functionality)
- `Start & End`: span information about the start and end indexes if the key found in the text
Replace the keys added to the flash keywords with their clean words if they exist in the document.
In this example FootBall and Apple will be replaced respectively with their clean words Sport and Fruit, but in the case of the key Banana, it doesn't get replaced because the key has no cleanWord.
import (
"fmt"
"github.com/ayoyu/flashtext/flash"
)
func main() {
flashKeys := flash.NewFlashKeywords(false)
flashKeys.AddKeyWord("Apple", "Fruit")
flashKeys.AddKeyWord("FootBall", "Sport")
flashKeys.AddKeyWord("🔥", "💪")
flashKeys.Add("Banana")
text := "I played football, while eating my Apple 🔥"
fmt.Println("New text: ", flashKeys.Replace(text))
}- Output
New text: i played sport, while eating my fruit 💪
import (
"fmt"
"github.com/ayoyu/flashtext/flash"
)
func main() {
flashKeys := flash.NewFlashKeywords(true)
flashKeys.AddKeyWord("Apple", "Fruit")
flashKeys.AddKeyWord("FootBall", "Sport")
flashKeys.AddKeyWord("🔥", "💪")
flashKeys.Add("Banana")
text := "I played football, while eating my Apple 🔥"
fmt.Println("New text: ", flashKeys.Replace(text))
}- Output:
New text: I played football, while eating my Fruit 💪
To check the documentation of all the methods and the functions in your browser, type in your terminal:
$ godoc -http=:8080
Then browse to: http://localhost:8080/pkg/github.com/ayoyu/flash
First generate some random data or use the ones that are located in the testdata folder
$ go run testdata/gen_testdata.go(cd to the benchmarks folder)
- Benchmark FlashText Search
$ go test -benchmem -run=^$ -bench ^BenchmarkFlashTextSearch$
Search on a corpus text of size: 669317
goos: linux
goarch: amd64
pkg: github.com/ayoyu/flashtext/benchmarks
cpu: Intel(R) Core(TM) i5-9300H CPU @ 2.40GHz
BenchmarkFlashTextSearch/key_size=10-8 162 7449500 ns/op 848 B/op 4 allocs/op
BenchmarkFlashTextSearch/key_size=1010-8 85 22213111 ns/op 2653670 B/op 19 allocs/op
BenchmarkFlashTextSearch/key_size=2010-8 50 31000992 ns/op 5930451 B/op 22 allocs/op
BenchmarkFlashTextSearch/key_size=3010-8 32 38498148 ns/op 7642576 B/op 23 allocs/op
BenchmarkFlashTextSearch/key_size=4010-8 39 33514732 ns/op 9797074 B/op 24 allocs/op
BenchmarkFlashTextSearch/key_size=5010-8 36 34218947 ns/op 12508624 B/op 25 allocs/op
BenchmarkFlashTextSearch/key_size=6010-8 33 42046412 ns/op 15916498 B/op 26 allocs/op
BenchmarkFlashTextSearch/key_size=7010-8 28 55664084 ns/op 15916496 B/op 26 allocs/op
BenchmarkFlashTextSearch/key_size=8010-8 30 36441695 ns/op 20192720 B/op 27 allocs/op
BenchmarkFlashTextSearch/key_size=9010-8 33 57783636 ns/op 20192720 B/op 27 allocs/op
BenchmarkFlashTextSearch/key_size=10010-8 37 45879421 ns/op 20192720 B/op 27 allocs/op
BenchmarkFlashTextSearch/key_size=11010-8 33 52046617 ns/op 25550288 B/op 28 allocs/op
BenchmarkFlashTextSearch/key_size=12010-8 27 38160474 ns/op 25550291 B/op 28 allocs/op
BenchmarkFlashTextSearch/key_size=13010-8 33 51766075 ns/op 25550290 B/op 28 allocs/op
BenchmarkFlashTextSearch/key_size=14010-8 30 52940062 ns/op 25550288 B/op 28 allocs/op
BenchmarkFlashTextSearch/key_size=15010-8 20 59203761 ns/op 32259536 B/op 29 allocs/op
BenchmarkFlashTextSearch/key_size=16010-8 31 45628554 ns/op 32259536 B/op 29 allocs/op
BenchmarkFlashTextSearch/key_size=17010-8 26 45658666 ns/op 32259536 B/op 29 allocs/op
BenchmarkFlashTextSearch/key_size=18010-8 27 39414959 ns/op 32259536 B/op 29 allocs/op
BenchmarkFlashTextSearch/key_size=19010-8 25 49583917 ns/op 32259536 B/op 29 allocs/op
BenchmarkFlashTextSearch/key_size=20010-8 27 47829831 ns/op 40664531 B/op 30 allocs/op
PASS
ok github.com/ayoyu/flashtext/benchmarks 42.381s
- Benchmark Regex Search
$ go test -benchmem -run=^$ -bench ^BenchmarkRegexSearch$
Search on a corpus text of size: 669317
goos: linux
goarch: amd64
pkg: github.com/ayoyu/flashtext/benchmarks
cpu: Intel(R) Core(TM) i5-9300H CPU @ 2.40GHz
BenchmarkRegexSearch/key_size=10-8 9 116758673 ns/op 2760 B/op 10 allocs/op
BenchmarkRegexSearch/key_size=1010-8 1 16455596425 ns/op 779272 B/op 2785 allocs/op
BenchmarkRegexSearch/key_size=2010-8 1 34206170218 ns/op 1505056 B/op 5548 allocs/op
BenchmarkRegexSearch/key_size=3010-8 1 52015943197 ns/op 2204912 B/op 8185 allocs/op
BenchmarkRegexSearch/key_size=4010-8 1 72208914019 ns/op 2968072 B/op 10852 allocs/op
BenchmarkRegexSearch/key_size=5010-8 1 87424594909 ns/op 3648112 B/op 13442 allocs/op
BenchmarkRegexSearch/key_size=6010-8 1 111404534897 ns/op 4469488 B/op 16163 allocs/op
BenchmarkRegexSearch/key_size=7010-8 1 129792983299 ns/op 5183560 B/op 18770 allocs/op
BenchmarkRegexSearch/key_size=8010-8 1 153291348479 ns/op 6027576 B/op 21370 allocs/op
SIGQUIT: quit
PC=0x468301 m=0 sigcode=0
goroutine 0 [idle]:
runtime.futex()
/usr/local/go/src/runtime/sys_linux_amd64.s:559 +0x21 fp=0x7ffda567d168 sp=0x7ffda567d160 pc=0x468301
runtime.futexsleep(0x467ef3?, 0xa567d1e8?, 0x444eeb?)
/usr/local/go/src/runtime/os_linux.go:69 +0x36 fp=0x7ffda567d1b8 sp=0x7ffda567d168 pc=0x432256
runtime.notesleep(0x60e5c8)
/usr/local/go/src/runtime/lock_futex.go:160 +0x87 fp=0x7ffda567d1f0 sp=0x7ffda567d1b8 pc=0x40b787
runtime.mPark(...)
/usr/local/go/src/runtime/proc.go:1457
runtime.stopm()
/usr/local/go/src/runtime/proc.go:2239 +0x8c fp=0x7ffda567d220 sp=0x7ffda567d1f0 pc=0x43cb2c
runtime.findRunnable()
/usr/local/go/src/runtime/proc.go:2866 +0x9e8 fp=0x7ffda567d310 sp=0x7ffda567d220 pc=0x43e1c8
runtime.schedule()
/usr/local/go/src/runtime/proc.go:3206 +0xbe fp=0x7ffda567d348 sp=0x7ffda567d310 pc=0x43effe
runtime.goschedImpl(0xc000202820)
/usr/local/go/src/runtime/proc.go:3370 +0xc5 fp=0x7ffda567d380 sp=0x7ffda567d348 pc=0x43f625
runtime.gosched_m(0xc000202820?)
/usr/local/go/src/runtime/proc.go:3378 +0x31 fp=0x7ffda567d3a0 sp=0x7ffda567d380 pc=0x43f7b1
runtime.mcall()
/usr/local/go/src/runtime/asm_amd64.s:448 +0x43 fp=0x7ffda567d3b0 sp=0x7ffda567d3a0 pc=0x464303
...
*** Test killed with quit: ran too long (11m0s).
exit status 2
FAIL github.com/ayoyu/flashtext/benchmarks 660.014s
$ go run simple_flash_regex_bench.go
keys_size | FlashText (s) | Regex (s)
---------------------------------------
10 | 0.00053121 | 0.007381449
1010 | 0.000902698 | 1.021105121
2010 | 0.001164453 | 2.155324188
3010 | 0.001272009 | 3.189556999
4010 | 0.001415052 | 4.489287341
5010 | 0.00151844 | 5.662644436
6010 | 0.001601235 | 6.820220812
7010 | 0.001711219 | 7.845579981
8010 | 0.001785076 | 9.740038207
...
- The python library: https://github.com/vi3k6i5/flashtext
- The original paper: https://arxiv.org/abs/1711.00046
@ARTICLE{2017arXiv171100046S,
author = {{Singh}, V.},
title = "{Replace or Retrieve Keywords In Documents at Scale}",
journal = {ArXiv e-prints},
archivePrefix = "arXiv",
eprint = {1711.00046},
primaryClass = "cs.DS",
keywords = {Computer Science - Data Structures and Algorithms},
year = 2017,
month = oct,
adsurl = {http://adsabs.harvard.edu/abs/2017arXiv171100046S},
adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}
- Make the data structure Thread-safe to be able to use it in a concurrent environment and call it
ConcurrentFlashTextsimilar name to theConcurrentHashMap.