github.com/aaronhanh/textrank
This source code is an implementation of textrank algorithm, under MIT licence.
The minimum requred Go version is 1.8.
If there was a program what could rank book size text's words, phrases and sentences continuously on multiple threads and it would be opened to modifing by objects, written in a simple, secure, static language and if it would be very well documented... Now, here it is.
The following link Recona is a simple, pre-programmed a.i. what uses this library to ranking raw texts. It visualizes how ranking works and it represents how it could be used for different purposes: Recona.app
You can install TextRank by Go's get:
go get github.com/DavidBelicza/TextRank
TextRank uses the DEP as vendoring tool, so the required dependencies are versioned under the vendor folder. The exact version number defined in the Gopkg.toml. If you want to reinstall the dependencies, use the DEP functions: flush the vendor folder and run:
dep ensure
Using Docker to TextRank isn't necessary, it's just an option.
Build image from the repository's root directory:
docker build -t go_text_rank_image .
Create container from the image:
docker run -dit --name textrank go_text_rank_image:latest
Run the go test -v . code inside the container:
docker exec -i -t textrank go test -v .
Stop, start or remove the container:
docker stop textrankdocker start textrankdocker rm textrankToo see how does it work, the easiest way is to use the sample text. Sample text can be found in the textrank_test.go file at this line. It's a short size text about Gnome Shell.
The most important phrases from the sample text are:
| Phrase | Occurrence | Weight |
|---|---|---|
| gnome - shell | 5 | 1 |
| extension - gnome | 3 | 0.50859946 |
| icons - tray | 3 | 0.49631447 |
| gnome - caffeine | 2 | 0.27027023 |
The gnome is the most often used word in this text and shell is also used multiple times. Two of them are used together as a phrase 5 times. This is the highest occurrence in this text, so this is the most important phrase.
The following two important phrases have same occurrence 3, however they are not equal. This is because the extension gnome phrase contains the word gnome, the most popular word in the text, and it increases the phrase's weight. It increases the weight of any word what is related to it, but not too much to overcome other important phrases what don't contain the gnome word.
The exact algorithm can be found in the algorithm.go file at this line.
Automatic summarization is the process of reducing a text document with a computer program in order to create a summary that retains the most important points of the original document. Technologies that can make a coherent summary take into account variables such as length, writing style and syntax. Automatic data summarization is part of machine learning and data mining. The main idea of summarization is to find a representative subset of the data, which contains the information of the entire set. Summarization technologies are used in a large number of sectors in industry today. - Wikipedia
This is the most basic and simplest usage of textrank.
package main
import (
"fmt"
"github.com/DavidBelicza/TextRank"
)
func main() {
rawText := "Your long raw text, it could be a book. Lorem ipsum..."
// TextRank object
tr := textrank.NewTextRank()
// Default Rule for parsing.
rule := textrank.NewDefaultRule()
// Default Language for filtering stop words.
language := textrank.NewDefaultLanguage()
// Default algorithm for ranking text.
algorithmDef := textrank.NewDefaultAlgorithm()
// Add text.
tr.Populate(rawText, language, rule)
// Run the ranking.
tr.Ranking(algorithmDef)
// Get all phrases by weight.
rankedPhrases := textrank.FindPhrases(tr)
// Most important phrase.
fmt.Println(rankedPhrases[0])
// Second important phrase.
fmt.Println(rankedPhrases[1])
}
After ranking, the graph contains a lot of valuable data. There are functions in textrank package what contains logic to retrieve those data from the graph.
package main
import (
"fmt"
"github.com/DavidBelicza/TextRank"
)
func main() {
rawText := "Your long raw text, it could be a book. Lorem ipsum..."
// TextRank object
tr := textrank.NewTextRank()
// Default Rule for parsing.
rule := textrank.NewDefaultRule()
// Default Language for filtering stop words.
language := textrank.NewDefaultLanguage()
// Default algorithm for ranking text.
algorithmDef := textrank.NewDefaultAlgorithm()
// Add text.
tr.Populate(rawText, language, rule)
// Run the ranking.
tr.Ranking(algorithmDef)
// Get all phrases order by weight.
rankedPhrases := textrank.FindPhrases(tr)
// Most important phrase.
fmt.Println(rankedPhrases[0])
// Get all words order by weight.
words := textrank.FindSingleWords(tr)
// Most important word.
fmt.Println(words[0])
// Get the most important 10 sentences. Importance by phrase weights.
sentences := textrank.FindSentencesByRelationWeight(tr, 10)
// Found sentences
fmt.Println(sentences)
// Get the most important 10 sentences. Importance by word occurrence.
sentences = textrank.FindSentencesByWordQtyWeight(tr, 10)
// Found sentences
fmt.Println(sentences)
// Get the first 10 sentences, start from 5th sentence.
sentences = textrank.FindSentencesFrom(tr, 5, 10)
// Found sentences
fmt.Println(sentences)
// Get sentences by phrase/word chains order by position in text.
sentencesPh := textrank.FindSentencesByPhraseChain(tr, []string{"gnome", "shell", "extension"})
// Found sentence.
fmt.Println(sentencesPh[0])
}
After ranking, the graph contains a lot of valuable data. The GetRank function allows access to the graph and every data can be retrieved from this structure.
package main
import (
"fmt"
"github.com/DavidBelicza/TextRank"
)
func main() {
rawText := "Your long raw text, it could be a book. Lorem ipsum..."
// TextRank object
tr := textrank.NewTextRank()
// Default Rule for parsing.
rule := textrank.NewDefaultRule()
// Default Language for filtering stop words.
language := textrank.NewDefaultLanguage()
// Default algorithm for ranking text.
algorithmDef := textrank.NewDefaultAlgorithm()
// Add text.
tr.Populate(rawText, language, rule)
// Run the ranking.
tr.Ranking(algorithmDef)
// Get the rank graph.
rankData := tr.GetRankData()
// Get word ID by token/word.
wordId := rankData.WordValID["gnome"]
// Word's weight.
fmt.Println(rankData.Words[wordId].Weight)
// Word's quantity/occurrence.
fmt.Println(rankData.Words[wordId].Qty)
// All sentences what contain the this word.
fmt.Println(rankData.Words[wordId].SentenceIDs)
// All other words what are related to this word on left side.
fmt.Println(rankData.Words[wordId].ConnectionLeft)
// All other words what are related to this word on right side.
fmt.Println(rankData.Words[wordId].ConnectionRight)
// The node of this word, it contains the related words and the relation weight.
fmt.Println(rankData.Relation.Node[wordId])
}
It is possibe to add more text after another texts already have been added. The Ranking function can merge these multiple texts and it can recalculate the weights and all related data.
package main
import (
"fmt"
"github.com/DavidBelicza/TextRank"
)
func main() {
rawText := "Your long raw text, it could be a book. Lorem ipsum..."
// TextRank object
tr := textrank.NewTextRank()
// Default Rule for parsing.
rule := textrank.NewDefaultRule()
// Default Language for filtering stop words.
language := textrank.NewDefaultLanguage()
// Default algorithm for ranking text.
algorithmDef := textrank.NewDefaultAlgorithm()
// Add text.
tr.Populate(rawText, language, rule)
// Run the ranking.
tr.Ranking(algorithmDef)
rawText2 := "Another book or article..."
rawText3 := "Third another book or article..."
// Add text to the previously added text.
tr.Populate(rawText2, language, rule)
// Add text to the previously added text.
tr.Populate(rawText3, language, rule)
// Run the ranking to the whole composed text.
tr.Ranking(algorithmDef)
// Get all phrases by weight.
rankedPhrases := textrank.FindPhrases(tr)
// Most important phrase.
fmt.Println(rankedPhrases[0])
// Second important phrase.
fmt.Println(rankedPhrases[1])
}
There are two algorithm has implemented, it is possible to write custom algorithm by Algorithm interface and use it instead of defaults.
package main
import (
"fmt"
"github.com/DavidBelicza/TextRank"
)
func main() {
rawText := "Your long raw text, it could be a book. Lorem ipsum..."
// TextRank object
tr := textrank.NewTextRank()
// Default Rule for parsing.
rule := textrank.NewDefaultRule()
// Default Language for filtering stop words.
language := textrank.NewDefaultLanguage()
// Using a little bit more complex algorithm to ranking text.
algorithmChain := textrank.NewChainAlgorithm()
// Add text.
tr.Populate(rawText, language, rule)
// Run the ranking.
tr.Ranking(algorithmChain)
// Get all phrases by weight.
rankedPhrases := textrank.FindPhrases(tr)
// Most important phrase.
fmt.Println(rankedPhrases[0])
// Second important phrase.
fmt.Println(rankedPhrases[1])
}
Graph ID exists because it is possible run multiple independent text ranking processes.
package main
import (
"fmt"
"github.com/DavidBelicza/TextRank"
)
func main() {
rawText := "Your long raw text, it could be a book. Lorem ipsum..."
// 1th TextRank object
tr1 := textrank.NewTextRank()
// Default Rule for parsing.
rule := textrank.NewDefaultRule()
// Default Language for filtering stop words.
language := textrank.NewDefaultLanguage()
// Default algorithm for ranking text.
algorithmDef := textrank.NewDefaultAlgorithm()
// Add text.
tr1.Populate(rawText, language, rule)
// Run the ranking.
tr1.Ranking(algorithmDef)
// 2nd TextRank object
tr2 := textrank.NewTextRank()
// Using a little bit more complex algorithm to ranking text.
algorithmChain := textrank.NewChainAlgorithm()
// Add text to the second graph.
tr2.Populate(rawText, language, rule)
// Run the ranking on the second graph.
tr2.Ranking(algorithmChain)
// Get all phrases by weight from first graph.
rankedPhrases := textrank.FindPhrases(tr1)
// Most important phrase from first graph.
fmt.Println(rankedPhrases[0])
// Second important phrase from first graph.
fmt.Println(rankedPhrases[1])
// Get all phrases by weight from second graph.
rankedPhrases2 := textrank.FindPhrases(tr2)
// Most important phrase from second graph.
fmt.Println(rankedPhrases2[0])
// Second important phrase from second graph.
fmt.Println(rankedPhrases2[1])
}
Engish is used by default but it is possible to add any language. To use other languages a stop word list is required what you can find here: https://github.com/stopwords-iso
package main
import (
"fmt"
"github.com/DavidBelicza/TextRank"
)
func main() {
rawText := "Your long raw text, it could be a book. Lorem ipsum..."
// TextRank object
tr := textrank.NewTextRank()
// Default Rule for parsing.
rule := textrank.NewDefaultRule()
// Default Language for filtering stop words.
language := textrank.NewDefaultLanguage()
// Add Spanish stop words (just some example).
language.SetWords("es", []string{"uno", "dos", "tres", "yo", "es", "eres"})
// Active the Spanish.
language.SetActiveLanguage("es")
// Default algorithm for ranking text.
algorithmDef := textrank.NewDefaultAlgorithm()
// Add text.
tr.Populate(rawText, language, rule)
// Run the ranking.
tr.Ranking(algorithmDef)
// Get all phrases by weight.
rankedPhrases := textrank.FindPhrases(tr)
// Most important phrase.
fmt.Println(rankedPhrases[0])
// Second important phrase.
fmt.Println(rankedPhrases[1])
}
It is thread safe. Independent graphs can receive texts in the same time and can be extended by more text also in the same time.
package main
import (
"fmt"
"time"
"github.com/DavidBelicza/TextRank"
)
func main() {
// A flag when program has to stop.
stopProgram := false
// Channel.
stream := make(chan string)
// TextRank object.
tr := textrank.NewTextRank()
// Open new thread/routine
go func(tr *textrank.TextRank) {
// 3 texts.
rawTexts := []string{
"Very long text...",
"Another very long text...",
"Second another very long text...",
}
// Add 3 texts to the stream channel, one by one.
for _, rawText := range rawTexts {
stream <- rawText
}
}(tr)
// Open new thread/routine
go func() {
// Counter how many times texts added to the ranking.
i := 1
for {
// Get text from stream channel when it got a new one.
rawText := <-stream
// Default Rule for parsing.
rule := textrank.NewDefaultRule()
// Default Language for filtering stop words.
language := textrank.NewDefaultLanguage()
// Default algorithm for ranking text.
algorithm := textrank.NewDefaultAlgorithm()
// Add text.
tr.Populate(rawText, language, rule)
// Run the ranking.
tr.Ranking(algorithm)
// Set stopProgram flag to true when all 3 text have been added.
if i == 3 {
stopProgram = true
}
i++
}
}()
// The main thread has to run while go-routines run. When stopProgram is
// true then the loop has finish.
for !stopProgram {
time.Sleep(time.Second * 1)
}
// Most important phrase.
phrases := textrank.FindPhrases(tr)
// Second important phrase.
fmt.Println(phrases[0])
}
The below image is a representation how works the simplest text ranking algorithm. This algorithm can be replaced by an another one by inject different Algorithm interface implementation.
FAQs
Unknown package
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Security News
In this segment of the Risky Business podcast, Feross Aboukhadijeh and Patrick Gray discuss the challenges of tracking malware discovered in open source softare.
Research
Security News
A threat actor's playbook for exploiting the npm ecosystem was exposed on the dark web, detailing how to build a blockchain-powered botnet.
Security News
NVD’s backlog surpasses 20,000 CVEs as analysis slows and NIST announces new system updates to address ongoing delays.