frequent-phrases v0.1.2-alpha

Frequent Phrases

Process large chunks of text into a node tree, which can then be traversed to grab phrases that match the given criteria.

To install:

npm install frequent-phrases

Basic Usage

The workflow is generally:

1. Construct FrequentPhrase instance
2. Define custom config (Optional)
3. Process text
4. Output frequent phrases

Construct

const FP = new FrequentPhrase();

Custom Config (more info HERE)

The default config object is as follows:

const defaultConfig = {
    maxPhraseLength: 6,
    selectionAlgorithm: 'dropOff',
    selectionConfig: {
        dropOff: {
            threshold: 0.5
        }
    },
    scoringAlgorithm: 'default',
    parserConfig: {
        chunkSentences: true,
        removeTypedSentences: true
    },
    preProcessing: {
        trim: 3
    },
    postProcessing: {
        uniqueWordAtCutoffDepth: 1
    }
}

Access the config property to modify this after instantiation, or construct a new config object and pass it in.

const FP = new FrequentPhrase();

FP.config = newConfigObject;
// or
FP.config.maxPhraseLength = 8; // etc.

Process Text And/Or Generate Phrases

The last bit can be separated out, or done altogether.

const speech = 'Five score years ago, a great American, in whose symbolic shadow' // ... MLK's I Have A Dream speech

To process text and then extract phrases:

await FP.process(speech);

// then get Frequent Phrases
await FP.getFrequentPhrases().then((res) => console.log(res))

To do both, just pass text in to getFrequentPhrases(). Note that this method overwrites previous tree data, and is best served if you are instantiating a new FrequentPhrase() everytime.

await FP.getFrequentPhrases(speech).then((res) => console.log(res));

Both methods will yield the same result:

// ^^^^^ console.log(res);
{
    ok: true
    msg: ''
    frequentPhrases: [
        { phrase: "", score: 0 },
        { phrase: "", score: 0 },
        { phrase: "", score: 0 },
        ...
    ]
    executionTime: '3.544ms'
}

***For More

Modifying the Library

To help understanding of best ways to modify for a specific use-case, the library works as follows: 1. Input corpus 2. Pre-process potential candidates 3. Select Candidates 4. Score selected Candidates 5. Post-process candidates 6. Output

Config

Pre-Processing

• trim - Trims candidate pool to only originate from the top trim starter words. Trim defaults to 0, or no trim.

Candidate Selection

• Selection Algorithm - Algorithm to use for selection algorithm. Default is a simple dropoff, which cuts off phrases based on their relative visits between child / parent.
• Selection Config - Stores constants to modify how selection algorithms perform. See here.

Candidate Scoring:

Defines what scoring algorithm is used. Default algo is based solely on averaged visits, meaning a higher visit average yields higher scores.

Post-Processing

• uniqueWordAtCutoffDepth - Trims scored candidates so that the highest-scored phrase from each starter word is represented.

Parser Config

• chunkSentences - convert a string into an array of it's contained sentences
• removeTypedSentences - find the unique, longest sentence amongst a gamut of typed copies of the same sentence.
  • e.g.: We are only interested in the sentence 'How are you?' but we have:
    • 'H'
    • 'Ho'
    • 'How'
    • ...
    • 'How are you?'

More

• .getFrequentPhrases(body)
• .process(body)
• .reset()

FrequentPhrase.getFrequentPhrases(body)

Return Frequent Phrases from data already processed.

Returns: Promise.<FP> - Frequent phrases present in the text

FrequentPhrase.process(body)

Process a string of sentences. Frequent phrases can only be extracted from processed text.

Returns: Promise<string[] | FPNode[]> - registry, rootNode

FrequentPhrase.reset()

Cleans out the sentence registry and destroys the node tree

Returns: Promise<string[] | FPNode[]> - registry, rootNode