npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2024 – Pkg Stats / Ryan Hefner

@storyscript/nlp

v1.0.0

Published

## Installation The NLP runs with Node and Typescript. To set it up, simply run; ```bash yarn ```

Downloads

3

Readme

Natural Language Processing

Installation

The NLP runs with Node and Typescript. To set it up, simply run;

yarn

To provide sentences via the standard input:

yarn prompt

To run the tests:

yarn test

The two Analysers

There are 2 different methods/algorithms to analyse a sentence, here's a quick description of both of them

Sentence Analyser (/src/analyser/)

What is it?

This was the first mechanism to be developed, it's more complete and does a better job at recognizing the parts of the sentence, but it's way less flexible and more scattered, which makes it hard to understand and modify it.

How does it work?

Everything is based on Grammar analysis, (see Gramex below).

We can identify specific grammar formations, and figure give them a label / group based on those formation. Once we labeled the groups, we use specific weighed words to find the meaning.
There are 2 limitations to this method:

  1. The POS analyser we use is far from being perfect, and makes a lot of mistakes (e.g. "text me after 5PM", "text" will be considered as a noun)

  2. Since we use specific, hard-coded words to understand the meaning of a group, we (obviously) don't cover the whole English language, and it's pretty easy to fall in edge cases.

What's valuable in there?

The most important/interesting part of this module is the informations analysis (/src/infos), especially the time/date analysis.
Note that the informations analysis is not linked in any way to the Sentence Analyser, it's its own module, So it can easily be moved around and used somewhere else.

Context Analyser (/src/contextAnalyser.ts)

What is it

This is basically augmented regex. This all began as a Science Day experiment, and I ended up switching to this analyser as it's much lighter, faster, and less complex than the first one.

How does it work?

Everything is based on the String Parser. We can define rules to match any type of sentence / grammar formation, and mix everything in a regular expression.

What's valuable in there?

If we want to completely switch to a new analysis mechanism (which will probably be the case, see Overall Limitations), we can keep this StringParser for the sentence tokenization (What is a when block? What is an instruction block?)

Overall limitations

The main problem with those analyser is that everything is based on hard-coded rules. Therefore, none of them can be enriched by deep learning.

Inner working

Gramex

The Gramex is a mix of Grammar and regex.
It allows to search for specific grammar formations in sentences, for example:

new WordGroup("This cat is so cute!").findGrammar("RB JJ");
// Result: so cute (so = adverb/RB, cute = adjective/JJ)

And we can of course use more advanced regex:

const str = "When I'm not in the house, lock all the doors and turn off the lights";
new WordGroup(str).findGrammar("( PRP| RB)? DT NN");
// Result: ["in  the house", "all the doors", "the lights"]
//           ==  === =====    === === =====    === ====== 
//           PRP  DT   NN     RB  RB   NN      DT    NN

Tokenization

We use 2 different types of tokenization: the first one is a word per word tokenization (it's basically a POS tagger).
For example:

new Tokenizer().wordPerWord("Turn on the heating system");
// Will give:
{
   "Turn on": "VB",
   "the": "DT",
   "heating": "NN",
   "system": "NN"
}

To do so, we use pos, which is a simple POS tagger Node module.

The second type of tokenization we use is a word grouping tokenization. It allows us to form noun or verb groups, based on specific gramex rules.
Those rules can be found at /data/grammar/wordGroupingRules.sp.json

An example of the word grouping tokenization:

new Toknenizer().groupWords("Turn off the old stove next to the washing machine")
{
    "Turn off": "G_VV",
    "the old stove": "G_NN",
    "next to": "PRP",
    "the washing machine": "G_NN"
}

Meaning

Each part of a sentence has a meaning (of the Meaning class).
A Meaning is composed of multiple components:

  • Action
  • Time
  • Subject
  • Item
  • Value
  • Target

Here's an example:

const str = 'Send me a text on the 5th of july';
new SentenceAnalyser(str).createMeanings();
// result:
Meaning {
	action: Action {
		verb: 'send'
		tense: 'present'
	},
	subject: Person {
		personTitle: 'SELF'
	},
	item: Item {
		itemName: 'text'
	},
	time: Time {
		day: 5,
		month: 7
	},
	target: {},
	value: {}
}

Data

For now, all the Data layer is handled by the Data class ( src/data/data.ts ).
The actual Data is located in the " data " directory, as JSON files.

There is a variety of files in the data folder, but they all follow one of the 3 following formats:

  • Balance (*.bal.json)
  • String Parser (*.sp.json)
  • Tree Parser (*.tp.json)

Each format has a class dedicated to reading it and parsing the file. The following section will describe those parsers more in depth.

Balance

To determine the nature of a noun group, we use a weighting system, handled by the class Balance.
The Balance is first given a setting, and then a string to weight.

A balance setting resembles the following example:

{
	"yes": {
		"definitely": 100, "most likely": 50, "probably": 20
	},
	"no": {
		"never": 100, "unlikely": 50
	}
}

And here's this setting in action:

const setting = {/*setting above*/};
const bal = new Balance(setting);
bal.categorize('I\'ll never go there'); // no
bal.categorize('Definitely, count me in!'); // yes
bal.categorize('It\'s unlikely, I\'ll probably stay at home'); // {yes: 20, no:50} => no 

We can rig also the balance:

bal.rig({"yes": 40});
bal.categorize('It\'s unlikely, I\'ll probably stay at home'); // {yes: 20+40, no:50} => yes 

This is useful in our case to infer a meaning from external elements, for example:

"Call my mom"

We know that the verb call is most likely followed by a person, so, when we weight my mom we can rig the balance to add weight to the person property

Tree Parser

The tree files respects the following format:

{
    "more": [
        "more", "above", "hotter", "higher", "warmer"
    ],
    "less": [
        "below", "under", "colder", "less", "lower"
    ]
}

It allows to find a parent by applying the children as regex to a sentence:

new TreeParser(data).findParentOfMatched("the temperature is under 15 degrees");
// result: "less"

To find a parent by testing a given regex on the children:

new TreeParser(data).findParentOfMatching(/above/gi);
// result: "more"

You can also retrieve children by searching the parents in the same fashion:

new TreeParser(data).findChildrenOfMatched("I need more gaz!"); // ["more", "above"...]
new TreeParser(data).findChildrenOfMatching(/less/gi); // ["below", "under", ...]

String Parser

The String parser is the most powerful, but also the most complicated of the 3 file formats.

Here's an example of the String parser file format: (part of the timeComponents.sp.json file)

 1 | {
 2 |     "definitions": {
 3 |         "this": "this|este|ce|この|今"
 4 |         "month": "month|mes|mois|月"
 5 |     },
 6 |     "matches": {
 7 |         "\\d+h(\\d+)": "m=$1",
 8 |         "(\\d+)(:\\d+)?PM": "h={={$1+12}}",
 9 |         "{{:this:}} {{:month:}}": "M={o0{month, 0}}"
10 |     }
11 | }

Let's analyse this line by line, and discover the features along the way:


 2 |     "definitions": {
 3 |         "this": "this|este|ce|この|今"
 4 |         "month": "month|mes|mois|月"
 5 |     },
     [...]
 9 |         "{{:this:}} {{:month:}}": "M={o0{month, 0}}"

Those are definitions, they can be referred to later in the code, for example here, this and month are used in the left part of the 9th line. Those definitions can include each other, and circular reference works to some extents. See the tests for more examples.


2 |    "\\d+h(\\d+)": "m=$1",

The fist thing we can see here, is that the key supports regex. It will match a specific part of the sentence, and give you its transformation.
Here, the transformation is m=$1, as in a replace, the $1 will be replaced by the captured group.

So, for example if we apply the above rule to "18h21", the transformation will be "m=21"

Let's look at the next line:

 8 |         "(\\d+)(:\\d+)?PM": "h={={$1+12}}",

Here, we find a new syntax: {={...}}. This syntax is used to make calculations. This will be replaced by the result of the operation.

So if we apply this rule to "6:18PM", two things will happen:

  • The $1 will be replaced by the captured value: "h={={6+12}}"
  • The calculation will be made: "h=18"

Now the final line:

 9 |         "{{:this:}} {{:month:}}": "M={o0{month, 0}}"

We find here yet another syntax: {o0{...}}. This allows the user to define his own operations. In this example, the user defined a function, which takes 2 arguments: a unit("month"), and a number(0).

In this example, let's imagine that the function is just a concatenation of the unit and the number. If applied to "this month", this rule will give the transformation "M=month0"

Here's an example of usage for the StringParser:

new StringParser(data)
    .parseString(
        "Send me flowers at 6:32PM this month",
        (match, transformation) => {
            // This function is called when a
            // rule is met
        },
        (nonMatch) => {
            // This function is called with all 
            // the parts that don't meet any rule
        },
        [
            (unit, num) => {
                // This is a custom operation function
                return unit + num;
            }
        ]
    );