nhunzaker / speakeasy

A simple natural language tool written for NodeJS

SpeakEasy

Simple Natural Language Processing

---

npm install speakeasy-nlp

---

Current commands:

1. *.classify : Roughly determines the action, subject, and owner (posessive object) of a sentence. As of 0.2.2 it also includes verbs, nouns, and adjectives
2. *.sentiment: A collection of methods to approximate the positive/negative affect of a statement (relative to the whole statement)
3. *.closest : Uses levenshtein distance to find the best match for a word given an array

---

var speak = require("./speakeasy-nlp");

// Analyze sentences at a basic level
// ------------------------------------- //
speak.classify("What is your name?")             //=> { action: "what", owner: "listener", subject: "name" }
speak.classify("Do you know what time it is?")   //=> { action: "what", owner: "it", subject: "time" }

// Sentiment analysis
// ------------------------------------- //
speak.sentiment.negativity("I hate your guts")   //=> { score: 1, words: [hate] }
speak.sentiment.positivity("I love you")         //=> { score: 1, words: [love] }

speak.sentiment.analyze("I love you, but you smell something aweful")  
// (Negative scores dictate a stronger influence of negative words)
//=> { score: -1, positive: { ... }, negative: { ... } }

// Closest word
// ------------------------------------- //
speak.closest("node", ["foo", "nodejs", "baz"])     //=> "nodejs"

Philosophy

The goal of this project is not to be the next final solution for natural language processing. There are plenty of other projects that do a significantly better job of this. SpeakEasy spawned out of another of my projects, Nodebot, as a method of processing user input to simulate the illusion of intelligence.

SpeakEasy's goal is to provide a library for NodeJS to perform simple language processing actions that perform well for 70%-80% of all cases.

License

speakeasy is released under the MIT License.

speakeasy's Issues

README EXAMPLE

The example

speak.classify("Do you know what time it is?") //=> { action: "what", owner: "it", subject: "time" }

is not correct.

.classify() never recognizes 'you' especially if leading the sentence

I'd expect it to return 'listener' and you see this behavior often when text contains "your", but curiously not pronouns like "you", which never is identified as the subject if the first word in a sentence:

const speak = require("speakeasy-nlp");
// below using speak.classify(String)

'What is your name?' 
// owner: 'listener', subject: 'name'   <--- Correct

'What name do you have?'
// owner: undefined, subject: 'you'

'You have a name?'
// owner: undefined, subject: 'name'

'You look lovely'
// owner: 'lovely', subject: 'look lovely' 

'I look lovely'
// owner: 'speaker', subject: 'look lovely' <--- Correct

code breaks when I extend Array.prototype

code breaks when I use these kind of code.
Array.prototype.somefuntion=function(){
//do something
}

error:
TypeError: words[i].toLowerCase is not a function
at POSTagger.tag (\node_modules\speakeasy-nlp\lib\classify\pos\POSTagger.js:52:40)

Possible to make corrective adjustments?

For example, someone said "i'm not so hot right now eh?" and that scored 2 for "hot" and "right", but it probably should've scored -1.

TypeError: string.match is not a function

Just running the module on a string i am getting this error , any ideas?

TypeError: string.match is not a function
at new LexerNode (web/node_modules/speakeasy-nlp/lib/classify/pos/lexer.js:32:31)
at module.exports.Lexer.lex (web/node_modules/speakeasy-nlp/lib/classify/pos/lexer.js:89:14)
at Object.classify (web/node_modules/speakeasy-nlp/lib/classify/index.js:86:21)
at Object.processMessage (web/src/server/Logics/ProcessMessage.js:12:20)
at web/src/server/Logics/SendMessage.js:69:28
at web/src/server/Models/UserModel.js:38:13
at model.Query. (web/node_modules/mongoose/lib/model.js:3731:16)
at web/node_modules/kareem/index.js:277:21
at web/node_modules/kareem/index.js:131:16
at _combinedTickCallback (internal/process/next_tick.js:67:7)
at process._tickCallback (internal/process/next_tick.js:98:9)

array.slice arguments error

I have found a issues with your code that I am unsure how to fix. You are supplying too many argument to the slice function in lib/classify/index.js on line 203. According to MDN, the slice function should only have 2 arguments, not 3.

 owner = words.slice(prepIndex + 1, -1, detIndex).join(" ");

Thanks!

Subject is wrong?

If I write "I am a dog" it says "dog" is the subject even though "I" am the subject.

Perhaps stop processing words/numbers in quotes?

I really like your project, i'm working on some kind of ifttt-clone for my personal usage, which should be configured as simple as possible, therfor i really like the idea of natural language configs, like

watch my server reinemuth.rocks with username [email protected] and password "password123password" for new mail

or

watch folder "/test123/test3" for new files

but words with number etc. in it are simply split into pieces, numbers are thrown away... Perhaps there could be some "stop-parsing" chars, maybe simple quotes?

Thanks,

Jens

pos-js is missing

Hi,

pos-js is just an emtpy folder

Client-side?

Is there a way to make this client-side and run in the browser?

License?

Hey there!

Very cool plugin -- noticed that there's not a license associated with this project. Is it MIT? GPL? Etc?

Hope this helps