@christophe77/sentiment-js

sentimentJS

SentimentsJS is build on to of tensorflow js. It allows you to add a sentiment analysis on your node or browser javascript app. With this package you'll be able to get the overall sentiment from a text and the toxicity classification.

Installation

yarn add @christophe77/sentiment-js

or

npm install @christophe77/sentiment-js

How to use

the "analyse" function takes an object as parameter.

The object has 2 properties :

• text : string
• type : 'sentiment' | "toxicity" | "both" | "combined"

The function will return the result like this :

		{
			"sentiment": {
				"score": 0.9931357502937317,
				"result": "positive"
			},
			"toxicity": [
				{"label": "insult", "result": true},
				{"label": "toxicity","result": true},
				{.......}
			]
		}

property "label" can have those values :

• identity_attack
• insult
• obscene
• severe_toxicity
• sexual_explicit
• threat
• toxicity

Exemple :

    import sjs from '@christophe77/sentiment-js';

    const checkThisText = async () => {
        // sentiment
	    const beerScore = await sjs.analyse({
		    text: 'I love cold beers',    
		    type: 'sentiment',    
	    });
	    console.log("'I love cold beers' : ", JSON.stringify(beerScore));

        // toxicity
	    const vegetableScore = await sjs.analyse({
		    text: 'I really hate those fucking vegetables',
		    type: 'toxicity',
		});
	    console.log("'I really hate those fucking vegetables' : ", JSON.stringify(vegetableScore));

        // both
	    const insultScore = await sjs.analyse({
		    text: 'You are just a fucking bitch. You just deserve to receive cum on your dirty shitty face',
		    type: 'both',
	    });
	    console.log( "'You are just a fucking bitch. You just deserve to receive cum on your dirty shitty face' : ",   JSON.stringify(insultScore) );

        // combined
	    const combinedScore = await sjs.analyse({
		    text: 'You are just a fucking bitch. You just deserve to receive cum on your dirty shitty face',
		    type: 'combined',
	    });
	    console.log( "'You are just a fucking bitch. You just deserve to receive cum on your dirty shitty face' : ",   JSON.stringify(combinedScore) );
	 };

Console output will be :

'I love cold beers' : {
	"sentiment": { score":0.8631731867790222,"result":"positive"},
	"toxicity":[]
}

'I really hate those fucking vegetables' : {
	"sentiment": {"score":0.1487867832183838,"result":"negative"},
	"toxicity":[{"label":"toxicity","result":true}]
}

'You are just a fucking bitch. You just deserve to receive cum on your 	dirty shitty face' : 
{
	"sentiment":{"score":0.9931357502937317,"result":"positive"},
	"toxicity":[
		{"label":"insult","result":true},
		{"label":"toxicity","result":true}
	]
}

'You are just a fucking bitch. You just deserve to receive cum on your 	dirty shitty face' : 
{
	"combined":"negative"
}

Has you can see on the 3rd exemple, the sentences are very vulgar but the sentiment is computed as "positive". That's why I added a "combined" classification so you can check both toxicity and sentiment result to determinate if it's a false positive.

TODO

• add a tensorflow backend to get better speed and performance.