1. use average of words who have sentiment score associated with them;
2. normalize this average with average sentence length (assume 15 words);
3. (hashtag score + this normalize score)/2 will be the final normalized score.

Refer to comments in #12.

hashtag is automatically identified by wink-tokenizer; remove the # and extract the balance word and lookup the score — this itself should be handle simple hashtags such as #win or #fail, etc.

We can even try regex to extract multiple words to perform look up separately for each extracted word; for example #FailedProduct can be split using /[A-Z]+[a-z]*|[0-9]+/g regex — this will on match yield [ 'Failed', 'Product' ] —> will result in a score of -2. If there are >1 sentiment words, then average of sentiment only words should be taken as the score.

Since hashtag is more like the category of the text, the final score from hashtags must be added with the final sentiment score of the text and the total should be divided by 2 to ensure hashtag gets its due credit!

Do you have any plan to support multilang ?

The tokenizer has moved to 3.2.0 from 3.0.0.

Hi! Thanks so much for creating this library! I was trying it out by creating a very simple React application that will do sentiment analysis for a given text input as the input changes.

The app is here -- https://react-sentiment-analyzer.netlify.com/ -- and/but although it's working pretty well on localhost in development, when I create a production build using the webpack setup in Create-React-App it appears that the way wink-sentiment tokenizes the input string doesn't work the same way it does in development. Essentially it seems to be splitting words into smaller pieces than it does in development, which makes me think that I may be importing something incorrectly.

For example, here's the local development version which shows the wink-sentiment output for the word "angry" via Redux-devtools on the right:

Note that it's parsing it all as a single word.

As a contrast, this is the Redux devtools output for the same word on https://react-sentiment-analyzer.netlify.com/ in a production build:

Here it's parsing the word "angry" into two tokens: a and ngry, and thereby not outputting the same sentiment.

The code I'm using to run wink-sentiment is https://github.com/kellyi/react-sentiment-analyzer/blob/master/src/utils/sentimentAnalyzer.js ->

import sentiment from 'wink-sentiment';

const sentimentAnalyzer = text => sentiment(text);

export default sentimentAnalyzer;

Do you have any idea why wink-sentiment might be tokenizing things one way in development and another in production?

Thanks again for creating this library!

refer to pr #3