Comment by sloria | Hacker News Reader

Top New Best Ask Show Jobs

Comment by sloria | Hacker News Reader

Parent Full thread

This is an extension to the Python programming language that makes it easier to analyze and manipulate text.

For example, an analyst might use sentiment analysis to see whether Facebook posts about a product are "positive" or "negative" in tone.

As another example, I hacked together this online sentiment analyzer using TextBlob: https://textfeel.herokuapp.com/

See also: NLP (Wikipedia): https://en.wikipedia.org/wiki/Natural_language_processing NLTK (a python library for NLP): http://nltk.org/ Twitter opinion mining using pattern: http://www.clips.ua.ac.be/pages/pattern-examples-elections

OK, so it's mostly to analyze text that's already been written? Can it also write natural language text based on data inputs?

> Can it also write natural language text based on data inputs?

from the features list it doesn't seem to.

What you're referring to is text generated using a Markov Chain algorithm. This will generate text that seems at first glance to be human generated. On closer inspection you'll find that it only follows common linguistic patterns, the actual content is gibberish.

Natural language generation is also an NLP task, but this particular library doesn't seem to tackle it at the moment.

This sounds interesting. Can you specify an example usecase (what would be the input data and how would generated natural language look like), and I will try to see if I can do it.

I think the only use case is spam. But it's a big one.

For example, financial data would be used as input to generate a daily stock market overview. Something along the lines of:

"Today, the Dow hit a high of 16,200, marking the first time it has crossed the 16,000 barrier. blah blah blah, etc"

Basically, use data points to create a market overview where readers wouldn't know that it was computer generated. That's one idea.

Good hack, now I'm following your github!

Reply on news.ycombinator.com