4 ms·
Great, simple concept. It might help a little if the tweet counts were normalized to control for the relative frequency of the terms. ie, people use the word "l
by hooande 13y ago
Great, simple concept. It might help a little if the tweet counts were normalized to control for the relative frequency of the terms. ie, people use the word "love" more often than "hopeless" in general. [1]
I'm not trying to nitpick, this is an inspired UI. It's just that a spike in something like "joyful" or "scared" could be interesting to see in relative terms, especially when calculating the positivity index at the top. I know that statistical accuracy isn't necessarily the point here, but working with twitter data can be tricky. I'm sure that engineering it was hard enough. What an interesting visualization.
[1] For example when looking at a standard corpora like the NLTK movie reviews data set, the term "love" appears 1,588 times vs "helpless" appearing 47 times.
- warcode 13y agoThe information on the page says this is directly related to counting #hashtags, and not any natural language analysis. I do not know if normalization would help in this instance as people are taking a deliberate action in hash-tagging with the related feeling. I've done some sentiment analysis and relationship/influence mapping on twitter data and the 140 character limit often means you have to create a specialized training set for your classifiers based on the group of people you are targeting. Simply weighting existing training sets offer very small benefits in accuracy.
- choult 13y agoClearly NLP/sentiment analysis on discrete units of 140 characters with no extra context is problematic. However, the state of the art is accurate enough that it's useful at higher volumes (eg. Twitter decahose scale). Still, love the simplicity of this UI and concept.