The Twitter Sentiment Index

Twitter graphing public opinion of presidential candidates

2 followers

Visit website
Backed By Research, The Twitter Sentiment Index (TSI) Is A Method For People To Track The Public's "Feeling" Of A Presidential Candidate in near real-time.
The Twitter Sentiment Index gallery image
The Twitter Sentiment Index gallery image
Launch tags:Web App•Politics•Artificial Intelligence
Launch Team
Framer
Framer
Launch websites with enterprise needs at startup speeds.
Promoted

What do you think? …

CJ Ezinne
Maker
Hey Everybody, My name is Chidubem Ezinne, call me CJ. Over the past 10 years, Twitter has become a prominent method of communication in the United States. And more recently it has taken center stage for its prominent use by significant politicians and this has led to a myriad of political commentary to take presence on the platform. After reading some older research on using twitter to analyze public sentiment; I decided to work on this as a side project and see what I could get built. What came out of this was The Twitter Sentiment Index. ✅How it works: - Analyze the latest tweets on each candidate every 15 min - Analyze each tweet's sentiment individually then save the overall average score - To reduce the "noisiness" of the data I've provided a graph based on the Moving Average ✅Issues and future works: - State Specific Analysis - What is the best method of displaying this data? - Implicit bias to the population with Twitter Accounts
Noah Mittman
@cjezinne1 How are you dealing with bots?
CJ Ezinne
Maker
@noahmittman Thanks for checking it out, Noah! Bots are tough to deal with, contextually the blend in well with normal tweets. As of right now, we've significantly limited bot input by focusing our twitter search to Text only tweets (No Retweets, or tweets with media). So far there are a lot of normal real political discussion than bots captured. Another attempt at slowing their effect is that by using a moving average, extreme sentiment scores can be repressed... now I did have an idea of utilizing a 20% Trimmed mean to cut back those all together, but I have to investigate that a little more.
Joe Ward
@noahmittman @cjezinne1 Although the volume of data is most likely prohibitive, you could look for characteristics of the accounts themselves to try to flag bot accounts based on some characteristics common to them. However, the work load may not be worth the effort.
Nils Siggurdsson
Wow, really cool! Would be so interesting if this could be correlated with FiveThiryEight.com data or other poll data to say something about how Twitter sentiment relates to future polling and/or results (if any). And: Would you be able to scrape profile data to understand more about the differences in Twitter sentiment among different demographics?
CJ Ezinne
Maker
@casinobonustip Wow, thanks for the reference to Five Thirty Eight, I'll see what I can do to integrate and improve upon the data. On the topic of predicting future polling, I have been a little wary of promoting this as a potential tool for that, however my referenced research have studied the matter with a similar solution like this for the last couple presidential elections and have had prominent results with older NLP Algorithms. I have linked to all the research on my personal site CEzinne.com Thanks for the comment, I truly appreciate it!
Terry Crandall
Cool idea! Given That primaries are local and create momentum in the election, have you considered localizing the data by primary calendar? You have Warren in first right now but she’s in life support in the early primaries. We might be interested in featuring this on votingsmarter.org. We are launching something here soon too.
CJ Ezinne
Maker
@terryatvotingsmarter I would love and appreciate the feature. if you do decide to feature can you also link and include my name and website (Chidubem (CJ) Ezinne www.CEzinne.com)? On the topic of Warren's lead. I am currently researching a method of weighing our data with current delegate numbers to see if any interesting trends appear. This data isn't meant to predict the next president, but to answer an interesting question. As of right now (or the last day), how have the public been talking about me? Positively or Negatively? Nonetheless, it is a work in progress and I always plan to iterate and add new features as needed.
CJ Ezinne
Maker
@terryatvotingsmarter I also want to add, randomly enough #PresidentWarren has been trending in the United States today and it's interesting how this chart picked up this uptick in public sentiment.
Hans Dekker
This is awesome. Can it track other tags such as company names?
CJ Ezinne
Maker
@hans_dekker That's actually really interesting... I'll add that to my list. Using this code base I can make a separate product to track user inputted topics.
Edfil Basan
Did you use anything in particular for sentiment analysis? Similar to your debate mode, I did something similar for Game of Thrones characters while the last season of the show was airing. I used TextBlob for that. https://edfilbasan.github.io/got...
CJ Ezinne
Maker
@edfil_basan Textblob looks interesting! For this project I used both Google's Cloud Natural Language API and Microsoft's Text Analytics API (Mostly because I have unused credits lol) and the results have been great. I'm currently learning more about the field of NLP and I'd love to make a custom model one day soon.
Kirsten Lambertsen
This is wonderful! Not that he's my pick, but I'm curious why Bloomberg isn't on it. Has he not met a threshold yet or something?
CJ Ezinne
Maker
@mspseudolus lol yea, that was my fault. Simply forgot. I’m adding him tomorrow but will have to wait at least 3 days to compile enough data to add him to the graph
Eileen Campbell
Need to get Bloomberg in there!
CJ Ezinne
Maker
@eileenatreidcampbell lol yea, that was my fault slip of the mind. Will need to wait at least 3 days worth of data before adding him to the chart!