Uses NLP algorithm trained on a RNN to analyze sentiment from a database of tweets. A website returns a pie chart of the distribution of sentiments on the selected keywords. Hosted on google cloud vm
Made by Calvin Madsen, Pravar Annadanam, John Slater, and Aadit Bennur.
From the frontend, we allow the user to meet the team, interact with and learn the project idea.
On the main page, the user can enter a keyword and a time range to run a query and gather data.
We posit that the data was supposed to run through flask into a python script that connected frontend and backend with the NLP model we built.
The algorithm:
We used tensorflow, trained it on a natural language dataset, and got it to analyze sentiment.
It takes tweets from a database that we compiled and returns one of 6 sentiments: joy, sadness, anger, surprise, love, and fear based on its analysis of the sentence(s).
With this data, we compile it and send it to the python script which reads it and formats it such that we can display a pie chart on the site of the results of the 6 sentiments that we get back.
Purposes:
Possible can be used by individuals and companies to see recent trends and the general perception of them, their business, or some keyword that they want to search for. With more time to develop our product, we could have added more features and functionality, but we definitely feel that we took great advantage of the time frame and that each one of us gained skills that will prove useful in the future.
Future Endeavors:
The next steps would be to train the model more given more time and more data. Furthermore, on the frontend, creating a way for users to export data would help. Finally, and most importantly to language on twitter, would be to integrate emojis into the overall processing of the algorithm and finding a way to truly extrapolate that data into real business logic and reason. Integrating it with bloomberg terminal, so that users could see the perception of stocks by the public, finding ways to segment the population of twitter in some ways based on language, and thereby making it possible to separate the data into strata would allow us to make more complex models, increase our accuracy and precision, and decrease the bias by grabbing data from many different groups.
Log in or sign up for Devpost to join the conversation.