Inspiration
The project is inspired from the idea of creating best of the content for tweets in least of the time. Most of the brands spend a lot on marketing on social media especially twitter where millions of people like and follow the content that is popular and engaging. For a small tweet, it becomes important to design it smartly such that it impacts maximum amount of people.
Through this project, we aim to help companies and brands come up with tweets that are projected to be highly popular and engaging as per the latest ongoing interests on twitter among the masses.
What it does
Given the name of the brand and target location, our project gives you the most relevant and engaging keywords, hashtags and tweets that can either be retweeted or these keywords and hashtags can be added to the new tweets company aims to design. We also provide the user with a summary of overall sentiment of the public associated with the recent tweets in order to introspect their social media performance and take needed next steps for marketing.
How we built it
We made use of IBM API for Sentiment Analysis of the relevant tweets for the brand and location. We extract the most relevant and popular tweets related to the brand and based on the location, interests of the people on twitter, popularity of the type of content, we come up with the keywords that have a positive sentiment associated with them and hence the brand. We use the NLP API for the relevance score and TF-IDF model for getting a more exhaustive list of hashtags. We provide with relevant hashtags based on how many times a particular hashtag is used in the latest tweet i.e. the popularity of the tag. The overall summary of the emotions and sentiments is assessed from the latest tweets and retweets people have done for the tag. Technologies Used: i) Flask (for backend) ii) React (for frontend) iii) Sklearn and Pytorch for NLP Models
Challenges we ran into
The major challenge we ran into was the need for the keywords to be relevant to the Brand we are looking for. Going simply by TF-IDF metrics did not ensure the keywords to be relevant and popular. So we use a transformer based NLP model to extract the relevance score of the keyword and move forward with the keyword only if the tweet considered has a positive sentiment associated with it.
Accomplishments that we're proud of
It was our first time working as a team completely online, where we managed to connect frontend and backend as well as make the relevant API calls in which we were not so experienced. As a team, it was nothing short of an exhilrating experience where we came up with a successful product at the end.
What we learned
We learned the importance of building a great supporting backend code which handles all the errors and edge cases and doesn't fail to output even in case of errors. We also learned to have a user experience at par with the best by adding functionalities such as summaries with pie charts, brand name and location suggestions while typing and entering the input on the project, etc.
What's next for Smart Tweet
The aim forward is to have complete recommendations of the tweet through sentences and phrases instead of just keywords, hashtags and sentiment summary. We are also looking forward to have suggestions for the tweets on the fly i.e. while creating or typing your tweet, where you need not do an analysis before starting to write a tweet itself.
Log in or sign up for Devpost to join the conversation.