Home

Twitter Sentiment Analysis

Objective

This doc describes the detailed design of our graduation project. The idea is to consume the Twitter API and perform a sentiment analysis over the tweets in order to classify each one to happy/sad or positive/negative -or even more categories in the future-. We could then store the classified tweets and construct some sort of a dashboard to visualize the data. The dashboard could be used in many ways, for example, users who are interested in understanding people’s opinion on a specific product or event could use the dashboard to search for the tweets about that topic and use the figures and diagrams that the dashboard offers to understand the general trend.

Overview

The project is consists of two main components, the first one is the model which classifies the tweets. And to build such model we need to go through three main steps:

Finding suitable training and testing data. Convert the text to a numerical vector of features that could be used to train the model. Try several classification models and choose the best one.

The second component is the dashboard, for this part, we found that elasticsearch could be used to store the classified tweets and Kibana to visualize the data. Kibana is a very powerful visualization tool, it gives us the freedom of choosing the best way to shape our data and it could be used to filter the data based on time, location or even specific text that users might be looking for.