How We Do It

In order to get an idea of what people think of the President we decided to analyse the tweet messages generated on Twitter. Twitter was selected because we considered it to be the largest open source of people’s opinions as over 300 million messages are generated in a year. It is also believed that 10% of the messages generated in Nigeria are related to political issues and politicians.

The Data

We source thousands of messages every day on Twitter that are related to President Buhari. We only select messages that are generated within a 50km radius of the borders of Nigeria and geo-locate the tweets to Nigerian states and major cities.

Our Calculations

We then apply sentiment analysis on the data to understand how effective both campaigns have been in getting their messages across and how their respective candidates match up when it comes to major political issues.

Each tweet is passed through our sentiment analysis engine which was built using a natural language toolkit. The sentiment analysis engine was trained with 5,000 tweets messages to identify the sentiment in each of the tweets. For each message, the engine gives two scores, the first gives the probability of the message having an overall positive sentiment and the second score indicates the probability of the message having a negative sentiment. After the training, in 70% of the cases monitored, the engine was correctly able to identify the sentiment of the message when the sentiment score was 55 or over.

We used the criteria of a score of 55 or over to select the messages which expressed a sentiment. To get the overall sentiment rate, we analyse the tweets from the previous two weeks from the day of the reporting, as it is considered the aggregated data gave more insight to identify the trend of sentiment rather than having a snapshot on one particular day.

The sentiment rate was determined by looking at the total number of people who sent a tweet that had been identified to have a positive or negative sentiment and then divide that by the total number of people who sent a tweet related to President Buhari over the same period.

When get the rates for each of the states or the cities, only the tweets that have been geolocated to the area being analysed.

SR(S_e) = (TP(S_e, P_gc, I_t) / TP(P_gc, I_t)) x 100

SR(S_e) = Rate for the particular sentiment being measured, either positive or negative

P_gc = Geolocation, if a particular geolopoitlical zone, state or city is subject of the measurement

I_t = The political issue to be tracked with the sentiment, if any.

TP(S_e, P_gc, I_t) = Total number of people who expressed the sentiment being measured.

TP(P_gc, I_t) = Total number of people who tweeted over the same period.