As can be seen, PTI has the highest percentage of Pro
predictions for each day, which considerably outnumber the
Pro predictions for both PMLN and MQM. Consequently,
both PMLN and MQM have very high Anti predictions. These
results clearly show that twitter sentiments predict PTI as the
winning party, and at the same time, predict PMLN and MQM
to lose heavily. In order to further verify our results, we
manually labeled the tweets in each test matrix (as Pro or
Anti) and compared these labels with the predicted ones. The
results are shown in Table X. We made manual Anti
predictions for MQM and PMLN, and manual Pro prediction
for PTI. Also, the predictions are expressed as average over 4
days (8th-11th May). We can see that the Pro prediction is still
the highest for PTI, followed by PMLN and the MQM. This
result strongly reinforces our statement that Twitter sentiments
predict PTI as the winning party. Nevertheless, we do observe
a drastic increase in the Pro prediction for PMLN, which
signals a type of “bias” in learning. However, this bias is
associated with the users’ sentiments, and may required
extensive experimentation with different parameters and
algorithms (which is outside the scope of this work).