Sorry for jumping like that before reading your blog post carefully but did you consider the fake user factor on your results? There are tons of accounts on Twitter posting 1-2 exactly the same or 2-3 character altered tweets from an unknown database per week. Those fake accounts mostly support AKP. I believe those are the ones that make your prediction inaccurate.
Hello yigit, Actually I did consider it. In short; out of the >200.000 users who posted Tweets regarding politics in the period of 1 May to 06 June there were a few dozen users who posted thousands of Tweets instead of only a few (<10). However, because there are only a few dozen users like this in a total set of >200.000 users, their actions do not affect the result much.
Unfortunately I did not have enough time to do a similar analysis on the Twitter data from Oktober. But I hope that the same conditions apply. In any case, I will take your concerns into account and try to do more analysis on the set of users from the dataset of September and Oktober if I have some time today or tomorrow.
I don't believe this will the result. The results of this snap election will be exactly like the last in which AKP won't be able to form government by themselves and a coalition will be required.