Twitter_bot_detection_using_lexicon

We are trying to build twitter bot detection model using sentiment analysis. Currently model worked on account level bot detection using adaboost, Random Forest, DecisionTree and Logistic Regression Classifier. But we are in the process of tweet level detection using recurrent neural network and LSTM.

Data

We used data from botometer(Link is provided below). Downloaded 2000 tweets for each 10K bots and 10K not bots. https://botometer.iuni.iu.edu/bot-repository/datasets.html

Feature engineering

a) Followers/Friends ratio

b) Tweets/day

c) Length of name

d) Length of screenname

e) Length of description

f) Number of words in name

g) Is their “bot” word in the name, description or screenname

h) Lexical diversity of description

i) In_Reply/tweet_downloaded

j) retweet_count/tweet_downloaded

k) fav_count/tweet_downloaded)

l) total_usermention/Tweet_downloaded

m) Variance in tweet rate

n) Number of nodes in network

o) Number of edges in network

Apart from this we genrated avrage of eight emotion and two sentiment for each account.

Model building

1. Decision Tree

2. Logistics Regression

3. Random Forest

4. Adaboost

Final result(on Adaboost classifier)

Results on test dataset are as follows: accuracy_score = 0.9345829428303655 cross_val_score = 0.9348178628684586 f1_score = 0.9344108250328885 confusion_matrix = [[2500 149] [ 200 2486]] precision recall f1-score support

      0       0.93      0.94      0.93      2649
      1       0.94      0.93      0.93      2686

avg / total 0.93 0.93 0.93 5335

auc = 0.934646097784269