A credibility analysis system for assessing information on twitter
Download as DOCX, PDF0 likes31 views
The document describes a credibility analysis system for assessing the credibility of tweets. It uses four components: 1) a reputation based component that calculates a reputation score based on account features like followers, retweets, favorites, etc., 2) a classifier engine like Random Forest or Naive Bayes to classify tweets as genuine or fake, 3) a user experience component using sentiment analysis to analyze the positivity or negativity of tweets, and 4) a feature ranking algorithm to select important tweet features. The system is implemented using a Twitter dataset, calculates reputation scores, extracts features, runs random forest and the proposed FeatureRank NB algorithm for classification, and compares their correctly classified instances.
A credibility analysis system for assessing information on twitter
- 1. Venkat Java Projects Mobile:+91 9966499110 Visit:www.venkatjavaprojects.com Email:[email protected] A Credibility Analysis System for AssessingInformation on Twitter Now-a-days twitter popularity is growing due to its fast dissemination (spreading information) of messages and for researchers finding Credibility (finding whether message is fake or genuine) of messages is a hot topic and lots of researchers already introduces some techniques to find whether given message in twitter is fake or genuine but all those techniques were not using all available information to determine fake or genuine messages. Sometime in all online social networks some malicious users will spread fake news and if this news publish on social media then it will put bad effect on human society. This malicious users will hire peoples or robots to create fake account and then using that fake account they will publish fake messages. This type of accounts will not have much profile features such as favourites, followers, following, hashtag, retweets etc. So by analysing those features we can determine whether tweets is credible (genuine) or non-credible (fake); In propose paper author is using four different components to find whether tweet message is credible (genuine) or non-credible (fake). Below are four components used in algorithm to find user reputation and tweets credibility. 1) Reputation Based Component: In this technique we will identify number of followers, following, retweet, hashtag and favourites from tweets dataset to calculate user’s reputation. For example if tweet is more genuine then more number of users show interest in that tweet topic and number of retweets, favourites count will be automatically increased. Followers/ following of such tweets will also be increased. If tweets is not genuine then very less number of users will follow it and reputation score will be less. Formula to calculate reputation Math.log(favourites)/math.log(max(U_followers,Total_retweet_for_topic) Similarly reputation score will be calculated for hashtag also, sum of favourites and hashtag will give reputation score. If user has less followers or retweets then its score will be less. If calculated value is less than 0.1 then we consider this user tweet as Non- Credible (fake) and if greater > 0.1 then consider as Credible (genuine)
- 2. Venkat Java Projects Mobile:+91 9966499110 Visit:www.venkatjavaprojects.com Email:[email protected] 2) Classifier Engine: using data mining algorithms such as Random Forest and Naïve Bayes or propose FeatureRank_NB we can classify/predict given tweet is genuine or fake. First with all existing data a train model will be generated with above algorithms and then this algorithms train model will be applied on new test tweet to determine whether given test tweet is genuine or fake. Propose FeatureRank_NB algorithm will apply ranking algorithm on all features such as favourite, followers, following, hashtag etc to determine relevant (important values) features to predict tweets are genuine or fake. Ranking algorithm will give high score to those attributes which occur more number of time and that attribute consider as important and more weight or rank assigned to those attributes. 3) User Experience Component: In this module we will apply sentiment analysis algorithm to determine whether given tweets contains more number of positive or negative words. If user is experience and genuine then he will used more number of positive words and if user is fake then he will used more number of negative words. By applying sentiment detection we can detect user experience and help us to identify whether user tweet is fake or genuine. To calculate sentiments we are using Stanford Natural Language Processing API’s which classify given tweets as Positive or Negative or Neutral. 4) Feature Ranking Algorithm: This module will be applied on all tweets features to select only those attributes which are important and has more rank. All algorithms given in paper will work on above describe four components. Implementation To implement above paper concept we are using tweets dataset from twitter. Screen shots. To run project double click on ‘run.bat’ file to get below screen
- 3. Venkat Java Projects Mobile:+91 9966499110 Visit:www.venkatjavaprojects.com Email:[email protected] In above screen click on ‘Upload Tweets Dataset’ button to upload tweets folder. Each file in tweet folder contains tweets and profile from one user After uploading tweets folder will get below screen
- 4. Venkat Java Projects Mobile:+91 9966499110 Visit:www.venkatjavaprojects.com Email:[email protected] In above screen we can see we extracted all features from all user profile. In above screen ‘Tweet Content’ column contains tweets text message. Now click on ‘Calculate User Reputation’ button to calculate user reputation from above extracted features In above screen for all tweets from each user we calculated reputation score and also calculated whether tweets contains positive or negative messages.
- 5. Venkat Java Projects Mobile:+91 9966499110 Visit:www.venkatjavaprojects.com Email:[email protected] Now click on ‘Extract Features’ button to extract features and to calculate credibility score from those tweets features In above screen for each user tweet we calculated credibility based on sentiment and reputation score. If score less than 0.60% then tweets will consider as non-credible tweet and > 0.60% consider as credible tweet. Now from above features we will train Random Forest Algorithm to build classification model and to calculate correctly classified instances (record). Now click on ‘Run Random Forest Algorithm’ button to build train model.
- 6. Venkat Java Projects Mobile:+91 9966499110 Visit:www.venkatjavaprojects.com Email:[email protected] In above screen selected text we can see total 36 tweets users found and out of that Random forest able to correctly classified 27 tweets. Now classify same thing with propose‘FeatureRank NB’ classifier, now click on ‘Run Feature Rank NB’ button to classify with propose features ranking technique In above screen we can see propose feature ranking technique able to correctly classified 28 records which are more than existing random forest technique. Now click on ‘Correctly Classified Instances Comparison Graph’ to see correctly classified instances between existing and propose technique
- 7. Venkat Java Projects Mobile:+91 9966499110 Visit:www.venkatjavaprojects.com Email:[email protected] In above graph x-axis represents algorithm name and y-axis represents correctly classified count. Existing random forest has less number of correctly classified instances compare to propose Feature Ranking NB algorithm