5
$\begingroup$

I am going to build machine learning algorithm to identify fake tweets. The data set has huge retweets which I think might be an issue. Do you think given that the focus is the original tweet, it is better to remove all the retweets?

Thank you,

$\endgroup$
1
  • 1
    $\begingroup$ It depends: what do you call a fake tweet? Is it a tweet not originally written by the user? $\endgroup$
    – Erwan
    Nov 12, 2019 at 23:42

5 Answers 5

1
$\begingroup$

No. I do not believe so and I can explain a few reasons why.

  • If an entity wants to create waves in twitter with false tweets retweets are probably apart of the plan.
  • If you want to detect tweets generated by bots looking at the statistical data on said tweets and retweets like time stamps could be relevant to detecting if the tweet is generated by a bot.
  • If You have a way of checking retweets by bots then removing all retweets would also remove that data.

You should remove retweets if.

  • The project is focused on analysis of text to determine if a tweet is bot or not.
  • There is no labeled human or bot retweet data.
$\endgroup$
1
$\begingroup$

There might be a chance that the retweet has an entirely different context compared to the original tweet. It is also possible that some retweets with different opinion/comment gain more popularity than the original one.

In these cases I don't think you can classify them as fake tweets.

You can classify tweets as fake when they are widely retweeted but with no context, One such example is retweets due to a giveaway or charity.

If you can figure out how to separate the spam retweets and original tweets it would help for better analysis and accurate results.

$\endgroup$
1
$\begingroup$

Removing retweets would make sense to me, if you want a model where each single original post has the same weight. However, it doesn't make sense to me if you want to measure each original tweets impact, except if you add one of the below features:

  • amount_of_total_people_reached (measured in e.g. total impressions from the original tweet and all retweets)
  • amount_of_people_reached_via_retweets (measured in e.g. total impressions from retweets)
  • amount_of_retweets

The exact purpose of the model and how you want it to fit to the real wolrd matters here I suppose.

$\endgroup$
0
$\begingroup$

To me it depends on what you want to focus on : do you want to create a model dealing with original posts that are fake news, and then make an algorithm finding the original from a retweet then applying your model ? Or do you just want a model that takes one tweet, not looking if it's a retweet or not, and trying to guess if it's fake or not.

In the first case, you should remove them, because you'll have many information about the people retweeting fake news, while you only want to find info about origin posters, which will make your model biaised. In the second case, of course, since that's exactly what your model aims to do, you should keep them.

$\endgroup$
0
$\begingroup$

Removing retweets from your data set could potentially help improve the model's performance by reducing noise and focusing on the original tweets, but it could also result in the loss of valuable information that could be useful for identifying fake tweets.

One approach you could consider is to keep the retweets in your data set, but to also include additional features in your model that capture information about the retweet, such as the number of times the tweet has been retweeted, the user who initially tweeted the message, and any other relevant metadata. This could allow your model to use the information from the retweets in a more structured and controlled manner, potentially improving its overall performance.

Experiment with different approaches and compare their performance to determine the best approach for your specific use case.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.