 who is leading the debate on social media, who is that one person that everybody mentions and that keeps coming up on our feed. Today we will learn how to construct a network of mentions and find the most popular account in the bunch. We will use the Twitter widget to retrieve 500 tweets in English with the hashtag machine learning. This will output a subset of the debate on machine learning with all the hashtags, ad mentions, text and so on. A quick look in a corpus viewer gives us an idea of what the corpus is about. Great! Now we are ready for some pre-processing. Our aim is to retrieve only ad mentions. First, I will remove everything I don't need lowercase transform and filtering for example. Secondly, I will select the tweet tokenizer since we are dealing with tweets. This will keep the ad mentions as they are. Finally, I will add a regular expression in the filtering section which will keep only words that start with ad. The expression can be copy pasted from the description below. Now our data is ready for the final part. I will use corpus to network widget to compute the network of mentions. The widget outputs two things. One is a network of documents where an edge is created between two documents if they share at least the number of words specified in the threshold parameter. The other is a network of words where an edge is created if the two words appear in at least the amount of documents specified in that same parameter. The latter is what we will use in this case as we wish to observe how words co-occur in our corpus. Also, we will set the window size to 100. This will create an edge between two words if they both appear within 100 words from each other. For tweets, this is sufficient to capture co-occurrence of ad mentions in the same tweet. To observe this network, we will use the network add-on which you can install in options add-ons. I will connect network explorer to corpus to network. Besides sending in the constructed network, I will also add node data which will give us interesting information about the nodes. In the network explorer, I will set the size of the node to word frequency. That is the number of times the word appears in the text. Finally, I will label the nodes with words. The network I get is interesting. It seems like there is a larger debate here with several people talking to each other. However, the most mentioned person does not get mentioned with the others. Today we have learned how to retain only mentions in preprocessing, construct a network of words from Twitter and how to explore the constructed network. Stay tuned for more text mining videos.