Privacy & Security in Online Social Media · End Term · Jan 2023 · AN

Suppose we build a co-occurrence matrix of size m × n, where each row corresponds to a word in the vocabulary and the columns corresponds to the context of the word. Which of the following could be a valid size of the co-occurrence matrix (select all correct answers)?

**[5 * 1 mark = 5]** 1. Give an example (other than the Facebook study that we saw in the class) for a natural phenomenon to study participants. 2. Give an example of HIPPA violation? 3. What are the advantages of word2vec (or similar algorithms) over bag of words representation? 4. Briefly mention a possible solution to reduce the tampering of the Twitter API. 5. State True/False. “Given the score received by post on Reddit, can we deterministically find out the number of upvotes or downvotes received by post?”

**[11* 2 marks =22]** 6. Give 2 example each for Push & Pull mechanisms on social media. 7. What is IRB? State its importance. 8. Discuss the issues in Twitter Sample API that led to biased samples? 9. An NLP pipeline consists of multiple stages, as we discussed in class. Specifically for social media text what steps are taken to convert raw text application-ready data?

10. Describe the challenges in identifying cyber bullying content from social media. 11. The y-axis depicts the visual clarity scores. Write a few lines of your interpretation from the graph (not more than 40 words).

12. Write one word for the following concepts. a. “The individual is separated from the group and freed from the observation of other persons. It is the most complete state of privacy that individuals can achieve. [1 M] b. “One man’s trash is another man’s treasure”. [1 M] 13. State atleast two differences among k-anonimity, l-diversity and t-closeness? 14. Give two examples of privacy enhancing technologies. 15. What could be the potential threat of sharing personal information such as birth date on social media? Give two potential threats. 16. How does YouTube promote radicalization?

**[4* 4 marks = 16]** 17. While discussing, linguistic characterization of divisive discussions, we saw that there were two methods used for predicting weather a discussion was contentious or not: a) TF-IDF features with Logistic Regression; b) BERT with Feed Forward Neural Network (MLP). a. Explain the rationale for choosing logistic regression? What advantage does it have over neural network classifiers? b. For a text classification problem, if you were to choose between BERT representations or TF-IDF, which one would you choose? Explain. 18. What are the different methods for data collection for human study? Name and describe any four methods. 19. Consider the graph below and compute the betweenness centrality of node 5. Show complete working in the solution.

20. Briefly describe the three types of validity in social data research with examples.

**[7 marks]** 21. Write a code to connect with Twitter’s streaming API. The stream should track tweets which mention the word “GDP”. When you hit the first tweet with the word “recession” in it, dump that tweet object to a Json file and disconnect the stream.