Summary

This paper focuses on the hate and violence activity of far-right YouTube channels and viewers. They used YouTube channels and videos that were related to Alex Jones and InfoWars who are categorized as far-right. As well as a baseline of the top 10 news channels on YouTube. Their analysis included research on the lexicon, topics, and bias of the videos and comments. They concluded that right-wing channels have more niche and targeted content that is both aggressive and violent (specifically the Muslim community).

Reflection

I think this paper is one of those candidates for “Everything is Obvious” where many probably wouldn’t be surprised with the results. But after seeing the data and methodology it’s clear that any reader could learn something new from the many of the results or reflections within the paper. Seeing the sheer number of statistics gives us a great idea of just exactly how much these channels’ behavior gets exacerbated with their interactions. But on the flip side of that, one thing that bothered me was that it seemed the authors chose data and methodologies that would compliment the results they were looking for.

I found their choice for a baseline to be a little odd. For the Right-wing channels they carefully reviewed every single channel in their category but just chose the first top 10 “news” channels for the baseline. The most popular, YouTube Spotlight, really doesn’t have much news (or anything politically related). A quick glance at their recent uploaded shows videos like the YouTube Rewind, fashion videos, and music videos. Even though the baseline channels aren’t intended to represent “neutral” users I’m not sure what they are supposed to represent.

Another thing I noted is that this paper also was similar to the first reflection read in which there is a lot of data dumping happening. They seemed to over explain the details and methods they used to an extent that went over my head. But also with that said, the amount of analysis that went into the research is astounding. They were able to pull out and compare the captions and comments to a greater extent than I could imagine. One example of this are the different terms under the negative and positive showing the aggression vs anger vs disgust percentage difference between their categories.

Regardless of some of the flaws, there definitely could be a ton of future work in this domain:

• Instead of this baseline, have different categories like the far-left, left, right, and neutrally perceived channels and see how they differ from each other.
• They also suggested a temporal look at the channels, but to add to it we could look at the extent to which hate and violence propagates or grows as channel gets more viewers and subscribers.
• Can we gauge also the reactions to these videos and comments? For this it might be useful to take a look at the number of dislikes or reports on these channels as well as banned users.

Read More

Early Public Responses to the Zika-Virus on YouTube: Prevalence of and Differences Between Conspiracy Theory and Informational Videos

This paper focuses its research on the initial user reaction to informational and conspiracy videos on YouTube about the Zika-Virus. Their data set was the 35 most popular videos on the issue. Through their analysis, they were able to find that most (23) of the videos were purely information while the remaining (12) offered different conspiracy theories. They found that user activity between the videos seemed to be the same, whether or not the content was informational or conspiracy, with low user engagement. The difference between the two was within the contents and sentiments of the reactions. Typically, informational responses focused on the disease itself and how it affects babies and women in Brazil while conspiracy responses seemed to to target certain people or demographics for starting the virus.

One thing that surprised me was how oddly specific the subject matter was. The authors chose to research YouTube videos from a virus that had appeared two years earlier. I thought that such a targeted subject would be very limited in its analysis but it does make some sense as most, if not all, of the content for the Zika-Virus has been uploaded and the data set is non-continuous. I would have liked to see them tackle multiple different health issues on YouTube in order to avoid different biases that may appear for specific subjects. That being said, the scope for the project was smaller as a result and might be a great example for choosing a feasible semester project.

One thing that didn’t make sense to me was why the authors chose only the “first phase” of the Zika-Virus in 2016. It seems they want to test the initial response of the population but don’t seem to explain why. How do the initial phases differ from the entire ordeal altogether? I think that the results and types of user interactions would have the same if not similar to the found results, but the data-set would be much larger and reliable than the 35 videos they were able to find.

Some additional research questions that could come out of this:

• Just given the comments of individual YouTube videos, can we use them to predict whether or a video is a conspiracy or informational video?
• Along the point about the “initial phase” could we compare the different phases of the Zika-Virus and the user interaction / dissemination?

Automated Hate Speech Detection and the Problem of Offensive Language

This paper researches the differences between hate speech and offensive language. Their data set consisted of 25,000 tweets that contained speech lexicon found on Hatebase.org. Through their analysis, they found that it was quite easy to classify offensive language, with a 91% prediction rate but had a little more trouble identifying hate speech with a 61% prediction rate. They found that those that were correctly identified as hate speech often were extreme and contained multiple racial and homophobic slurs.

What I would have liked to see are more explanations on the features that they used. What are the TF-IDF, POS, and reading score methods they decided to use and why are they helpful to the research? This is something I found useful in some of the other papers we have read so far.

Some additional research questions that could come out of this:

• Using the terms generally associated with hate speech. Can we find a percentage of tweets that are hateful / offensive that contain them?
• As noted in the conclusions, another topic would be to look at the users who have been known to use hate speech by looking at their social structures and motivations. But also I think it would be interesting to see data on their demographics and how they might change over time as well considering the responses to their hate speech.

Read More

Summary

The goal of this paper was to identify specific statistical differences between fakes news and real news. The primary concern of this paper focuses more on categorizing the different features of fakes news rather than similar papers that may try to classify only the difference between satirical pieces. Using a wide variety of stylistic, complexity, and features, they were able identify fake news 71% and 78% of the time when compared to real news. Along these lines, they also could correctly classify 91% of satire news to real news. Lastly, they had a harder time identifying satire piece to fake with a 55% and 67% success rate.

Reflection

Much like the last paper, I didn’t find much of the results from this paper to be surprising either. But even so, I believe they did do what they set out to study in the first place, which was to be able to find features the can categorize fake news from real news. Even though I wasn’t surprised, I felt as though the paper was well written. The authors followed the data science workflow and succinctly described everything they were doing and why. Specifically, their explanations of the limitations on their datasets and how they tried to combat them was quite thorough. One thing I did notice was that they explained their reasoning with using ANOVA, but completely left off why the Wilcoxon rank sum would work when ANOVA wouldn’t. As we read more articles, I think this paper could be a great reference for structure and linguistics going into the semester project.

One thing I felt was missing from the paper was reflection. They did very well in telling us the who, what, how, and why but don’t really offer ideas on what all of these conclusions might mean and what this research could lead to. I have a few idea on different research branches that could come from this:

• Maybe try to do the same type of classification work, but use news articles from different geographical areas or languages. This could be used to test whether or not fake news can be generalized and studied in a broader light.
• Although a bit tangent, I think it would be interesting to specifically look at the positive sentiment differences between the three types of news pieces they defined. They looked at negative sentiment but I wonder if they were to add positive sentiment into the mix, would they be better able to distinguish satire from fake news?

In regards to the upcoming semester project, I liked the general methodologies and statistical analyses that the authors used in their research. I might look to do something that may be similar to this type of research with natural language processing and lexical analysis. Using many different types of features and combining them into a coherent argument looks impressive to me and I wonder how they were able to combine all of these factors together and even form a classifier from the results.

Read More