The abstract of Project 3 (Alexey Khrustalev, Ira Kokoshko, Anastasiia Todoshchuk)
Research Questions
The main research question of ours is to find out if the same topic (Climate Change) is discussed differently on YouTube at various moments in time. While using the language model and clustering algorithms, as well as during the manual analysis of the YouTube comments, we discovered that the discussion changes the topic from the original one to politics or some social issues. The main world news and events clearly affect the rhetoric and arguments of the commentators, as well as the overall mood of a discussion. It was also noticed that the political views of the commentators greatly influence their attitude towards the discussion. That’s why it is interesting to compare the opinions on the same topic from different times and to find out whether new scientific discoveries or political events have an impact on its discussion.
Material and Data
The time period of ten years with five-year intervals was chosen, as we believe that a lot of different events happened during this time that could heavily affect the discussion. That being said, we will compare the comments from 2010, 2015, and 2020. We will see to what extent the topics such as elections, crises, pandemics, etc if they are parts of the discussion, influence the debates. The case when there is no such influence or where we can’t clearly see it is also possible. In this case, we at least will get the information about the quantitative distribution of the comments across the topics.
Technology and Methods
The language model, which was used in the previous assignments, will be used. In addition to the model, the clustering algorithm with different parameters will be applied. These tools will help us to answer the different questions, e.g. which topics surround the main one (climate change).
The model and clustering algorithm would also be useful to retrieve the information about the most/least popular (according to the number of likes) comments, the most/least supported ones (according to the number of relatively same comments).
Analytical Approach
We don’t know if our starting assumption will hold as we proceed with the exercise. Grounded theory methods would allow us to analyze the data already at the early stages of the work and adjust the strategy and questions along the research path. Since the Grounded theory guidelines let construct theories at the early research stage from available data, we can begin with the initial data and ideas and switch between data and analysis using comparative methods. Thus, new insights gained during the work on the project will contribute to the formulation of new, additional questions and hypotheses.
Moreover, there are a few approaches to comments analysis. On the one hand, we can analyze the comments for the selected videos without filtering the comments by time range and define the most frequent topics of discussion for each year-related video. On the other hand, we can filter the comments by date and analyze them separately for each video. In this way, we can take into account the time range, as the set of topics for discussion could be changed with time for every video due to the appearance of new commentators that have a modern point of view on old events/videos. The second approach can be especially useful for videos released in 2010 and 2015.
In our opinion, it would be best to combine all three approaches discussed in the article “Mapping Controversies with Social Media: The Case for Symmetry”.
A precautionary approach would be very useful to get rid of the biases caused by YouTube. For example, we know that YouTube has its own algorithm for video advertising, which may drastically influence the topics to be discussed. It would probably make sense to search not only for the most popular video but try to find the videos representing different points of view.
An affirmative approach would come in handy to not waste time on the videos without a discussion and debates in the comments. One can use specific signs that show that a particular video is not suitable for our research goals. For example, the number of comments, the number of views, the number of likes and dislikes, and so on.
An empirical approach could be helpful in understanding what type of discussion one could hope for. For example, we, as researchers, may need to take a look not only at the video itself but also at the channel and description section. If there are links to some scientific research, this could lead to the discussion to be more scientific, and maybe more comments would be on the topic stated in the video. Furthermore, we have to take into account the sorting algorithm of YouTube. It shows us the most popular comments or the newest comments first. This could affect the impression one could get from the comment section. Luckily, we can deal with it by using the language model and clustering algorithm.
The abstract of Project 3 (Alexey Khrustalev, Ira Kokoshko, Anastasiia Todoshchuk)
Research Questions The main research question of ours is to find out if the same topic (Climate Change) is discussed differently on YouTube at various moments in time. While using the language model and clustering algorithms, as well as during the manual analysis of the YouTube comments, we discovered that the discussion changes the topic from the original one to politics or some social issues. The main world news and events clearly affect the rhetoric and arguments of the commentators, as well as the overall mood of a discussion. It was also noticed that the political views of the commentators greatly influence their attitude towards the discussion. That’s why it is interesting to compare the opinions on the same topic from different times and to find out whether new scientific discoveries or political events have an impact on its discussion.
Material and Data The time period of ten years with five-year intervals was chosen, as we believe that a lot of different events happened during this time that could heavily affect the discussion. That being said, we will compare the comments from 2010, 2015, and 2020. We will see to what extent the topics such as elections, crises, pandemics, etc if they are parts of the discussion, influence the debates. The case when there is no such influence or where we can’t clearly see it is also possible. In this case, we at least will get the information about the quantitative distribution of the comments across the topics.
Technology and Methods The language model, which was used in the previous assignments, will be used. In addition to the model, the clustering algorithm with different parameters will be applied. These tools will help us to answer the different questions, e.g. which topics surround the main one (climate change). The model and clustering algorithm would also be useful to retrieve the information about the most/least popular (according to the number of likes) comments, the most/least supported ones (according to the number of relatively same comments).
Analytical Approach We don’t know if our starting assumption will hold as we proceed with the exercise. Grounded theory methods would allow us to analyze the data already at the early stages of the work and adjust the strategy and questions along the research path. Since the Grounded theory guidelines let construct theories at the early research stage from available data, we can begin with the initial data and ideas and switch between data and analysis using comparative methods. Thus, new insights gained during the work on the project will contribute to the formulation of new, additional questions and hypotheses.
Moreover, there are a few approaches to comments analysis. On the one hand, we can analyze the comments for the selected videos without filtering the comments by time range and define the most frequent topics of discussion for each year-related video. On the other hand, we can filter the comments by date and analyze them separately for each video. In this way, we can take into account the time range, as the set of topics for discussion could be changed with time for every video due to the appearance of new commentators that have a modern point of view on old events/videos. The second approach can be especially useful for videos released in 2010 and 2015.
In our opinion, it would be best to combine all three approaches discussed in the article “Mapping Controversies with Social Media: The Case for Symmetry”. A precautionary approach would be very useful to get rid of the biases caused by YouTube. For example, we know that YouTube has its own algorithm for video advertising, which may drastically influence the topics to be discussed. It would probably make sense to search not only for the most popular video but try to find the videos representing different points of view.
An affirmative approach would come in handy to not waste time on the videos without a discussion and debates in the comments. One can use specific signs that show that a particular video is not suitable for our research goals. For example, the number of comments, the number of views, the number of likes and dislikes, and so on.
An empirical approach could be helpful in understanding what type of discussion one could hope for. For example, we, as researchers, may need to take a look not only at the video itself but also at the channel and description section. If there are links to some scientific research, this could lead to the discussion to be more scientific, and maybe more comments would be on the topic stated in the video. Furthermore, we have to take into account the sorting algorithm of YouTube. It shows us the most popular comments or the newest comments first. This could affect the impression one could get from the comment section. Luckily, we can deal with it by using the language model and clustering algorithm.
References: [1] Marres, Noortje, and Moats, David, Mapping Controversies with Social Media: The Case for Symmetry (February 20, 2015). Available at SSRN: https://ssrn.com/abstract=2567929 or http://dx.doi.org/10.2139/ssrn.2567929