- Last 7 days
-
wu-msds-capstones.github.io wu-msds-capstones.github.io
-
observed survival time in the dataset, and the bold line represents the predicted survival times by our Weibull model.
Should I be concerned that these really don't look like they match up well at all?
-
ollowing table (Figure 11).
This is a table, not a Figure. The are referenced and numbered independently.
-
Figure 10
No mention of what the dashed vs solid lines mean in the figure or the caption.
-
Weibull Model of Survival Probability by Genre overlaid on KM Curve of Survival Probability by Genre
why should a reader care about this?
-
Weibull model
This was not defined earlier
-
In Figure 10, the d
explain before figure
-
We then fit a parametric model to this survival curve. We created a Weibull model and overlaid the resultant model predictions onto our survival plot (Figure 10).
Why did you do that here but not for the earlier situation? What is the motivation behind this?
-
We ran a log rank test on our genre Kaplan-Meier curve as well, and this test resulted in a p value of <2e-16. As such, we can say that genre absolutely plays a significant role in the sustained popularity of a given show.
Does this also mean that the drama and docu-series curves are individually significantly different?
-
Kaplan-Meier curve of how many weeks shows took to go from maximum (100) to minimum (0) Google Trends rating, based on their genre
what should a reader take away from this? Also, why the shading? That wasn't apparent in the previous plot.
-
This Kaplan Meier curve
explain before figure.
-
As such, we can say with confidence that there is a statistically significant difference in how long a show’s popularity is sustained based on the streaming platform.
This is across all the platforms correct. The log rank test works both across groups or across pairs?
-
s the Kaplan-Meier curve in Figure 8 shows, there is some difference between the streaming platforms. The two worst performing streaming platforms in terms of sustaining popularity are Peacock and Disney+. Peacock lags way behind the other streaming platforms; this can be attributed to Peacock’s relative unpopularity amongst streaming platforms. Peacock is the newest of all the streaming platforms tested, launching in July of 2020 (Comcast, 2020). As such, it has not had the time to build up an audience or a reputation as a platform with good original programming. Disney+ is a more interesting example. Disney+ is a very popular streaming platform, but not very good at creating original content that sustains popularity. We hypothesize that this highlights the strengths of Disney+: the catalog of Disney content. Disney+ houses all sorts of Disney Channel shows, Star Wars films, and Marvel films, all of which could be argued are the main draw of Disney+ as a platform, not their original content.
Explain before the image.
-
Kaplan-Meier curve of how many weeks shows took to go from maximum (100) to minimum (0) Google Trends rating, based on their streaming platform
And so what should a reader be concluding from this image?
-
Figure 8
It would probably be worth lighboxing these images so that they don't take you to a different page when clicked, but instead get larger on this page.
-
To do this, we decided to use Google Trends data.
Why this over the ratings, for instance?
-
Kaplan-Meier Curves
You can write definitions in markdown with
markdown word : definitionwhich might be better here than subheadings. -
To ensure clarity and understanding, we have provided these definitions below.
I appreciate this, as I haven't taken all these classes!
-
Descriptive Statistics
I saw no statistics in this section. Only EDA visualization. So don't conflate the two.
-
Figure 6:
You need a legend explaining the blue and green on these still.
-
Figures 6 and 7 highlight the Google Trends search interest and placement on the Nielsen top 10 streaming rankings for two streaming TV shows, both post-apocalyptic dramas based on video game franchises. Fallout was released in full on Prime Video on April 10, 2024. The Last of Us premiered on HBO (simulcast on HBO Max) on January 15, 2023 and ran until March 12. Fallout was on the Nielsen rankings for 7 weeks, spending 4 of those at #1 before falling to #7. The Last of Us was on the Nielsen rankings for 9 weeks, spending just one week at #1, but fell no farther than #4. Interest for both shows peaked when they premiered, but by a week later interest in Fallout had fallen and it was half as popular; it continued to fall, although showing small peaks. Interest in The Last of Us jumped weekly as new episodes aired, not dipping below the halfway mark until almost a week after the finale.
Move the explanation before the figures.
-
Fallout (Figure 6) and HBO’s The Last of Us (Figure 7).
I would stack these into the same figure if what you are wanting to do is compare across them.
To get a consistent x-axis I'd change it to days before or after release. Then the two should be comparable.
-
Figure 5
This is better than Fig 4, but it is still difficult to see some of the smaller variations b/c of the dominance of Netflix's all-at-once. Would a vertical log scale help this at all? That would seem preferential to making a cut on the vertical scale.
-
* *the shows whose release schedule changed were classified as ‘Changed’ on this graphic
No sense of footnoting when it shows up on the next line. Just include your explanation as part of the caption.
-
The breakdown of shows is heavily skewed towards the ‘all at once’ shows. This is mainly due to the dominance of Netflix. Breaking down the release schedules by streaming platform paints a very clear picture (Figure 5).
I think I'd just skip straight to Figure 5, as Figure 4 is mostly not giving anything important because, as mentioned, of the Netflix dominance.
-
The distribution of shows by genre is relatively straightforward and makes sense. Initially, we were surprised by the comparatively low number of comedy shows compared with drama and even docuseries shows. However, we realized that this was due to the fact that the only shows whose data we gathered were shows that were present on a top 10 ranking. And as the majority of critically acclaimed TV shows are dramas in some capacity, this difference in genre counts makes sense based on our data collection methods.
Above figure.
Also, this figure I have a little bit of a harder time understanding exactly how it is going to factor into your research question. So all the more reason to perhaps explain that.
-
Looking at the breakdown of shows, it is clear to see that Netflix has considerably more shows in the dataset than any other streaming platform
It is important to relate your EDA back to your research question. What are the implications of this massive lead Netflix has on your research question? (Because I suspect there very much are implications) Keep a reader thinking about the data in the context that you want to evaluate it.
-
Figure 2
Are you referencing these in the Quarto method? Usually it will link them if you are doing that.
-
Breakdown of shows in dataset based on Streaming Platform
Your caption should also explain what the main thing you wanted a reader to get out of the image was. Here presumably that is the huge lead Netflix has in this area.
-
Looking at the breakdown of shows, it is clear to see that Netflix has considerably more shows in the dataset than any other streaming platform; in fact, it has more than all other platforms combined. This is due to two main reasons. First, Netflix’s weekly top 10 data covers a timeframe of 3.5 years, whereas the other platforms weekly top 10 data covers around 0.75 years. Therefore, we have an additional 2+ years worth of data for specifically Netflix shows than the other streaming platforms, leading to more shows. Second, Netflix is the most popular streaming platform, and as such, both has the funding to make much more original content than its counterparts and dominates the combined streaming ratings on Nielsen. According to DigitalTrends, Netflix has 270 million monthly subscribers, whereas Hulu, Max, Paramount+, and Apple TV+ all have under 100 million subscribers (Nickinson, 2024). That large of a difference in revenue allows Netflix to create more original content than its competitors. Additionally, as we were only gathering data for shows that appeared on a top 10 list, Netflix’s large subscriber base helped them dominate the Nielsen rankings. And, perhaps most obviously, Netflix is the oldest streaming platform, having been launched in 2007 (Barnes, 2019). As it is the oldest streaming service, it has also had the most time to create original content.
Again, I'd move the discussion about the image to before the image actually appears.
-
We use the show_id as our main connecting piece, allowing us to connect our shows table with the various ratings tables, the gtrends table (containing Google Trends data), and the releaseinfo and binned_genres tables. Instead of having unique Primary Keys for each table, we primarily use compound keys as unique identifiers for each of our tables. Only the shows and platforms tables have single value primary keys. Every other table has a compound primary key. The compound keys were chosen to have a unique identifier that we could use to identify each row. For example, the netflixtop10 table’s compound key consists of the show_id, date, and season_title. This is because Netflix’s Top 10 data breaks down a show by season, so a single show could appear on the same week’s top 10 list multiple times. An example of this is Bridgerton. The Netflix Top 10 list for the week of June 6, 2024 contains Bridgerton Season 3, Bridgerton Season 2, and Bridgerton Season 1. As such, a compound key of just the show_id and date would not create a unique compound key. Therefore, season_title was also included to differentiate between the different seasons of a show. The season_number was included in the compound key of the releaseinfo table for similar reasons; as the release schedule for a show can change over the lifetime of the show, we needed to include the season_number to the compound key alongside the show_id to ensure that each row had a unique identifier. Most of the other tables such as gtrends, reaperratings, yearlyratings, and nielsenratings use a compound key of just the date and show_id as they do not differentiate between seasons, and therefore each show would have at maximum one data point per week, making our compound keys unique for these tables. The binned_genres table required including both the show_id and the binned_genre as part of the compound key. This is due to the fact that, as mentioned in the Show Genre section, there are a small number of shows whose genre listed both ‘comedy’ and ‘drama’, and as such, we classified them as both. Therefore, within the binned_genres table, there are a few show_ids that are repeated, so we must additionally include the binned_genre column in our compound key to ensure that each row is unique.
I think I'd move much of this up ABOVE your Figure. It is always better to explain the figure fully in the text before showing it. If someone wants to look ahead at the figure as they read then, that is fine. But it keeps someone from looking at a figure, being confused and taking a bunch of time to figure it out, only to continue reading and have you explain it all.
-
Results
You should basically always introduce a section with some text before an image, other section heading, table or etc appears.
Also, I think I'd change this name to Analysis, or Data Analysis.
-
The way that we structured our database tables is shown in the Entity Relationship Diagram of Figure 1.
Good mention, but then you need to explain what is happening in the figure in the text.
-
This ended up being a pretty considerable effort, taking a few days’ worth of work to complete.
Yeah, this is always tricky and difficult to automate.
-
3NF
write this out at least the first time you mention it
-
Data Ethics
I would try to center a bit more of this around how this data interacts around your research question and possible ethical concerns in that direction. This isn't initially a topic that screams out ethics, so I think you need to think a bit harder about this. Maybe think about what the results of your study might be and if they could affect the streaming landscape in a way that might be unethical? If it turns out that all-at-once is substantially better for generating buzz, could that have unequal ramifications in some way to customers? Is the act of binge-watching entire seasons the healthiest way to consume entertainment content? I'm not 100% sure here, I'm just thinking about possible side-effects that might fall under ethical concerns. But I think you can expand this some, even if just to mention that you have considered some of these things and found there to be no ethical problems.
-
As we were essentially duplicating these shows, we separated the binned genres into a different table in our database so we don’t have duplicate shows in the shows table and can maintain having a single row for each show in our shows table.
This feels like a many-to-many relationship type of dynamic, where you'd frequently use a "middle table" to store the connections, where it would just have two columns: show id and genre id, and if a show had multiple it would just show up on multiple rows (the connecting table often has no primary key)
-
This meant that there were a small, but not insignificant number of shows in our dataset that are in a foreign language.
I'm assuming this means as the only available language? Because many you can get dubbed or with sub-titles? Or does this just mean the original language?
-
respective category
Could one show be in multiple of these bins? Like a dramatic comedy?
EDIT: Oh ok, you address this later.
-
superhero teen drama
thetvdb gives this one Action, Drama, Science Fiction. So maybe those would have been more generalizable at least. Oh, and sub-genre of superhero
-
comedy’, ‘drama’, or ‘docuseries’
Are single quotes just rendering in this odd way? These actually feel like back-tics to me, except that they are going the wrong direction.
-
once again turned to Wikipedia.
Also, on thetvdb, though I don't know how they compare
-
from a third party source
Need to name that source!
-
An argument can be made that Wikipedia’s information could be inaccurate. However, it was the only source that we could find that contained the release information for each individual episode of a show, whereas other sources would only contain when the season was released.
thetvdb has an API and includes release dates of every single episode.
-
yearly ratings
Are these ratings of the streaming platform itself? Or like an average rating of shows shown on the platform? What is being rated here I guess is my question.
-
Netflix Weekly Top 10 Streaming Ratings
I don't love subheadings for blocks of text that only end up being a paragraph long. You might reconsider how you've structured this with the subheadings or see if there is more you could like to add about this and FlixPatrol.
-
show
show name?
-
CSV file from the Google Trends website
might be worth pointing out that Google Trends does not offer an API, so this was the best (and only) approach
-
This is interesting, as the different streaming platforms treat this release type differently. On Apple, the hybrid release is essentially the default release type at this point in time, whereas on Netflix, hybrid release schedules are reserved for the most high-performing and popular shows on the platform, such as Stranger Things and Bridgerton. This likely is the cause for the relative low performance of the hybrid release type.
From your earlier descriptions, it seemed like most of the hybrid release systems did some sort of bulk release at the start before switching to weekly or similar. So it seems especially surprising that they do so much worse so quickly, since the initial episodes seem like they are essentially all-at-once released.
-
last 5 years for each show,
what are the oldest shows that you gathered data on? You haven't made that clear yet.
-
Shows like Succession, The Last of Us, and The White Lotus, all of which aired weekly on HBO at the same time they were released on HBO Max (later Max),
Given the timelines you mentioned above, having a timeframe for this would help me place it within that context
-
and Background
you can just leave this off and call this the Introduction. That is totally fine
-
the combination of release schedule with these other factors influenced their survival probabilities.
Though release schedule and platform are highly correlated in this dataset correct?
-
Kaplan-Meier Curve showing Survival Probabilities of shows based on their release schedule
what should a reader take away from this?
-
This graphic
explain first.
-
survival probabilities of the different release schedules
good, since this kinda felt like the whole objective based on your research question.
-
This graphic
explain first
-
Kaplan-Meier Curve showing Survival Probabilities of shows based on their genre and release schedule
why should a reader care?
-
Surprisingly, docuseries with weekly releases do the best at sustaining popularity. We were surprised by this, as we expected to see a higher popularity of all-at-once docuseries, due in part to their oftentimes low number of episodes and are covering time-sensitive issues. This likely has to do with the fact that there are only 9 total shows that are weekly released docuseries in our dataset, and two of those shows are extremely popular shows: Last Week Tonight and The Grand Tour
Yeah, this felt very surprising to me, since docuseries in general was among the worst performing earlier. And yet here all of its variants seem to perform better. What would cause that?
-
First and foremost, i
Explain first.
-
Figure 13
Are the shaded regions indicating confidence or error? Or just arbitrary? I can't figure it out.
Also, I wonder if this graph would be more readable with panels, similar to the below figure.
-
Figure 14
How much of some of these curves are influenced by small numbers of shows fitting in that bin? Would it be helpful to show counts of the number of shows analyzed in each figure?
-
Kaplan-Meier Curve showing Survival Probabilities of shows based on their streaming platform and release schedule
What should a reader take away from this graphic?
-
8.91% (61/684) had their release schedules changed at some point in the show’s lifetime.
This is of all shows that have multiple seasons? Or were some show's scheduled changed mid-season? You'd only mentioned the season changes so far I think.
-
this difference
Exactly what difference are you talking about here. Presumably between shows that have changed and those that have done, but what is the variable you are comparing?
-
Chi-Squared value was 24.482
how to interpret this?
-
the top two features are whether the show was released on Peacock or Netflix
Oh, so these categories have been encoded as different features now?
-
In order to predict the in_yearly column, we provided XGBoost with the following columns: platform_id (the id of the streaming platform the show is on), miniseries (a binary column that states whether a show was a miniseries or not), foreign_language (either states ‘N’ if the show is in English, or the primary spoken language(s) of the show), binned_genre (the show’s genre that was assigned; either drama, comedy, or docuseries), release_changed (a binary column that states whether a show had its release schedule changed over its lifetime), and weeks_in_nielsen (the cumulative number of weeks that the show appeared in the Nielsen Weekly Top 10 Streaming Show Rankings).
It is interesting that you chose to not include the release schedule in this, given the general target of your research question. Why was that?
-
To ensure that our results are as accurate as possible, we performed a test train split on our data to ensure that we have a test data set to compare our model against. Additionally, we also ran a cross validated score test for the same purpose.
Two sentences isn't much of a paragraph.
-
exhibit the two extremes of number of shows in the dataset.
We are still looking at original content right? So you are saying that Peacock had the least amount of original content and Netflix the most.
-
they are on the Yearly Top 10 Rankings for Peacock.
Wait, maybe I am confused about how these yearly top 10 ratings are working. They are broken down by platform? So you are asking if the show shows up on that particular platform's top 10? Isn't the platform then always going to play a strong role, given the relative amount of non-sharing of content that goes on?
-
Label 0 = Show did not appear in Yearly Top 10; Label 1 = Show did appear in Yearly Top 10*
I'd adjust the label on your plot to make this more clear.
Also, these values come to 173, which is a far cry from the numbers mentioned earlier. Why is this so much smaller?
-
Our model does a wonderful job of predicting shows that will NOT be in a Yearly Top 10 Ranking, but does a much worse job of predicting shows that WILL be in a Yearly Top 10 Ranking.
How balanced is your data though?
-
comparatively small number of shows that appeared in a Yearly Top 10 Ranking in our testing dataset.
Clarify this: you happen to have a small number in your testing dataset? Even over runs with different seeds? Or compared to the total number of shows relatively few make top 10 status? (Which I'm still not sure if it is by platform or not)
-
We looked at multiple contributing factors, including genre and streaming platform.
Ironically you do not meant release schedule here, even after saying that was your research question. So maybe you wanted something like: "In the end, we also ended up looking at extra potentially contributing factors, including genre and streaming platform."
-
popularity
continuing popularity. Not necessarily for its popularity at any one time, because you didn't really measure that correct?
-
how good the show is
at least by some metric.
-
I think you can flesh your conclusions out a bit more. What could be done differently in the future in make this sort of analysis easier, more reliable, faster, etc. What new areas would you like to add. Is critical reception one of those areas? Or is that something that is tough to quantify and thus you want to stay away from it? What should the takeaway be for streaming platforms?
-
Ultimately, we found that weekly releases appear to be ideal for shows with preexisting fanbases, but are less satisfactory for newly released shows without preexisting fanbases.
Oooh, where is the real evidence of this? I don't think this was called out to me at any point during the analysis. Make sure you aren't making conclusions that you have not adequately shown or supported.
-