A Big Data Analysis on YouTube Trending Videos in the United States¶

Mohammad A. Kazemivarnamkhasti, Rami Shehadeh, Wenxuan Wan

Introduction¶

It’s hard imaging modern society operating YouTube, the video sharing platform goliath. YouTube is the second most popular website on Earth with 1.8 billion user visits every month, falling short to only Google, its parent website. With the massive amount of traffic YouTube receives, it’s hard for a single video to get any attention. However YouTube superstars like Mr.Beast, T-Series, and PewDiePie seem to effortlessly soar into the trending videos with ease. This begs the question, what makes a video go viral? What kinds of video would have the highest chances of reaching the most people? What kinds of videos would stir up the most activity in your comments? Let’s first take a step back. Why do we even want to succeed on YouTube? Many content creators pour hundreds of hours into a single video just to hope that people watch it and stay engaged. Creative drive aside, having a successful YouTube channel would mean a stream of ad revenue and sponsorships, successful YouTube channels can even appear on national television or other traditional media as guests. However, this is all contingent on your videos doing well. In this analysis we seek to answer the traits do success videos have in common? What are their audience engagements like, and what topics do they cover?

Data Collection¶

For this anayslsis we collected our YouTube trending dataset from Kaggle. More specifically this link. This takes you to a dataset containing trending YouTube data from 2020 to 2022. This data is comprised of multiple .csv files. Since we’re mainly interested in the US data, we’ll be using US_youtube_trending_data.csv. Pandas and Numpy are used to do basic data science and data manipulation.

In [1]:
import pandas as pd
import numpy as np
import math
from matplotlib import pyplot as plt
import seaborn as sns
from scipy import stats

import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)

Cleaning the Data¶

Before any analysis can be done we first must clean the data so that its more workable and presentable.

Lets first see what our data looks like

In [2]:
df = pd.read_csv("US_youtube_trending_data.csv")
df.head()
Out[2]:
video_id title publishedAt channelId channelTitle categoryId trending_date tags view_count likes dislikes comment_count thumbnail_link comments_disabled ratings_disabled description
0 3C66w5Z0ixs I ASKED HER TO BE MY GIRLFRIEND... 2020-08-11T19:20:14Z UCvtRTOMP2TqYqu51xNrqAzg Brawadis 22 2020-08-12T00:00:00Z brawadis|prank|basketball|skits|ghost|funny vi... 1514614 156908 5855 35313 https://i.ytimg.com/vi/3C66w5Z0ixs/default.jpg False False SUBSCRIBE to BRAWADIS ▶ http://bit.ly/Subscrib...
1 M9Pmf9AB4Mo Apex Legends | Stories from the Outlands – “Th... 2020-08-11T17:00:10Z UC0ZV6M2THA81QT9hrVWJG3A Apex Legends 20 2020-08-12T00:00:00Z Apex Legends|Apex Legends characters|new Apex ... 2381688 146739 2794 16549 https://i.ytimg.com/vi/M9Pmf9AB4Mo/default.jpg False False While running her own modding shop, Ramya Pare...
2 J78aPJ3VyNs I left youtube for a month and THIS is what ha... 2020-08-11T16:34:06Z UCYzPXprvl5Y-Sf0g4vX-m6g jacksepticeye 24 2020-08-12T00:00:00Z jacksepticeye|funny|funny meme|memes|jacksepti... 2038853 353787 2628 40221 https://i.ytimg.com/vi/J78aPJ3VyNs/default.jpg False False I left youtube for a month and this is what ha...
3 kXLn3HkpjaA XXL 2020 Freshman Class Revealed - Official An... 2020-08-11T16:38:55Z UCbg_UMjlHJg_19SZckaKajg XXL 10 2020-08-12T00:00:00Z xxl freshman|xxl freshmen|2020 xxl freshman|20... 496771 23251 1856 7647 https://i.ytimg.com/vi/kXLn3HkpjaA/default.jpg False False Subscribe to XXL → http://bit.ly/subscribe-xxl...
4 VIUo6yapDbc Ultimate DIY Home Movie Theater for The LaBran... 2020-08-11T15:10:05Z UCDVPcEbVLQgLZX0Rt6jo34A Mr. Kate 26 2020-08-12T00:00:00Z The LaBrant Family|DIY|Interior Design|Makeove... 1123889 45802 964 2196 https://i.ytimg.com/vi/VIUo6yapDbc/default.jpg False False Transforming The LaBrant Family's empty white ...

We see that there are alot of information present and many are abbreviated to numbers. Lets first translte the categoryId into english tags. This is done by referring to the YouTube video category ID list found here: https://mixedanalytics.com/youtube-video-category-id-list/

In [3]:
cat_id = {
    1 :"Film & Animation",
    2 : "Autos & Vehicles",
    10: "Music",
    15: "Pets & Animals",
    17: "Sports",
    18: "Short Movies",
    19: "Travel & Events",
    20: "Gaming",
    21: "Videoblogging",
    22: "People & Blogs",
    23: "Comedy",
    24: "Entertainment",
    25: "News & Politics",
    26: "Howto & Style",
    27: "Education",
    28: "Science & Technology",
    29: "Nonprofits & Activism",
    30: "Movies",
    31: "Anime/Animation",
    32: "Action/Adventure",
    33: "Classics",
    34: "Comedy",
    35: "Documentary",
    36: "Drama",
    37: "Family",
    38: "Foreign",
    39: "Horror",
    40: "Sci-Fi/Fantasy",
    41: "Thriller",
    42: "Shorts",
    43: "Shows",
    44: "Trailers"
}
In [4]:
df_c = df['categoryId'].tolist()
cat_names = [None] * len(df_c)
for i in range(len(df_c)):
    cat_names[i] = cat_id[df_c[i]]
df.drop(columns = ['categoryId'], inplace= True)
df['category'] = cat_names

There's still alot of information that are not import to us such as the video id, thumbnail url, or description. Lets remove them and the dataframe.

In [5]:
df.drop(columns = ['video_id', 'channelId', 'description','thumbnail_link'], inplace = True)
df
Out[5]:
title publishedAt channelTitle trending_date tags view_count likes dislikes comment_count comments_disabled ratings_disabled category
0 I ASKED HER TO BE MY GIRLFRIEND... 2020-08-11T19:20:14Z Brawadis 2020-08-12T00:00:00Z brawadis|prank|basketball|skits|ghost|funny vi... 1514614 156908 5855 35313 False False People & Blogs
1 Apex Legends | Stories from the Outlands – “Th... 2020-08-11T17:00:10Z Apex Legends 2020-08-12T00:00:00Z Apex Legends|Apex Legends characters|new Apex ... 2381688 146739 2794 16549 False False Gaming
2 I left youtube for a month and THIS is what ha... 2020-08-11T16:34:06Z jacksepticeye 2020-08-12T00:00:00Z jacksepticeye|funny|funny meme|memes|jacksepti... 2038853 353787 2628 40221 False False Entertainment
3 XXL 2020 Freshman Class Revealed - Official An... 2020-08-11T16:38:55Z XXL 2020-08-12T00:00:00Z xxl freshman|xxl freshmen|2020 xxl freshman|20... 496771 23251 1856 7647 False False Music
4 Ultimate DIY Home Movie Theater for The LaBran... 2020-08-11T15:10:05Z Mr. Kate 2020-08-12T00:00:00Z The LaBrant Family|DIY|Interior Design|Makeove... 1123889 45802 964 2196 False False Howto & Style
... ... ... ... ... ... ... ... ... ... ... ... ...
172185 Minecraft but I have 1,000,000 XP 2022-12-11T15:00:03Z Craftee 2022-12-16T00:00:00Z minecraft|minecraft but|new minecraft|craftee|... 1190457 23172 0 1497 False False Gaming
172186 Paddy Pimblett Octagon Interview | UFC 282 2022-12-11T05:33:54Z UFC - Ultimate Fighting Championship 2022-12-16T00:00:00Z ufc|mma|ultimate fighting championship|UFC 282... 1446095 11069 0 7736 False False Sports
172187 REVEALING Our BABY'S NAME! 😱 2022-12-10T17:46:12Z The Royalty Family 2022-12-16T00:00:00Z The Royalty Family|Royalty Family|Andrea Espad... 3103034 76518 0 18022 False False People & Blogs
172188 Star Wars Jedi: Survivor - Official Reveal Tra... 2022-12-09T01:52:37Z EA Star Wars 2022-12-16T00:00:00Z star wars|star wars jedi survivor|jedi survivo... 5286961 174278 0 12748 False False Gaming
172189 Chris Brown - It's Giving Christmas 2022-12-09T17:00:26Z ChrisBrownVEVO 2022-12-16T00:00:00Z Chris brown its giving Christmas|chris brown t... 889375 60867 0 2918 False False Music

172190 rows × 12 columns

Views Vs Like Ratio¶

Intuitively we think that popular and trending videos have high views counts and like like ratio. Lets see how related thet actually are. We first need to check if there are any missing values in the columns we are planning to look at.

In [6]:
nan_likes = df['likes'].isnull().values.any()
nan_dislikes =  df['dislikes'].isnull().values.any()
nan_views =  df['view_count'].isnull().values.any()

print(nan_likes)
print(nan_dislikes)
print(nan_views)

False
False
False

Perfect, there aren't any null or missing values. However this dataset isn't going to be THAT nice to us. YouTube changed their dislike so that any video uploaded after November 10 2021 has a zero for their dislike value. However We believe that looking at the remaining data (where dislikes were included) will still provide valuable insights. Data with zero dislikes will be dropped

In [7]:
### Lets find out more about the viewcounts and like to dislike videos

likes_to_dislike_ratio = [None]* len(df)
likes_count = df['likes']
dislikes_count = df['dislikes']
i = 0
for l, d in zip(likes_count, dislikes_count):
    if l+d == 0 or d == 0:
        likes_to_dislike_ratio[i] = np.nan
        i += 1
        continue
    likes_to_dislike_ratio[i] = (l/(l+d)) * 100
    i += 1
df['likes_ratio'] = likes_to_dislike_ratio
df = df.dropna()
df.head()
Out[7]:
title publishedAt channelTitle trending_date tags view_count likes dislikes comment_count comments_disabled ratings_disabled category likes_ratio
0 I ASKED HER TO BE MY GIRLFRIEND... 2020-08-11T19:20:14Z Brawadis 2020-08-12T00:00:00Z brawadis|prank|basketball|skits|ghost|funny vi... 1514614 156908 5855 35313 False False People & Blogs 96.402745
1 Apex Legends | Stories from the Outlands – “Th... 2020-08-11T17:00:10Z Apex Legends 2020-08-12T00:00:00Z Apex Legends|Apex Legends characters|new Apex ... 2381688 146739 2794 16549 False False Gaming 98.131516
2 I left youtube for a month and THIS is what ha... 2020-08-11T16:34:06Z jacksepticeye 2020-08-12T00:00:00Z jacksepticeye|funny|funny meme|memes|jacksepti... 2038853 353787 2628 40221 False False Entertainment 99.262657
3 XXL 2020 Freshman Class Revealed - Official An... 2020-08-11T16:38:55Z XXL 2020-08-12T00:00:00Z xxl freshman|xxl freshmen|2020 xxl freshman|20... 496771 23251 1856 7647 False False Music 92.607639
4 Ultimate DIY Home Movie Theater for The LaBran... 2020-08-11T15:10:05Z Mr. Kate 2020-08-12T00:00:00Z The LaBrant Family|DIY|Interior Design|Makeove... 1123889 45802 964 2196 False False Howto & Style 97.938673

Lets verify that our code above worked

In [8]:
count_zero = (df['dislikes'] == 0).sum()
count_ratio = (df['likes_ratio'] == 0).sum()
print(count_zero)
print(count_ratio)

0
0

Lets plot a graph showing our like ratio as a percent and the number of views a video has

In [9]:
plt.scatter(df['likes_ratio'], df['view_count'])
plt.xlabel(r"Like Ratio (%)")
plt.ylabel("Views")
plt.title("Like Ratio vs Views")
plt.show()

This is a bit messy; Let's try another approach. This one using seaborn and using its regression.

In [10]:
sns.regplot(x='view_count',y='likes',data=df)
plt.show()

With this we see a more linear relation between views and likes. This is expected but we can look more into by taking the log(base 10) of each axis

In [11]:
views = df['view_count']
likes = df['likes']

views_log= []
likes_log = []
for i,v in enumerate(zip(views, likes)):
    if v[0] == 0 or v[1] == 0:
        continue
    views_log.append(math.log(v[0],10))
    likes_log.append(math.log(v[1],10))
In [12]:
log_df = pd.DataFrame()
log_df['likes_log'] = likes_log
log_df['views_log'] = views_log
sns.regplot(x='views_log',y='likes_log',data=log_df)
plt.show()

This is more interesting, but there's too many points. Let's plot the first 1% of the points. This may seem like a tiny amount but keep in mind we have almost 100,000 rows of data, 1% of this is still a thousand rows.

In [13]:
log_df = pd.DataFrame()
log_df['likes_log'] = likes_log[0:len(likes_log)//100]
log_df['views_log'] = views_log[0:len(likes_log)//100]
sns.regplot(x='views_log',y='likes_log',data=log_df)
plt.show()

What percent of trending videos have a high like ratio¶

Next lets see if most of the videos that made it into trending has a high like ratio

In [14]:
plt.hist(df['likes_ratio'])
plt.xlabel(r"Like Ratio (%)")
plt.ylabel("Number of Videos")
plt.title("What percent of popular video has a high like ratio")
plt.show()

We see that the overwhelming majority of the views in trending are in fact very well received! This means that sensational or controversal content (these content would typically have lower like ratios) would have a harder time breaking into trending videos.

What types of videos are the most popular?¶

If you wanted to make a video to break into trending, what category of video should you make?

In [15]:
temp = {}
for c in cat_id.values():
    temp[c] = 0

for data in df['category']:
    temp[data] += 1

categories = {}
for c in temp:
    if temp[c] > 100:
        categories[c] = temp[c]
fig = plt.figure(figsize =(25, 15))    
plt.bar(categories.keys(), categories.values())
plt.xlabel(r"Categories")
plt.ylabel("Number of Videos")
#plt.title("")
plt.show()

We see that the 3 most popular categories are Music, Sports, and Entertainment

Trending categories per month¶

We want to see if theres a relationship between the trending date(months) and the chances of a video making it to trending

Preprocessing¶

In this part of the project, we'll be looking into the most popular trending categories of YouTube videos per month; however, from looking at the head of our dataframe in the cell above, you can see that the format of the 'trending_date' column is odd and needs some cleaning. Once this column is modified, it can be converted to standard datetime format for further analysis.

In [16]:
# taking a substring of the first ten characters of each trending data
# effectively removing the extra characters at the end
for i, row in df.iterrows():
  df.at[i, "trending_date"] = row["trending_date"][0:10]

# converting trending_date col to datetime format
df['trending_date'] = pd.to_datetime(df['trending_date'], format='%Y.%m.%d')
df.head()

C:\Users\Steve\AppData\Local\Temp\ipykernel_13156\156249899.py:7: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  df['trending_date'] = pd.to_datetime(df['trending_date'], format='%Y.%m.%d')
Out[16]:
title publishedAt channelTitle trending_date tags view_count likes dislikes comment_count comments_disabled ratings_disabled category likes_ratio
0 I ASKED HER TO BE MY GIRLFRIEND... 2020-08-11T19:20:14Z Brawadis 2020-08-12 brawadis|prank|basketball|skits|ghost|funny vi... 1514614 156908 5855 35313 False False People & Blogs 96.402745
1 Apex Legends | Stories from the Outlands – “Th... 2020-08-11T17:00:10Z Apex Legends 2020-08-12 Apex Legends|Apex Legends characters|new Apex ... 2381688 146739 2794 16549 False False Gaming 98.131516
2 I left youtube for a month and THIS is what ha... 2020-08-11T16:34:06Z jacksepticeye 2020-08-12 jacksepticeye|funny|funny meme|memes|jacksepti... 2038853 353787 2628 40221 False False Entertainment 99.262657
3 XXL 2020 Freshman Class Revealed - Official An... 2020-08-11T16:38:55Z XXL 2020-08-12 xxl freshman|xxl freshmen|2020 xxl freshman|20... 496771 23251 1856 7647 False False Music 92.607639
4 Ultimate DIY Home Movie Theater for The LaBran... 2020-08-11T15:10:05Z Mr. Kate 2020-08-12 The LaBrant Family|DIY|Interior Design|Makeove... 1123889 45802 964 2196 False False Howto & Style 97.938673

Now we want to start analyzing by looking into which video categories are trending in this dataset.

In [17]:
byCat = df.groupby(by=['category']).sum().reset_index()
display(byCat)
category view_count likes dislikes comment_count comments_disabled ratings_disabled likes_ratio
0 Autos & Vehicles 1914785329 88772670 1408858 7713271 0 0 1.707036e+05
1 Comedy 10511614324 718422436 11174623 39961401 5 0 5.400268e+05
2 Education 3337783703 209396163 4040488 15941268 21 0 2.124092e+05
3 Entertainment 65440917179 3346084989 62549431 223442458 356 0 1.913180e+06
4 Film & Animation 8162675980 374676085 6686643 30691115 16 0 3.655510e+05
5 Gaming 38245601154 2112545311 38395646 169568027 98 0 1.657954e+06
6 Howto & Style 4627675942 222634390 7266989 18647080 47 0 3.095635e+05
7 Music 83492867653 5706361753 106002834 661578759 28 0 1.750666e+06
8 News & Politics 6084677056 80036755 9979392 22407985 314 0 3.068019e+05
9 Nonprofits & Activism 131399890 11096562 112968 397858 0 0 8.624558e+03
10 People & Blogs 18465758586 1108635493 31225806 74720498 217 0 8.246492e+05
11 Pets & Animals 631401256 30929729 395656 2200316 0 0 4.994657e+04
12 Science & Technology 9356091709 406897672 9343119 25099157 265 0 3.378653e+05
13 Sports 21981874704 551473512 15843036 46554551 44 0 1.016712e+06
14 Travel & Events 408323428 21577650 453556 1289756 0 0 4.004674e+04

We see that from the set of 44 categories, our dataset contains only videos from 15 categories, so we'll only focus on these for this portion of data exploration. Next, we can divide up the rows of our dataframe by each of these 15 categories and go on from there.

In [18]:
# getting each category's data as a separate df
grouped = df.groupby(df.category)
auto = grouped.get_group("Autos & Vehicles")
com = grouped.get_group("Comedy")
edu = grouped.get_group("Education")
entertain = grouped.get_group("Entertainment")
film = grouped.get_group("Film & Animation")
gaming = grouped.get_group("Gaming")
howto = grouped.get_group("Howto & Style")
music = grouped.get_group("Music")
news = grouped.get_group("News & Politics")
activism = grouped.get_group("Nonprofits & Activism")
blogs = grouped.get_group("People & Blogs")
animals = grouped.get_group("Pets & Animals")
scitech = grouped.get_group("Science & Technology")
sports = grouped.get_group("Sports")
travel = grouped.get_group("Travel & Events")
auto.head()
Out[18]:
title publishedAt channelTitle trending_date tags view_count likes dislikes comment_count comments_disabled ratings_disabled category likes_ratio
105 $3000 vs. $500 Suspension Upgrade 2020-08-09T16:00:12Z Donut Media 2020-08-12 donut media|hi low|james pumphrey|nolan skyes|... 1354257 62586 560 3888 False False Autos & Vehicles 99.113166
127 Road Rage – Why Motorcycle Riders Smash Windows 2020-08-09T15:00:01Z FortNine 2020-08-12 motorcycle road rage compilation|motorcycle cr... 342870 30254 580 3124 False False Autos & Vehicles 98.118960
162 First Shreds! Testing Our Duramax Swapped Cama... 2020-08-07T15:00:32Z Hoonigan Project Cars 2020-08-12 Knuckle Busters|Knuckle Busters 2|Edelbrock|Ed... 466200 15363 316 1272 False False Autos & Vehicles 97.984565
187 Ford Sticks It To Jeep By Taking The New Bronc... 2020-08-06T21:04:51Z TFLoffroad 2020-08-12 TFLoffroad|TFL|2021 ford bronco|2021|ford bron... 511098 6700 761 2191 False False Autos & Vehicles 89.800295
215 We Buy a Donk Project. Goal: 1000HP! Part 1 of 69 2020-08-12T16:00:08Z Hoonigan 2020-08-13 Donk|Box|Bubble|Donk magazine|RIDES Magazine|C... 112081 6115 101 2021 False False Autos & Vehicles 98.375161

In order to properly identify the most popular categories per month, we can use the split-up groups of rows (one for each category) and compute the total monthly view count. The total views per month of the 'Autos & Vehicals' category is shown as an example. (Likes are added along with view count solely so that the resultant objects will be dataframes, which is what we want, and not series.)

In [19]:
categories = [auto, com, edu, entertain, film, gaming, howto, music, news, activism, blogs, animals, scitech, sports, travel]
names = ['auto', 'com', 'edu', 'entertain', 'film', 'gaming', 'howto', 'music', 'news', 'activism', 'blogs', 'animals', 'scitech', 'sports', 'travel']
sums = {}
for i in range(len(categories)):
    sums[names[i]] = categories[i].groupby(categories[i].trending_date.dt.month)['view_count','likes'].sum()
sums['auto']
Out[19]:
view_count likes
trending_date
1 113389596 4060354
2 187585344 3962879
3 71786108 2725930
4 97722045 2306466
5 82213650 5587252
6 297415169 19399228
7 83316454 4023365
8 178427336 7179923
9 203266951 15289605
10 203936100 7045607
11 191377634 8666578
12 204348942 8525483

In this step, we make a dictionary mapping each month to the the total view count per appropriate category, for every category. Each key is a number corresponding to a month and each key is a set of tuple: (view_count, category). If the appropriate key already exits in the map, then the tuple/value is appended to that key; otherwise, the key-value pair is initialized in the dictionary.

In [20]:
months = {}
for key in sums:
    # key is the name of the category
        
    for index, row in sums[key].iterrows():
        if index in months:
            months[index].append((row['view_count'], key))
        else:
            months[index] = [(row['view_count'], key)]

# sorting months accordingly (1-12)
months = {key: val for key, val in sorted(months.items(), key = lambda ele: ele[0])}
#months

Now that we have a dictionary mapping each month to its views by category, we can perform a descending sort on these to find the category with most views from each month, which has been the objective of this portion.

In [21]:
# top categories (by views) from each month of the year
for i in range(1,13):
  months[i].sort(reverse=True)
  print('month', i,':',months[i][0][1])

month 1 : music
month 2 : music
month 3 : music
month 4 : entertain
month 5 : music
month 6 : entertain
month 7 : entertain
month 8 : music
month 9 : music
month 10 : music
month 11 : music
month 12 : music
In [22]:
data = [ [1, months[1][0][0]], [3, months[3][0][0]], [5, months[5][0][0]],
         [8, months[8][0][0]], [9, months[9][0][0]], 
         [10, months[10][0][0]], [11, months[11][0][0]]]
cols = ['Jan', 'Mar','May','Aug','Sep','Oct','Nov']

df2 = pd.DataFrame(data, cols)

ax = df2.plot(kind='bar', title='Most Popular Categories per Month - Music', color=['g', 'b', 'r', 'c'])
plt.legend(['Entertainment','Music'])
ax.set_xlabel("Month of Year")
ax.set_ylabel("Views (hundreds of millions)")
plt.show()
In [23]:
data = [[2, months[2][0][0]], [4, months[3][0][0]], 
        [6, months[6][0][0]], [7, months[7][0][0]], [12, months[12][0][0]]]
cols = ['Feb', 'Apr','Jun','Jul','Dec']

df3 = pd.DataFrame(data, cols)

ax = df3.plot(kind='bar', title='Most Popular Categories per Month - Entertainment', color=['b', 'g', 'r', 'c'])
plt.legend(['Music','Entertainment'])
ax.set_xlabel("Month of Year")
ax.set_ylabel("Views (hundreds of millions)")
plt.show()

We can see that only two of the 15 categories within our dataset had the most views in a month. The 'Entertainment' category had the most views in the months of February, April, June, July, and December, while the 'Music' category had more views than any other category in the other months of the year.

Controversial Trending Videos¶

Now, we are going to be looling at controversial videos. We are going to define controversial videos as having:

  • like ratio below 50%
  • comment count in the upper 40th percentile

The reason for choosing this criteria is it allows us to limit the data set to find trending videos that caused controversey, and at the same time not limiting it too much so that we have enough data to work with.

In [24]:
#controversial videos: high comments, low like ratio


df2 = pd.DataFrame()

 # for each row, if the like ratio is below 50, and comments in upper 40th percentile
for index, row in df.iterrows():
  if row['likes_ratio'] < 50 and stats.percentileofscore(df['comment_count'], row['comment_count']) >= 60: 
    df2 = df2.append(row) 

#remove duplicate titles (same video trending at a different time)
df2 = df2.drop_duplicates(subset=['title'])

We have limited our data set to only contain controversial trending videos based on our criteria. However, we can see that we are not left with that many results. What this mainly tells us is that videos of controverisal nature do not commonly go on the YouTube trending page.

Let's examine this data further. What are the main categories of trending videos that are controvesial?

In [25]:
### Bar plot of the number of controversial videos per category
sns.set(rc={'figure.figsize':(13,10)})
plot = sns.countplot(x="category", data=df2)
plot.set(title='Controversial Trending YouTube Videos per Category')

plt.show()

From the plot above, we can clearly see that the main category of controversial videos that go on the trending page is "News & Politics".

This tells us that videos regarding politics or world news are more likely to generate a lot of discussion in the comment section, in addition to having a rather low percentage of people liking the video.

This should not be surprising. However, we can also look at the category "News & Politics" in general to see how many of those videos in our dataset are controversial (based on our criteria).

In [26]:
### split "News & Politics" into 'controversial' and 'non_controversial'

df['is_controversial'] = np.nan
df['is_controversial']=df['is_controversial'].astype('boolean')

for index, row in df.iterrows():
  if row['category'] == "News & Politics":
    if row['likes_ratio'] < 50 and stats.percentileofscore(df['comment_count'], row['comment_count']) >= 60: 
      df.at[index, 'is_controversial'] = True
      
    else:
      df.at[index, 'is_controversial'] = False
      
#plot the two 

df = df.dropna(subset=['is_controversial'])
sns.set(rc={'figure.figsize':(4,10)})
ax = sns.countplot(x="is_controversial", data=df)
ax.set(title = 'Controversial Videos in The News & Politics Category',xlabel='Controversial', ylabel='Count')

plt.show()

C:\Users\Steve\AppData\Local\Temp\ipykernel_13156\3310834883.py:3: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  df['is_controversial'] = np.nan
C:\Users\Steve\AppData\Local\Temp\ipykernel_13156\3310834883.py:4: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  df['is_controversial']=df['is_controversial'].astype('boolean')

From the bar plot above, we can see that very few trending videos with the category "News & Politics" are controversial. This tells us that despite most of the controversial videos being related to News & Politics, this is a very low percentage when compared to the entire category.

Conclusion¶

So from our tutorial we see that to make a trending video we want to first keep the audience engaged. This can take the form of a high ratio of likes and a high ratio of likes to view. Videos in specific categories tend to trend more often (e.g., Music, Entertainment) so it’s more worthwhile to focus on content in those fields. It’s important to note that trending videos are rarely controversial as seen by the very low number of controversial videos in the “News & Politics” category; Note that this category was the one with the highest number of controversial videos so it’s safe extrapolate to all categories.

Hope this helps you make a banger YouTube video channel and I wish to see you on the trend page.

In [ ]: