Diving into YouTube Analytics

Deepnote presented a live stream featuring a showcase of how to use Google's YouTube API. We had a few Deepnote coders and a Deepnote user Allan who helped crack the API and start getting data. Below is a project inspired from that stream!

In this notebook we are going to dive in and check out the popular YouTube channel Good Mythical Morning.

Want the data without making the API calls? GMM data csv can be downloaded here. Github

Imports

In this section we are going to import our libraries.

from googleapiclient.discovery import build import os import pandas as pd import seaborn as sns import matplotlib.pyplot as plt

Setup

First you will need to create an API key by going to API signup. Then I created an enviroment variable in Deepnote by selecting integrations on the side menu. Call it whatever you would like and paste in your key. Now we can grab our API key out of our Deepnote integration and setup our youtube variable.

API_KEY = os.environ["API"] CHANNEL_ID = "UC4PooiX37Pld1T8J5SYT-SQ" youtube = build('youtube', 'v3', developerKey=API_KEY)

API Functions

In this section we setup some API functions to help us gather data. We create a function to gather channel stats and video stats. To do this in a effecient way and save API calls we do not use the search function. We pull in GMM's upload playlist ( containing all of their videos ) and use that to build a video list. Then grab the data from each video.

# Function to get the channels stats # It will also contain the upload playlist ID we can use to grab videos. def get_channel_stats(youtube, channel_id): request = youtube.channels().list( part="snippet,contentDetails,statistics", id=channel_id ) response = request.execute() return response['items']

# This will get us a list of videos from a playlist. # Note a page of results has a max value of 50 so we will # need to loop over our results with a pageToken def get_video_list(youtube, upload_id): video_list = [] request = youtube.playlistItems().list( part="snippet,contentDetails", playlistId=upload_id, maxResults=50 ) next_page = True while next_page: response = request.execute() data = response['items'] for video in data: video_id = video['contentDetails']['videoId'] if video_id not in video_list: video_list.append(video_id) # Do we have more pages? if 'nextPageToken' in response.keys(): next_page = True request = youtube.playlistItems().list( part="snippet,contentDetails", playlistId=upload_id, pageToken=response['nextPageToken'], maxResults=50 ) else: next_page = False return video_list

# Once we have our video list we can pass it to this function to get details. # Again we have a max of 50 at a time so we will use a for loop to break up our list. def get_video_details(youtube, video_list): stats_list=[] # Can only get 50 videos at a time. for i in range(0, len(video_list), 50): request= youtube.videos().list( part="snippet,contentDetails,statistics", id=video_list[i:i+50] ) data = request.execute() for video in data['items']: title=video['snippet']['title'] published=video['snippet']['publishedAt'] description=video['snippet']['description'] tag_count= len(video['snippet']['tags']) view_count=video['statistics'].get('viewCount',0) like_count=video['statistics'].get('likeCount',0) dislike_count=video['statistics'].get('dislikeCount',0) comment_count=video['statistics'].get('commentCount',0) stats_dict=dict(title=title, description=description, published=published, tag_count=tag_count, view_count=view_count, like_count=like_count, dislike_count=dislike_count, comment_count=comment_count) stats_list.append(stats_dict) return stats_list

Create our Channel Stats

channel_stats = get_channel_stats(youtube, CHANNEL_ID)

Find our Upload Playlist ( will contain all video uploads )

upload_id = channel_stats[0]['contentDetails']['relatedPlaylists']['uploads'] upload_id

video_list = get_video_list(youtube, upload_id)

Get our Video Details

Finally we will get all of our videos details returned in a dictionary.

video_data = get_video_details(youtube, video_list)

Creating Visualizations

In this section we will covert our data to pandas dataframe and start visualizing. I have also created a CSV of data in the cell below if you don't want to do the API calls yourself.

df=pd.DataFrame(video_data) df['title_length'] = df['title'].str.len() df["view_count"] = pd.to_numeric(df["view_count"]) df["like_count"] = pd.to_numeric(df["like_count"]) df["dislike_count"] = pd.to_numeric(df["dislike_count"]) df["comment_count"] = pd.to_numeric(df["comment_count"]) # reaction used later add up likes + dislikes + comments df["reactions"] = df["like_count"] + df["dislike_count"] + df["comment_count"] + df["comment_count"] df.to_csv("GMM-Data.csv") df.head()

Plotting Number of Tags Per Video

# Histogram of tag count. chart = sns.distplot( a=df["tag_count"], hist=True, kde=False, rug=False ) chart.set(xlabel='Number of Tags Used', ylabel='Occurrences')

Title Length Used

chart = sns.distplot( a=df["title_length"], hist=True, kde=False, rug=False ) chart.set(xlabel='Title Length', ylabel='Occurrences')

Most Viewed Videos

df_highest_views = df.nlargest(10, 'view_count') df_highest_views['title'] = df_highest_views['title'].str[:40] df_highest_views['view_count_millions'] = df_highest_views['view_count'] / 1000000 df_highest_views

sns.set(rc={'figure.figsize':(20,10)}) plot = sns.barplot(x="view_count_millions", y="title", data=df_highest_views, palette="bright") plot.set(xlabel='Millions of Views', ylabel='') plot.set_title('Most Viewed Videos') plt.xlim(20, 32)

Most Reacted to Videos (Likes + Thumbs Down + Comments)

# Most Interations ( Like + Dislikes + Comments) df_highest_reactions = df.nlargest(10, 'reactions') df_highest_reactions['title'] = df_highest_reactions['title'].str[:40] df_highest_reactions['reactions'] = df_highest_reactions['reactions'] / 1000 df_highest_reactions

sns.set(rc={'figure.figsize':(20,10)}) plot = sns.barplot(y="title", x="reactions", data=df_highest_reactions, palette="bright") plot.set(xlabel='Thousands of Reactions', ylabel='') plt.xticks(rotation=45) plot.set_title('Most Video Reactions') plt.xlim(300, 700)

Most Thumbs Downed Videos

df_highest_disliked = df.nlargest(10, 'dislike_count') df_highest_disliked['title'] = df_highest_disliked['title'].str[:40] df_highest_disliked['dislike_count'] = df_highest_disliked['dislike_count'] / 1000 df_highest_disliked

sns.set(rc={'figure.figsize':(20,10)}) plot = sns.barplot(y="title", x="dislike_count", data=df_highest_disliked, palette="bright") plot.set(xlabel='Thousands of Thumbs Down', ylabel='') plt.xticks(rotation=45) plot.set_title('Most Thumbed Down Videos') plt.xlim(10, 55)

Diving into YouTube Analytics

Imports

Setup

API Functions

Create our Channel Stats

Find our Upload Playlist ( will contain all video uploads )

Get our Video Details

Creating Visualizations

Plotting Number of Tags Per Video

Title Length Used

Most Viewed Videos

Most Reacted to Videos (Likes + Thumbs Down + Comments)

Most Thumbs Downed Videos

Recommended on Deepnote

Stock market analysis

10 ways to create NumPy arrays

Wide Residual Networks

Diving into YouTube Analytics

Imports

Setup

API Functions

Create our Channel Stats

Find our Upload Playlist ( will contain all video uploads )

Get our Video List

Get our Video Details

Creating Visualizations

Plotting Number of Tags Per Video

Title Length Used

Most Viewed Videos

Most Reacted to Videos (Likes + Thumbs Down + Comments)

Most Thumbs Downed Videos

Recommended on Deepnote

Stock market analysis

10 ways to create NumPy arrays

Wide Residual Networks