Read Me

Objective

To create a dash app that uses networkX and plotly to generate a graph of featured channels for a select list of channels as our point of origin, and to conduct statstical analysis on significant nodes in the network and on the connectivity of the graph.

Purpose

The purpose of this project is visualize graphs of YouTube channels by crawling through featured channels listed on profile pages. This graph can then be used to analyze each channel's relative significance in that network.

In this approach, we start off with a list of channels as our point of origin, extract the list of channels featured on each profile page, and then repeat the process n times.

This will produce a directional graph, which means each edge has directionality involved. For instance channel A might point to channel B, but channel B might not point to channel A. If both directions exist, than they are both counted as two separate edges.

Note: Due to the crawling nature of this project, the analysis is relative to this subset of channels in the YouTube universe, therefore the statistics will change for different sized networks containing different points of origin.

---

Background Context

Each channel has an option to feature other youtube channels on their profile page. This appears on their profile page as a tab. As in this example for Google's YouTube channel.

Many channels do not feature other channels on their profile pages. Such as with Google's Webmasters YouTube channel

Some channels exclusively feature channels within their business network. For example BBC.

Although a lot of channels feature channels with similar contexts, channels of friends, and/or collaborating channels. Such as Corridor Crew, The Slow Mo Guys, and Smarter Every Day.

---

Collecting Data

The data come from Google's Youtube Data Api v3. I created a GCP project, generated an API key, and used two API endpoints; specifically youtube.search.list and youtube.channels.list.

The API-key lives in a config.py file, which was excluded from this repo for security purposes. If you would like to replicate this project: create a GCP project, generate an API key, and write a config.py file to reference that API-key. The quota limit is 10,000 units per day.

• youtube.search.list costs 100 units per request
• youtube.channels.list costs 1 unit per request

The Data - using jsons and structuring data

The youtube.channels.list endpoint returns a json with a variety of parts about the channelId requested.

['kind', 'etag', 'id', 'snippet', 'contentDetails', 'statistics', 'topicDetails', 'status', 'brandingSettings', 'contentOwnerDetails']

Graphs

We used NetworkX to graph the network of channels. Specifically using networkX's DiGraph object. We then used plotly to visualize the graph by individually plotting each node and edge in two separate traces as Scatter Plots.

The graphs are visualized using a force-directed graph drawing utilizing the Kamada-Kawai algorithm for determining the position of each node, instead of the standard spring layout (Fruchterman-Ringold algorithm).

Graph of Corridor Digital's 3-distance featured network

Graph of Corridor Digital's strongly connected components in their 3-distance featured network

Statstics

Degree, In Degree, Out Degree, Betweenness Centrality, In Degree Centrality, Page Rank

Top 8 channels by Betweenness Centrality;

	title	page_rank	out_degree	degree	in_degree	b_centrality	id-centrality	subCount	viewCount
1	Corridor	0.0337084	11	29	18	0.197719	0.0608108	8,080,000	1,469,507,306
2	Corridor Crew	0.0399649	12	20	8	0.112594	0.027027	4,110,000	713,277,250
10	devinsupertramp	0.0127246	10	25	15	0.10919	0.0506757	5,920,000	1,438,824,815
22	Nukazooka	0.02434	8	19	11	0.0826987	0.0371622	2,260,000	670,559,796
23	Mike Diva	0.0112591	7	15	8	0.0796148	0.027027	650,000	142,893,349
9	Film Riot	0.0155565	13	23	10	0.0792312	0.0337838	1,670,000	191,825,386
15	SoKrispyMedia	0.0152009	10	18	8	0.0741729	0.027027	1,150,000	319,054,747
0	Corridor Cast	0.0219452	10	14	4	0.0515002	0.0135135	111,000	5,011,173

Top 8 channels by Indegree Centrality;

	title	page_rank	out_degree	degree	in_degree	b_centrality	id-centrality	subCount	viewCount
1	Corridor	0.0337084	11	29	18	0.197719	0.0608108	8080000	1469507306
10	devinsupertramp	0.0127246	10	25	15	0.10919	0.0506757	5920000	1438824815
92	TomSka	0.0167781	15	28	13	0.0492833	0.0439189	6440000	1657474725
44	LetsPlay	3.69413e-17	1	12	11	1.71782e-05	0.0371622	3850000	2437731719
22	Nukazooka	0.02434	8	19	11	0.0826987	0.0371622	2260000	670559796
20	RocketJump	1.44605e-10	3	13	10	0.0258455	0.0337838	9000000	1904596631
9	Film Riot	0.0155565	13	23	10	0.0792312	0.0337838	1670000	191825386
14	Rooster Teeth	0.000467877	12	21	9	0.018583	0.0304054	9470000	6081509776

Top 8 channels by Page Rank;

	title	page_rank	out_degree	degree	in_degree	b_centrality	id-centrality	subCount	viewCount
35	TechLinked	0.0509535	6	12	6	5.72607e-06	0.0202703	1260000	183056413
50	LMG Clips	0.0509535	6	12	6	5.72607e-06	0.0202703	178000	20307801
65	Carpool Critics	0.0509535	6	11	5	0	0.0168919	42400	425975
76	ShortCircuit	0.0509535	6	11	5	0	0.0168919	948000	56517459
18	Linus Tech Tips	0.0509535	6	13	7	0.0079077	0.0236486	11500000	3924848893
89	Techquickie	0.0509535	6	12	6	5.72607e-06	0.0202703	3530000	527112792
2	Corridor Crew	0.0399649	12	20	8	0.112594	0.027027	4110000	713277250
105	Channel Super Fun	0.035747	4	10	6	0	0.0202703	740000	97736777

Dash Appplication Use

This tool can be used to assess any community of channels and their subsequent n-distance network.

---

Conclusion

Next Steps

There are a lot of questions that this analysis does not answer. There are also a lot of features that I can add to this tool.

Next Questions

• Analyzing the universe of YouTube Videos.
• Expanding the size of nodes.
• Adding a weight to page rank and other metrics based on subscriber count and total views.

Next Features

• I would like to allow a user to select a variety of differing channels and watch how they link between separate networks.
• Take advantage of selection feature in Plotly.