mininao / BI-SQL-Homework

Homework for a College Business Intelligence course

Rendu de projet BI - Youtube

Maxime Ghermaoui - Christophe Loeur - Vincent Salahkar - Maxence Aïci

Modèle de données

• Le modèle de données ci-dessus est disponible ici : DataModel.png
• Le modèle original a été créé avec le logiciel Visual Paradigm, le fichier source est DataModel.vpp
• Le script permettant de créer les tables et leur contraintes est DataModel.sql

Données de test

• Les données de test on été générées avec Mockaroo, Permettant d'en créer autant que souhaité
• Nous avons tenté d'obtenir un maximum de cohérence dans ces données ainsi qu'un volume important pour pouvoir bien tester nos indicateurs.
• Les scripts sql créant chacune des tables sont dans le dossier FakeData
• Un dump complet de la base de données est également disponible dans dump.sql.gz

Voici la quantité de données ajoutées :

Table	Lignes
Channel	500
Comment	1000
Interaction	10000
Like	1000
Publicite	40
Pubview	3000
Subscription	2000
User	1000
Video	100
Videoview	5000

Indicateurs

1. Indicateurs Simples

• Vidéos les plus vues
• Pubs qui poussent à cliquer
• Chaînes avec le plus d'abonnés

2. Indicateurs Moyens

• Vidéos les plus populaires
• Vidéos qui rapportent le plus d'argent
• Genre de vidéos qui poussent à les regarder en entier

3. Indicateurs Difficiles

• Vidéos les plus populaires avec pondération temporelle
• Chaines qui rapportent le plus d'argent
• Chaines qui poussent à rester sur youtube

4. Indicateurs DataMart

• Revenu Journalier
• Durée Journalière regardée totale

Indicateurs Simples

Vidéos les plus vues

SELECT Video.video_id, Video.name, COUNT(Videoview.videoview_id) AS Vues
FROM Videoview
INNER JOIN Interaction
INNER JOIN Video
WHERE Videoview.interaction_id = Interaction.interaction_id
AND Interaction.video_id = Video.video_id
AND Interaction.start_date BETWEEN date_sub(now(),INTERVAL 1 WEEK) AND now()
GROUP BY Video.video_id
ORDER BY Vues DESC, Video.name ASC
LIMIT 12;

Pubs qui poussent à cliquer

SELECT S1.publicite_id, S1.name, S2.Clics/S1.Vues AS Pourcentage
FROM (SELECT Publicite.name,Publicite.publicite_id, COUNT(Pubview.pubview_id) AS Vues
FROM Pubview
INNER JOIN Publicite
WHERE Pubview.publicite_id = Publicite.publicite_id
GROUP BY Pubview.publicite_id) AS S1
INNER JOIN (SELECT Publicite.name,Publicite.publicite_id, SUM(Pubview.clicked) AS Clics
FROM Pubview
INNER JOIN Publicite
WHERE Pubview.publicite_id = Publicite.publicite_id
GROUP BY Pubview.publicite_id) AS S2
WHERE S1.publicite_id = S2.publicite_id
GROUP BY S1.publicite_id
ORDER BY Pourcentage DESC
LIMIT 12;

Chaînes avec le plus d'abonnés

SELECT Channel.channel_id, Channel.channel_name, COUNT(Subscription.subscription_id) AS Subs
FROM Subscription
INNER JOIN Channel
WHERE Subscription.channel_id = Channel.channel_id AND Subscription.end_date IS NULL
GROUP BY Subscription.channel_id
ORDER BY Subs DESC, Channel.channel_name ASC
LIMIT 10;

Indicateurs Moyens

Vidéos les plus populaires

SELECT S1.video_id, S1.name, ((3*S1.Comments + S2.Likes)*100/S3.`Views` + S3.`Views`) AS Popularite, S1.Comments, S2.Likes, S3.`Views`
FROM(SELECT Video.video_id, Video.name, COUNT(Comment.comment_id) AS Comments
FROM Comment
INNER JOIN Interaction
INNER JOIN Video
WHERE Comment.interaction_id = Interaction.interaction_id AND Interaction.video_id = Video.video_id AND Comment.end_date IS NULL
GROUP BY Interaction.video_id) AS S1

INNER JOIN(SELECT Video.video_id, Video.name, COUNT(`Like`.`like_id`) AS `Likes`
FROM `Like`
INNER JOIN Interaction
INNER JOIN Video
WHERE `Like`.interaction_id = Interaction.interaction_id AND Interaction.video_id = Video.video_id AND `Like`.end_date IS NULL
GROUP BY Interaction.video_id) AS S2

INNER JOIN (SELECT Video.video_id, Video.name, COUNT(Videoview.videoview_id) AS `Views`
FROM Videoview
INNER JOIN Interaction
INNER JOIN Video
WHERE Videoview.interaction_id = Interaction.interaction_id AND Interaction.video_id = Video.video_id
GROUP BY Interaction.video_id) AS S3

WHERE S1.video_id = S2.video_id AND S2.video_id = S3.video_id
ORDER BY Popularite DESC
LIMIT 12;

Vidéos qui rapportent le plus d'argent

Select Video.video_id,Video.name,
COUNT(S1.interaction_id) AS PubViews,
SUM(S1.clicked) AS PubClics,
AVG(S1.CPC) AS AverageCPC,
AVG(S1.CPV) AS AverageCPV,
SUM(Revenue) AS TotalRevenue
FROM
(Select Pubview.clicked, Publicite.CPV, Publicite.CPC, Pubview.interaction_id, (Publicite.CPV + Pubview.clicked * Publicite.CPC) AS Revenue
FROM Pubview
INNER JOIN Publicite
WHERE Pubview.publicite_id = Publicite.publicite_id) AS S1

INNER JOIN Interaction ON S1.interaction_id = Interaction.interaction_id
INNER JOIN Video ON Interaction.video_id = Video.video_id
GROUP BY Interaction.video_id
ORDER BY TotalRevenue DESC, Video.name ASC
LIMIT 10;

Genre de vidéos qui poussent à les regarder en entier

SELECT Video.genre,
AVG(Videoview.duration / Video.duration) AS AverageViewRatio,
AVG(Video.duration) AS AverageVideoDuration,
AVG(Videoview.duration) AS AverageViewDuration
FROM Videoview
INNER JOIN Interaction ON Videoview.interaction_id = Interaction.interaction_id
INNER JOIN Video ON Interaction.video_id = Video.video_id
GROUP BY Video.genre
ORDER BY AverageViewRatio DESC, Video.Genre ASC
LIMIT 10;

Indicateurs Difficiles

Vidéos les plus populaires avec pondération temporelle

SELECT S1.video_id, S1.name, ((3*S1.FreshComments + S2.FreshLikes)*100/S3.FreshViews + S3.FreshViews) AS FreshPopularite, S1.FreshComments, S2.FreshLikes, S3.FreshViews
FROM(SELECT Video.video_id, Video.name, COUNT(Comment.comment_id) AS Comments,
SUM(GREATEST(
  TIMESTAMPDIFF(SECOND,Interaction.start_date,date_sub(now(),INTERVAL 1 MONTH))
/ TIMESTAMPDIFF(SECOND,now()                 ,date_sub(now(),INTERVAL 1 MONTH))
,0)) AS FreshComments
FROM Comment
INNER JOIN Interaction
INNER JOIN Video
WHERE Comment.interaction_id = Interaction.interaction_id AND Interaction.video_id = Video.video_id AND Comment.end_date IS NULL
GROUP BY Interaction.video_id) AS S1

INNER JOIN(SELECT Video.video_id, Video.name, COUNT(`Like`.like_id) AS Likes,
SUM(GREATEST(
  TIMESTAMPDIFF(SECOND,Interaction.start_date,date_sub(now(),INTERVAL 1 MONTH))
/ TIMESTAMPDIFF(SECOND,now()                 ,date_sub(now(),INTERVAL 1 MONTH))
,0)) AS FreshLikes
FROM `Like`
INNER JOIN Interaction
INNER JOIN Video
WHERE `Like`.interaction_id = Interaction.interaction_id AND Interaction.video_id = Video.video_id AND `Like`.end_date IS NULL
GROUP BY Interaction.video_id) AS S2

INNER JOIN (SELECT Video.video_id, Video.name, COUNT(Videoview.videoview_id) AS `Views`,
SUM(GREATEST(
  TIMESTAMPDIFF(SECOND,Interaction.start_date,date_sub(now(),INTERVAL 1 MONTH))
/ TIMESTAMPDIFF(SECOND,now()                 ,date_sub(now(),INTERVAL 1 MONTH))
,0)) AS FreshViews
FROM Videoview
INNER JOIN Interaction
INNER JOIN Video
WHERE Videoview.interaction_id = Interaction.interaction_id AND Interaction.video_id = Video.video_id
GROUP BY Interaction.video_id) AS S3

WHERE S1.video_id = S2.video_id AND S2.video_id = S3.video_id
ORDER BY FreshPopularite DESC
LIMIT 12;

Chaines qui rapportent le plus d'argent

Select Channel.channel_id, Channel.channel_name,
COUNT(S1.interaction_id) AS PubViews,
SUM(S1.sponsored) AS SponsoredPubViews,
SUM(S1.clicked) AS PubClics,
AVG(S1.CPC) AS AverageCPC,
AVG(S1.CPV) AS AverageCPV,
COUNT(DISTINCT Video.video_id) AS Videos,
SUM(Revenue) AS TotalRevenue
FROM
(Select Pubview.clicked,Publicite.sponsored, Publicite.CPV, Publicite.CPC, Pubview.interaction_id, (Publicite.CPV + Pubview.clicked * Publicite.CPC) AS Revenue
FROM Pubview
INNER JOIN Publicite
WHERE Pubview.publicite_id = Publicite.publicite_id) AS S1

INNER JOIN Interaction ON S1.interaction_id = Interaction.interaction_id
INNER JOIN Video ON Interaction.video_id = Video.video_id
INNER JOIN Channel ON Video.channel_id = Channel.channel_id
GROUP BY Channel.channel_id
ORDER BY TotalRevenue DESC, Video.name ASC
LIMIT 10;

Chaines qui poussent à rester sur youtube

SELECT Channel.channel_id, Channel.channel_name,
ViewCounter.Views,
COUNT(Videoview.videoview_id) AS CreatedViews,
GREATEST(1-(COUNT(Videoview.videoview_id)/ViewCounter.Views),0) AS BounceRate
FROM Videoview
INNER JOIN Video ON Videoview.previous_video_id = Video.video_id
INNER JOIN Channel ON Video.channel_id = Channel.channel_id

INNER JOIN
(SELECT Channel.channel_name, Channel.channel_id, COUNT(Videoview.videoview_id) AS `Views`
FROM Videoview
INNER JOIN Interaction ON Videoview.interaction_id = Interaction.interaction_id
INNER JOIN Video ON Interaction.video_id = Video.video_id
INNER JOIN Channel ON Video.channel_id = Channel.channel_id
GROUP BY Video.channel_id)
AS ViewCounter ON ViewCounter.channel_id = Channel.channel_id

GROUP BY Channel.channel_id
ORDER BY BounceRate ASC, Channel.channel_name ASC
LIMIT 10;

Indicateurs DataMart

Nous avons intégré au datamart deux indicateurs supplémentaires : le revenu publicitaire gagné chaque jour et la durée passée par tous les utilisateurs à regarder des vidéos chaque jour (en secondes).

Ces deux indicateurs devront être représentés à l'aide d'une courbe présentant le temps en abscisses et la quantité en ordonnées, comme celle-ci :

Revenu Journalier

Select date(Interaction.start_date) AS revenue_date,
SUM(Publicite.CPV + Pubview.clicked * Publicite.CPC) AS revenue_amount
FROM Pubview
INNER JOIN Publicite ON Pubview.publicite_id = Publicite.publicite_id
INNER JOIN Interaction ON Pubview.interaction_id = Interaction.interaction_id
GROUP BY date(Interaction.start_date)

Durée Journalière regardée totale

Select date(Interaction.start_date) AS duration_date,
SUM(Videoview.duration) AS duration_amount
FROM Videoview
INNER JOIN Interaction ON Videoview.interaction_id = Interaction.interaction_id
GROUP BY date(Interaction.start_date)