TV3.0 New TV frontiers.

Consumer eXperienceNew TV frontier

2

Present the actual status of the Proof of Concept for IPTV customer, based on ScaleOut technoogy

3

Proof of Concept status• Manage STB behaviour events

– STBOn– STBChangeChannel– VoDEnter– VoDSelectItem– VoDExit– STBOff

• Alarms– Manage STB alarms– Manage Network alarms

• Monitor Channel Statistics– Channel visualizations global and per postal code (number users)– VoD visualizations (number users)

• Monitor STB– STB state– STB real time channel and session visualizations– STB Session Ads printings– STB recommended Items– STB online user with same likes

• Monitor Alarms– Monitor network alarms – Monitor node status and linked STB statuses

• Batch (MapReduce analysis)– Channels Programs share– Ads prints per channel, per brand, per time stamp as user printed– Item recommendations maps– User with similar likes map

• Recommendations engines– User based recommendation engine ( online user similar to you are now watching …)– Item based recommendations (who bought this item also bought …)

4

PoC: Architecture

ConsumereXperience

CDN alarms

STB alarms

STB configuration

TV Guide.xml

Ads scheduling.xml

VoD.xml

5

PoC: Tracking STBs• Real Time management of STB state, user behaviour …• Real Time channel visualization tracking, ads printing visualization tracking• Enter/Exit Video On Demand catalogue

TV Guide

Ads scheduling

Events:• STBOn• STBChangeChannel• VoDEnter• VoDSelect• VoDExit• STBOff

VoD

6

PoC: Store STBs data

• Store in repository all events produced by every STB for batch analysis

ID_290384, STBOn, 12/2/215-10:00ID_290384, STBChannelChange, 12/2/215-10:02…

Events:• STBOn• STBChangeChannel• VoDEnter• VoDSelect• VoDExit• STBOff

7

In-Memory Data Grid intelligently manages IPTV bandwidth allocation for set-top boxes based on updates from CDN on bandwidth usage:• IMDG tracks active set-top

boxes, including viewer’sparameters, events, and box’s CDN node.

• IMDG can apply policies (e.g.,allowed b/w) and notify boxafter viewer event.

• CDN delivers periodic node-specific updates on bandwidthusage and overload condition.

• For each CDN update, IMDGupdates all affected boxes,applies policies, and notifiesbox to adjust b/w if necessary.

• All affected boxes are notified.

PoC: Managing STBs configuration

8

PoC: Store CDN events

• Store in repository all alarms produced by CDN and STB alarms for batch analysis

ID_290384, STBAlarm, Type1 ,Level1,12/2/215-10:00ID_290374, STBAlarm, Type1 ,Level2,12/2/215-10:20…

Alarms:• STBAlarm

CDN alarms

NodeID_84, NodeAlarm, Type1 ,Level1,12/2/215-10:00…

9

Simple GUI with REAL TIME monitoring of individual and aggregated statistics• Individual STB state• STB Ads printing• STB Channel visualization• STBs alarms• Aggregated Channel statistics• CDN Node state• …

PoC: Monitoring

10

PoC: Monitoring

Real Time Programs statistics

11

PoC: Monitoring

Advertising tracking

Program Tracking

12

PoC: MonitoringCDN alarms

management

13

Advance batch analysis with Map/Reduce Big Data techniques• Channel visualization statistics• Program share statistics• Ads printing counting• …

PoC: Batch analysis

14

PoC: Batch analysis• Ej: Program audience share report

ID_290384, STBOn, 12/2/215-10:00ID_290384, STBChanneChange, 12/2/215-10:02…



STB events stored in Log files or

Per STB session Program

visualization analysis

Map

Reduce Per Program reduce







Per Program reduce

Per Program reduce

Programs share report

15

PoC: Batch analysis• Ej: Ads printing visualization




STB events stored in Log files or

Per STB session Ad

printing analysis

Map

Reduce Per Ad reduce

Per STB session Ad

printing analysis

Per STB session Ad

printing analysis

Per STB session Ad

printing analysis

Per Ad reduce

Per Ad reduce

Ads visualization report

16

PoC: Batch analysis• Ej: Ads printing visualization. Number of

impresions (user that have seen the Ad)

17

Discursion of several posibilities for Innovative Services and their possible implementation in the PoC

Some services are not implemented in actual PoC release.User based recomendations and Item based recommendations are partialy released

18

Two user scenarios

TV channel watching Video On Demand watching

19

Consumer eXperience

The sum of experiences at all interaction point between the customer and the TV service during all the duration of a TV session.

20

Consumer eXperience services

• One2One marketing• One2One recommendations• Users based recommendation• Item based recommendation • Online pop-up notifications• Rates&Reviews• Gamification• Social TV

21

Value Added services

• Ads personalization• Ads bidding• Identity Management• Big Data Analitycs

22

One-to-one marketing

• Propose a different user experience for EACH USER, and EACH TIME the user enters.

• Total personalization of the user interface• Personalization of marketing campaigns– Set dozens/hundreds of campaign rules by

marketing– Analyzed on real time for each user

23


ConsumereXperience

Marketing Rules engine

1. If condition offer item2. If … offer … 3. If … offer … 4. If … offer … 5. If … offer … 6. If … offer … 7. If … offer …

All data from each user:- User Profile- Past consumer actions

Marketing Rules

Condition any logical combination of user profile, past consumer experience and online actions

Item any product, film, program, discount … from internal products catalog or from external sources

24


Hello Sara, we have this offering for you

In your next film

25


Hello Sara, we recommend to Buy first

HDSEE HD UPGRADE TO

CAR ISSURANCE CUPON

174095E6T63

26

One-to-one recommendations

• This is a recommendations model based on marketing interests, so is marketing who decides under specific circumstances which item from catalogue to to recommend the user.

• Can be based on items characteristics (ej: type of film, director, ..)

• Can be based on customer past behavior (ej: if the last film watched was a Disney one offer new Disney films)

• or both types … combined or not

27


ConsumereXperience

Recommendations engine


Recommendations relationships

X X

X

X

Relations between items base on catalogue section, item characteristics, actors etc

28


Other similar filmsBrowsing

29


ConsumereXperience

Recommendations engine


Recommendations engine (combined)

Batch Hadoop process.Based on the previous customer events, last films watched and the relations catalogue

X X

X

X X

X

30


FREE with 5600p

Our recommendations for you

31

Collective Intelligence

• Offer not what marketing wants to sell but what the user wants to buy

• No marketing interaction• Only based on users behavior• Most algorithm parts done in batch processing

best with Map/Reduce • Algorithms does know nothing about users

profiles, or products characteristics• Persistence storage could be build NoSQL database

32

Collective Intelligence

(ID_USER1, ID_ITEM4)

USERID ITEMID

245345 3264

262456 45654

262456 4546

262456 456

345343 45

132312 324

132312 234

ConsumereXperience

33

Online users based recommendation

• Use of Collective Intelligence• Is based on what is call neighborhood, which

are the users that have similar “likes” to the one selected user

• Each user has its own neighborhood, as many different neighborhood as users

• Respond to the question“ What are other users like me watching now ?”

34


ConsumereXperience

Recc. EngineStep 2


Batch Hadoop step 1

UserNeighborhood object

Neighborhoodalgorithm…

3,2 6,7 8,2 3,2 4,7 2,3

6,7 4,7 6,8

8,2 2,3 6,8

USERID ITEMID

245345 3264

262456 45654

262456 4546

262456 456

345343 45

132312 324

132312 234

35


• Batch Hadoop Step 1• Batch process to find for each user the most

similar users based on previous buying experience (neighborhood)

• NxN/2 algorithm• The result is for each user a weighted distance

with the rest of users

36


• Real time recommendation engine Step 2• Or one online user, the step 2 gets the

neighborhood (ej: 100) of this user and in real time asks those neighborhood users what are watching now.

• Collects, weights, normalizes and order results to offer a small set of recommended programs

37


Other people (similar to you ) are watching now

38

Item based recommendation

• Use of Collective Intelligence• Creates a neighborhood of items for each item• Respond to the question“ Who bought this item also bought …”

39


ConsumereXperience


Batch Hadoop step 1

ItemNeighbourhood database

Neighborhoodalgorithm

Rec. Engine step 2

3,2 6,7 8,2 4,7 2,3

6,8

USERID ITEMID

245345 3264

262456 45654

262456 4546

262456 456

345343 45

132312 324

132312 234

40


• Hadoop Step 1 process to find based on the user neighborhood the relation of most close items to each other

• NxN/2 algorithm• The algorithm converts the user neighborhood

in a item neighborhood • The result is for each item a weighted distance

with the rest of item

41


• Recommendation Engine Step 2 – when the user selects one item (ej: film) the in memory representation objects passes that info to Recommendation Engine who asks for ej: 4 recommended items for the selected one.

• Is a real time query to batch Map/Reduce prepared information

42


Who viewed this film also viewed

43

Recommendation refinements

• All presented recommendation engines should have refinements, probably the most important is NOT to recommend any product that the user has actually bought in the past.

• So item recommendations engines should be combined with a “filter” engine that for a user and a set of items removes the the items previously bought.

• This could be a recommendation engine inside component or maybe an external component

• This filtering is done in Real Time

44

Recommendation refinements

ConsumereXperience

Any recommendation

engine


Items setTo recommend

Items filter

45

Recommendation refinements II

• Other refinement to algorithms could come from when the commendation is queried (time/date/day of week)

• Up to now same person would receive same recommendation Friday night 9:00, than Saturday morning 10:00, and that may not have sense, so for more relevant recommendations, day of week and time could be consider as a key factor in recommendation algorithm

46


(ID_USER1, ID_ITEM4)

USERID Time/Date ITEMID

245345 10/10:00 3264

262456 10/11:00 45654

262456 10/11:01 4546

262456 10/12:01 456

345343 11/12:01 45

132312 12/18:01 324

132312 12/18:05 234

ConsumereXperience

47


ConsumereXperience

Recc. EngineStep 2


Batch Hadoop step 1

Time frame based UserNeighborhoods

Neighborhoodalgorithm…

3,2 6,7 8,2 3,2 4,7 2,3

6,7 4,7 6,8

8,2 2,3 6,8

USERID Time/Date

ITEMID

245345 10/10:00 3264

262456 10/11:00 45654

262456 10/11:01 4546

262456 10/12:01 456

345343 11/12:01 45

132312 12/18:01 324

132312 12/18:05 234

3,2 6,7 8,2 3,2 4,7 2,3

6,7 4,7 6,8

8,2 2,3 6,8

3,2 6,7 8,2 3,2 4,7 2,3

6,7 4,7 6,8

8,2 2,3 6,8

48


ConsumereXperience


Batch Hadoop step 1

ItemNeighbourhood database

Neighborhoodalgorithm

Rec. Engine step 2

3,2 6,7 8,2 4,7 2,3

6,8

USERID Time/Date

ITEMID

245345 10/10:00 3264

262456 10/11:00 45654

262456 10/11:01 4546

262456 10/12:01 456

345343 11/12:01 45

132312 12/18:01 324

132312 12/18:05 234

3,2 6,7 8,2 4,7 2,3

6,8

3,2 6,7 8,2 4,7 2,3

6,8

49

More recommendation refinements

• More and more sophisticated engines could be implemented based in more sophisticated algorithms introducing– Behavioral aspects (recommendation based on

actual navigation and searches)– User profile analysis (Identity360)– Aggregated Geo Information or other statistical

information– ...

50

Online pop-up notifications

• Consider that one user is browsing the Catalogue or watching a program or a film.

• The system has determined that in your past watching experience you always see a program in a channel and is about to start right now

• The system will send you a notification alert that the program you always see is going to start in that moment and recommend to switch to that channel and not lose the beginning of the program

51


Hi Sara, you always watch and is going to start right now, want to switch ?

52


• Consider that one user is browsing the Catalogue or watching a program or a film.

• The algorithm could determine that in your neighbourhood of similar users, people is switching at this right moment massively to a specific program in other channel

• The system will send you a notification alert that the program other similar users to you are moving to that TV program (maybe MasterChef) that is starting right now

53


Hi Sara, many people is switching to right now, do you want to switch too ?

54

Online pop-up notifications• Consider that one user watching a soccer match. The first

period ends and ads will take 15 minutes. The user starts zapping.

• The algorithm could determine that the users is watching the match and after 15 minutes the second period starts

• If when the second period starts the users is watching other program, the system will send a notification alert that the match is stating second period and ask if watch to switch

• In general is applicable to films, programs etc

55


Hi Sara, the second period of the soccer match is to start, do you want to switch ?

56

Rates&Reviews

• People buys based on other users reviews• In music or films is a key decision point• Independent evaluation of films by other users

helps a more trustable buying decision• Rates could be general or weighted by user

neighbourhood • Reviews could be selected online or offline by

call center independent personnel

57

Rates&Reviews

ConsumereXperience

R&R engine


3 4 1 8

5 6

3

2 3

8

User ratings

Batch Hadoop process

58

Rates&Reviews refinement

• While rating of items could be calculated with the total community of user rating (which is the normal case in all rating models) as an arithmetic average, is more relevant id we consider only the rating of the neighbourhood of the user.

• So this refinement only consider the arithmetic average of the user that has similar likes to the one to be given item rates.

• So 2 different users will see different ratings for same product

59

Rates&Reviews

ConsumereXperience

R&R engine


5 6

3

2 3

8

3,2 6,7 8,2 4,7 2,3

6,8

Recommendation engineneighbourhood

User ratings

3 4 1 8

Batch Hadoop process

Personalized rating

60

Rates&Reviews

61

Gamification• Introduces gaming engagement concepts to consumer

experience• Those may include reputation point, budgets, virtual money,

rankings …• User may receive those benefits from, buying a film, rating a

program, …• Make programs visible based on reputation point or budgets

or virtual money …• Make gamification info social• Gaming info can be include in one to one marketing

campaigns and the opposite…

62

Gamification

ConsumereXperience

Gamification engine


• Points for buy• Points for recomm• Points for rating• Expert point …• Master points…• …• …

Gamification Rules

Marketing Rules engine

63

Gamification

Need 2100

point to open this

film

You have 1800 points

FREE with 5600p

Congrats you can watch this film

Only Grand Master

64

Gamification


Last films opened

Your friends points

4000 points

500 points

2200 points

800 points

3400 points

2200 points

1200 points

1900 points

65

Social TV

• People likes to expose on social networks their likes, actions and share those info with friends

66

Facebook Integration

• So Facebook integration is a first must in social TV.– Connect TV with user Facebook account– Share films seen to friends– Share point and gadgets– Share films reviews– See friends actions (programs or films they have

seen with their opinion)

67


ConsumereXperience

Facebook integration


19/02/15

68


Sara recommends you the film

Sara reached 1800 point

69

Facebook IntegrationYour Facebook friends online are watching :

70

Social TVYour Facebook friends recommendations for you:

71

IMDb Integration

• Other social integrations could also be possible , for example IMDb integration

• IMDb is the world best database of films information

• Maybe al alternative/complement for the presented Rates module could be used instead the IMDb rates & reviews

72

IMDb Integration

73

Twitter Integration

• People likes to know what other people thinks on a specific matter or program

• So Twitter integration is a must in social TV.– Connect TV with user Twitter– A TV program is linked to a Twitter tag – See Twits on real time on TV linked to a program

74

Twitter Integration

ConsumereXperience

Twitter integration

75

Twitter Integration

75

76

Social Chat TV

• Real time communication with other users• In a similar way to Whatsapp or thought

Facebooks groups a user could establish online communication channel with contacts, Facebook friends or whatever groups or contacts platforms defines as available

77

Social Chat TV

ConsumereXperience


Peer-To-Peer communication

Hi friend, now starting Master Chef in channel 6

78

Social TV

78



79

All services working together in REAL TIME

80

All services running together

Gamification engine

Personalization engine

Recommendations engines

Rates&Reviews engine

Social engines

ConsumereXperience

Online communications

81

All services togetherYou have 1800 points

Hello Sara, we have this offering for you

Our recommendations for you today

Other people is watching now

82

All services together


Your ratings

Want to know more

Last programs you have seen

Specials offers for you

Our recommendations for you today

Other people is watching now Your friends are watching

83



SEE REVIEWS

YOU WILL RECIVE 250 POINTS

Who bought this film also bought

HD

TO SEE HD UPGRADE TO

See more films from

Our recommendations for you

Your second Disney film

Want to see what is happening in

84


You have 2050 pointsYou have added 250 POINTS

Here is your cupon: 2gn3487634g48

Want to rate this film? Get 200 point

50% off Your next Disney film

Want to recommend this film to your friends? Get 100 point

85


85

Hi Sara, the second period of the soccer match is to start, do you want to switch ?

86


A set of Value Added Services for the TV operators that increase the benefits and open a new set of strategic posibilities

87


• Ads personalization• Ads bidding• Identity Management• Batch Big Data Analitycs

88

Ads personalization• Google AdWorks model has disrupted the way publishing ads• In a similar way, TV advertising model could be based on:

– Pay per print– Personalization

• That means that EACH person watches a different Ad at the same time, based on different factors like its profile, Identity information, past behaviour, current scenario, date/time etc…

• So we need three main elements:– A Real Time powerfull personalization Engine that should be able to select

the best fit Ad for each user each time– A Real Time tracking engine to count ads printing per user– A behaviour engine to track the user behaviour in Ads to feedback system

89

Ads personalization

89

TV Guide

Ads scheduling

Ads catalogAds tracking

Adds personalization

engine

The Ads personalization engine selects the BEST matching Ad for every user every time

90

Ads bidding

• The possibility to track and visualize in Real Time the number of user watching a TV channel allows to create and offer a Real Time bidding platform for Real Time ads insertion.

91

Identity management• Identity Management means Record, Connect and Analyze

ALL available information from users to create a complete user profile called Identity360– Contract information– User behavioral information

• What, when are user preferences and interests (programs, films…)• Watching, zapping and TV usage behavior• Navigation and search behavior• Buying behavior

– Social information (Facebook, twitter…)– Service contacts and call center iterations– …

92

Big Data analytics

Big Data analysis opens a broad range of new possibilities• Customer analytics– Customer better understanding, marketing One2One, better

Ads insertion, UpSelling, CrossSelling …• Strategic statistics– Aggregated customers behavior, GeoAnalysis, Behavioral

Analysis, Better Segmentation Analysis• Infrastructure– Better CDN provisioning and better customer quality

experience and support

93

THANKS for your time

www.qualityobjects.com

César [email protected]

http://www.qualityobjects.com/

TV3.0 New TV frontiers.

Technology