Perv a ds-rr13

PervADsSemantic Pervasive Advertising

Lorenzo Carrara1, Giorgio Orsi2, and Letizia Tanca3

(1) Cubica s.r.l(2) IFoC, Oxford Martin School, University of Oxford

(3) DEIB, Politecnico di Milano

Advertising

A form of communication that creates awareness of an offer about a product or a service.

Advertising

advertiserconsumer

A form of communication that creates awareness of an offer about a product or a service.

Advertising

advertiserconsumer

offer

ads

A form of communication that creates awareness of an offer about a product or a service.

Advertising: key elements

Targeting

too broad/too frequent: ads perceived as background noise.

too focused: you might miss someone.

Advertising: key elements

Targeting

too broad/too frequent: ads perceived as background noise.

too focused: you might miss someone.

Cost

TV/radio: effective, easy to plan, but expensive.

billboarding: less effective, harder to plan, not sensibly cheaper.

Internet: cheap (for now), effective (measurable), but invasive.

Pervasive Advertising

Electronic advertising that targets consumers during their everyday activities.

Pervasive Advertising

Electronic advertising that targets consumers during their everyday activities.

Targeted and context aware

profile-based crafting of ads.

activity-based targeting.

Pervasive Advertising

Electronic advertising that targets consumers during their everyday activities.

Pervasive

mobile devices: tablets, smartphones, …, Google glasses

smart billboards: do not move but that’s pervasive too!

Targeted and context aware

profile-based crafting of ads.

activity-based targeting.

Pervasive (and Targeted) Advertising: Issues

Opacity

effective [1,2] (>50% thanTV/radio) but debated data.

in the hands of few companies (Yahoo, Google, Facebook).

third-parties might bias offers (e.g., Expedia / Booking.com).

[1] Howard Beales. The Value of Behavioral Targeting. NAI.[2] M. Sala, K. Partridge, L. Jacobson, and J. Begole, “An exploration into activity-informed physical advertising using pest,” in Proc. of Pervasive, 2007, pp. 73–90.

Pervasive (and Targeted) Advertising: Issues

Privacy

someone has to know about you and what you do.

consumers both tracked and profiled (activities, demography, behaviour).

[1] Howard Beales. The Value of Behavioral Targeting. NAI.[2] M. Sala, K. Partridge, L. Jacobson, and J. Begole, “An exploration into activity-informed physical advertising using pest,” in Proc. of Pervasive, 2007, pp. 73–90.

Opacity

effective [1,2] (>50% thanTV/radio) but debated data.

in the hands of few companies (Yahoo, Google, Facebook).

third-parties might bias offers (e.g., Expedia / Booking.com).

Direct communication between businesses and consumers.

PervADs

Direct communication between businesses and consumers.

Monitor effectiveness ads campaign

PervADs

Direct communication between businesses and consumers.

Monitor effectiveness ads campaign.

Private and local exploitation of richer user data.

PervADs

PervADs: A typical scenario

PervADs: Structure

OfferDescription

ContextSpecification

Human ReadableAdvert

Context and data modelling [1]

CDO (Context Dimension Ontology): context of ads and consumers.

data about the offer and the product (Schema.org, Good-Relations).

[1] C. Bolchini, C. Curino, G. Orsi, E. Quintarelli, R. Rossato, F. A. Schreiber, L. Tanca: And what can context do for data? Commun. ACM 52(11): 136-140 (2009)

PervADs

Context matching and reasoning [2]

matching context of offers and consumers context containment.

requires inference.

[1] C. Bolchini, C. Curino, G. Orsi, E. Quintarelli, R. Rossato, F. A. Schreiber, L. Tanca: And what can context do for data? Commun. ACM 52(11): 136-140 (2009)[2] G. Orsi, L. Tanca: Context Modelling and Context-Aware Querying - (Can Datalog Be of Help?). Datalog 2.0. 2010: 225-244

PervADs

Context and data modelling [1]

CDO (Context Dimension Ontology): context of ads and consumers.

data about the offer and the product (Schema.org, Good-Relations).

Context Model: Example

Context Model: Example

In summary

schema set of FO constraints (DL-Lite)

instance set of assignments

context containment fact inference

Context Inference: Example

Context Matching

Checking only containment is open to cheating.

Context Matching

Checking only containment is open to cheating.

Penalize contexts that are too broad or too specific.

Context Matching

Penalize contexts that are too broad or too specific.

per-dimension similarity

userinstance

pervadsinstance

Checking only containment is open to cheating.

,

Context Matching

Penalize contexts that are too broad or too specific.

per-dimension similarity

userinstance

pervadsinstance

and

Checking only containment is open to cheating.

,

Context Matching

Penalize contexts that are too broad or too specific.

per-dimension similarity

aggregate (e.g., avg) over all dimensions.

userinstance

pervadsinstance

and

Checking only containment is open to cheating.

,

Dimension Similarity: Example

Matching performance vs dimensions

Performance vs signal strength

PervADs GUI: Queries

PervADs GUI: Matching

PervADs core (https://code.google.com/p/pervads/)

Android client.

Server application (OpenWRT routers).

Get and develop PervADs-like stuff

Mobile ontology management (https://code.google.com/p/androjena/)

AndroJena / μJena.

ARQoid.

LucenOid.

TDBoid.

Get and develop PervADs-like stuff

Mobile ontology management (https://code.google.com/p/androjena/)

AndroJena / μJena.

ARQoid.

LucenOid.

TDBoid.

Apache license

PervADs core (https://code.google.com/p/pervads/)

Android client.

Server application (OpenWRT routers).

Thank you!

http://pervasiveadvertising.org/More on Pervasive Advertising:

http://www.cs.ox.ac.uk/files/4735/RR-11-07.pdfMore on PervADs: