Market Blended Insight

1
Market Blended Insight

• Modeling propensity to buy with the Semantic Web

Manuel Salvadores, Landong Zuo, SM Hazzaz Imtiaz,
John Darlington, Nicholas Gibbins, Nigel R
Shadbolt, and James Dobree
ISWC 2008
2

• Introduction
• Motivation
• Datasets
• Use cases
• Micro Segmentation
• Value chain
• Conclusions and future work

3
Introduction

• The MBI project focuses its research in marketing
  strategies for the B2B sector.
• The project is extending world class Semantic Web
  research from the EPSRCs Advanced Knowledge
  Technologies IRC
• The project plans to aggregate a broad range of
  business in- formation, providing unparalleled
  insight into UK business activity and develop
  rich semantic search and navigation tools.

4
Introduction

• Real data, real B2B processes to ensure real
  scenarios for the undertaken research

5
Introduction

• Context problem

to overcome the problem that traditional
marketing techniques have broad push without
knowing if the recipient has a propensity to buy
6
Introduction

• Innovation
• To create a source of information based on the
  3.7 million companies that constitute the UK
  economy.
• To create a collection of ontologies that covers
  not just company information but a broad range of
  B2B scenarios too.
• To identify the semantic relations and queries
  required to determine propensity to buy.

7
Motivation

• Micro Segmentation
• to classify or to segment potential customers by
  clustering those with common needs

8
Motivation

• Value chain
• defined as a series of value generating
  activities. Products pass through all activities
  of the chain in order, and at each activity the
  product gains some value

Inbound Logistics
Operations
Outbound Logistics
Marketing And Sales
Service
Porters Value Chain Framework
9
Datasets in the 1st prototype

• A backbone of the UK companies within the London
  boroughs of Lewisham and Camden (83 500 companies
  / 12 million RDF triples)
• Ordnance Survey Address Layer from Lewisham and
  Camden (50 million triples)
• Ordnance PointX dataset with point of interest on
  the mentioned areas.
• Extracted data
• MyCamden website (93k RDF triples)
• Architects Journal (105k RDF triples)
• SIC(92) industrial classification, a hierarchy
  with 6k nodes represented in 62k RDF triples.

10
Micro Segmentation

• SIC(92) standard industrial classification does
  not provide finer enough description of
  companies economic activity.

Total market
11
Micro Segmentation

• Italian or Chinese restaurant ?
• That piece of data is out there.

12
Micro Segmentation
hasSIC92
hasName
/SIC92/2367
Trattoria Luca
../company/1
rdfslabel
Restaurant
Initial company information from the backbone
owlsameAs
rdfssubClass
13
Micro Segmentation
14
Micro Segmentation
15
Micro Segmentation

• 5 014 companies with added information
• 4 406 from PointX
• 608 from MyCamden
• 843 new micro segments
• 777 from PointX
• 66 from MyCamden
• Second prototype will scale this scenario from
  Lewisham and Camden London boroughs to the all
  UK.

16
Value Chain

• Relative to a company there are many
  relationships.

Company - customer
Trade Association - member
Company
Company - supplier
Company Director - person
Shareholder
Relationships might be involved into a Value
Chain process.
17
Value Chain
Value Chain
Company - customer
Trade Association - member
Company
Company - supplier
Company Director - person
Shareholder
18
Value Chain
Value Chain
Supplier (local distributor)
Manufacturer
Many network patterns depending on business sector
Suppliers
19
Value Chain

• Finding relationships in the Building and
  Construction industry.

20
Value Chain
21
Value Chain

• Pre-inference data view

22
Value Chain
23
Value Chain
24
Value Chain

• From Architects Journal (105k RDF triples)
• 4 000 suppliers.
• 600 building and construction projects
• 6 000 products
• From the inferred data (30 038 RDF triples) we
  detected 24 287 relationships between companies.

25
Conclusions

• It is possible to enhance companies data
  portfolio by extracting and thus linking
  information from the Internet.
• Complex B2B processes can be defined by ontology
  modeling and therefore use reasoning to infer new
  concepts.
• Validation with the Consortium has concluded that
  both Segmentation and Value Chain scenarios can
  significantly improve their marketing analysis.
• There is a trade-off between reasoning and query
  performance.

26
Future Work (2nd prototype)
27

• Questions ?