“Big Data & Artificial Intelligence” ----How to Achieve …cyberc.org/Content/pdf/Keynotes...

Preview:

Citation preview

“Big Data & Artificial Intelligence”

----How to Achieve Accurate Sales

Chongqing University of Posts and Telecommunications, Chongqing, China

xugx@cqupt.edu.cn

1/30

Prof. Guangxia Xu

Outline

1. Background

2. How to Achieve Accurate Sales

3. Applications in Other Industries

4. Future outlook

2/30

Outline

1. Background

2. How to Achieve Accurate Sales

3. Applications in Other Industries

4. Future outlook

3/30

Sales to brand strategy

Internet technology and sales integration

“AI & Big Data “and sales Fusion

1.Background

4/30

1.Background

5/30

Significance and

Function

Update Sales Strategy In Time

Reduce Enterprise Sales Cost

Higher Performance Price Ratio

Enhancing Customer Satisfaction

6

Data Processing Takes A Long Time

Error In Analysis Result

Miss the Best Time of Sales Data

Quality Is Too Low

Data Filtering Is Difficult

Data Processing Error

Data Value Is Not Represented

1.Background

The Plight of Accurate

Sales

6/30

显示

7/30

Bank Insurance Online Shopping

AI

1.Background

8/30

Data Collection AI

Analysis Result

Accurate Sales

……Data Preprocessing

1.Background

9/30

1.Background

he

User Data Filtering

01

User Data Analysis

02 User Feature

Extraction

03

Match the Product

04

User Match

05

Marketing Evaluation

Effect

06

Marketing Closed Loop

Process

Outline

1. Background

2. How to Achieve Accurate Sales

3. Applications in Other Industries

4.Future outlook

10/30

2. How to Achieve Accurate Sales

Mail Recommendation

SMS Recommendation

Social Software Recommendation

Recommendation System

User Recommendation

UserManagement

Accurate Sales

Insurance Industry

Financial Industry

FourismIndustry

Industry Users

Insurance Industry

Financial Industry

FourismIndustry

Other Industry

Industry Customers

Sales Demand

Big Data Platform

Sales Users Group

Own Channel Sales

Business

Market Report

Ability Output

Ability Output

Accurate Sales Platform

Data Cleaning

User portrait extraction technology

pre-trained deep neural network

11/30

2. How to Achieve Accurate Sales

Potential User Value MiningMarketing

Model building

Feature Variable Selection

Basic Attributes

Financial Credit Hobby APP

PreferenceLocation

PreferenceUser

Portrait

Application Layer

Data Collection

Information Exchange

Terminal Information

Financial Credit

APP Application

Loss of Customer Retention

Value Customer Conversion

Personalized Recommendation

Stock User Interaction

Combined Marketing

Electricity Consumption

Business Layer

Data Layer

Correlation Analysis

Identify Smart

Variables

Building A Mining

Model

Output Target Group

12/30

Step1: Data cleansing

13/30

2. How to Achieve Accurate Sales

Data cleansing

14/30

Data cleansing

Target Data Set

Graph Data

V1

V2 V3 V2

V4

V3

V5 V4 V5 V5 V4

Graph Data Processing Storage

15/30

16/30

2. How to Achieve Accurate Sales

Step2: Personas

Personas

17/30

User Demographics

Text Data

Network Data

Cleaning and FilteringCrawling

Step1:Data Preprocessing

……

PersonasStep2:Features Extraction

Features One

Features Two

Features Three

Features ...

Personas

18/30

High-level Features

Low-level Features

……

Multi-source data

Personas A

Personas B

Personas C

Personas D

Personas

Personas

User Initial Data

KeywordVector

Latent Dirichlet Allocation

DataProcessing

Data Cleansing

PMICharacteristic

Distribution

User Feature Database

19/30

Personas

User Feature

Database

Feature Sequence

Feature Sampling

Feature Selection

User FeatureGroup

Residual CHaracteristics

PMI

User Portrait

20/30

21/30

2. How to Achieve Accurate Sales

Step3: Recommendation model

22/30

Recommendation model

{W1,b1}

Input x

{W2,b2}

h1 h2

Hidden layers

Output y

23/30

...

...

...

...

...

...

...

...

...

...

...

...

...

Autoencoder 1 Autoencoder 2 Autoencoder 3

UnsupervisedPre-training

SupervisedFine-tuning

Recommendation model

24/30

Recommendation model

g DNN

X1,X2…XN(random)

Y1,Y2…YN

Personas

(w,b)

25/30

Recommendation model

Wage earner

SAE

Stratum

Outline

1. Background

2. How to Achieve Accurate Sales

3. Applications in Other Industries

4.Future outlook

26/30

3.Applications in Other Industries

Intelligent Diagnosis

Drug Selection

Equipment Purchase

Personnel Employed

Doctors Distribution

“Big Data + Artificial Intelligence”

Target UsersMedical Industry

27/30

3.Applications in Other Industries

Equity Market

Insurance Industry

VC Industry

Financial Industry “Big Data + Artificial Intelligence”

Target Users

28/30

Outline

1. Background

2. How to Achieve Accurate Sales

3. Applications in Other Industries

4.Future outlook

29/30

Data Mining

Database

Statistics

Machine Learning

How to solve?

Timeliness

Inefficient

Sample Deviation

…Application

Single

Useless Data

4.Future outlook

1. In the Personas construction, the method of Machine Learning is introduced to adjust data parameters. 2. Preprocessing the data to avoid the curse of dimensionality. 3. Integrated use of cross domain data to break data dependencies.

30/30

Thank You Q & A