You are on page 1of 29

Multi-Agent Framework for

Aspect Level Sentiment


Analysis in CRM Systems
Sales of drones
E.g. Thank you for the demo of the Phantom F21
drone you did the other day. I loved the product.
However I do have some concerns related to the HD
Camera resolution

The overall sentiment of the email is positive


The aspect level sentiment of HD Camera is
negative.
Recommendation
Overall sentiment is positive
Aspect-Level sentiment of HD Camera is negative

Recommendation

Offer free upgrade to a better/higher resolution HD


Camera.
Purpose
The purpose is to describe the design and
development of a multi-agent recommendation
system that uses aspect level sentiment analysis to
convert prospects into customers in CRM Systems.
CRM Customer Relationship Management

Customer Customer Customer Customer


Identification Attraction Retention Development
CRM
Operational CRM

Analytical CRM

Collaborative CRM
CRM Terminology
From Lead to Prospect

Lead Qualification Prospect


Conversation Rate

CR=NS/NP*100

CR- Conversation Rate


NS Number of Total Sales
NP Number of Total Prospects
Sales Pipeline
Initial Communication

First Interview

Qualification/ Opportunity Analysis

Solution Presentation/Demo

Customer Evaluation

Negotiation
Commitment to
Buy
Sale
Sales Lifecycle
Initial
Contact

Close Qualification

Sales
Cycle

Proposal Meeting
Sales Lifecycle
L(t)=CS(t)-IC(t)
Length
IC Initial Contact
CS Closed Sale

L->0
A business would like L to converge asymptotically to 0
A business wants
High Conversation Rate
CR->100%

Low Sales Cycle


L(t)->0
Sales Representatives

Want to sell fast

Want to sell more

Want to know if there is deviation from norm


Agent
Autonomy of decision
Re-activity to environment changes
Pro-activity in solving problems
Social ability to interact with other agents
Multi-Agent Architecture
Dispersed information
Email
Database
Notes
Instant Messaging

Distributed Features
Different activities different agents
Monitoring
Sentiment Analysis
Online Learning
Facts storage
Recommendation
Multi-Agent Architecture
Robustness
Simplicity
Asynchronous execution
Multi-tenant CRM

Can extend the CRM


Distribute system
Intelligent features
Flexibility
Recommendation System
Get knowledge
from experts
historical data
using machine learning

Online learning
Sentiment Analysis
Text-mining of
opinions
sentiments
emotions
expressed in text

Purpose
Generate knowledge
Interesting patterns
Sentiment Analysis
Natural
Language
Processing

Machine Sentiment Data/Text


Learning Analysis Mining

Information
Retrieval
Aspect Level Sentiment Analysis

Context of CRM Systems


Extract sentiments from emails
related to aspects or features of
sold products or services
Aspect Level Sentiment Analysis

(s,g,h,t)
s-sentiment
g-target
h-holder
t-time
Example
I have some concerns related to the HD Camera Resolution.

s-negative
g-Camera Resolution
g-customer
Purpose
The purpose is to describe the design and development of a multi-agent recommendation
system that uses aspect level sentiment analysis to convert prospects into customers in CRM
Systems.
Recommendation Agent Knowledge Base

CRM System

Architecture
Database of
Aspect Level
Sentiment Facts

Monitoring Agents Aspect Level Sentiment Analysis Agents

Product 1 Product 3 Product Aspects


Aspects Aspects Database
Monitoring

Product 2
Aspects ...
...

Data Sources

Email boxes
Messages CRM Notes
State of the art
Olszak, Celina. Multi-agent framework for social customer relationship
management systems. Proceedings of the Informing Science
and Information Technology Education Conference. Vol. 2013. No.
1. 2013.
Hu, Minqing, and Bing Liu. Mining opinion features in customer
reviews. In AAAI, vol. 4, no. 4, pp. 755-760. 2004.
Xiao, Yanshan, et al. Multi-agent system for customer relationship
management with SVMs tool. International Journal of Intelligent
Information and Database Systems 4.2 (2010): 121-136.
Q&A