A talk delivered at
Understanding Consumers in Digital Era
IIM Lucknow, Noida Campus
DECODING RATINGS FOR SUPERIOR SERVICE IN RESTAURANTS
Using text to understand customers
BROAD AGENDA
• CHALLENGES OF DATA
• HOW TO TACKLE IT
• SOME TERMINOLOGY
• CASE STUDY : RESTAURANT + TEXT ANALYTICS
• WHY RATINGS ARE NOT HOLY
• LUNCHBOX – AN INTELLIGENT RESTAURANT APP
Did you know? Total traffic to restaurant review
sites exceeds 100 million per month in India.
CHALLENGES
You are drowning in too much data – social channels, feedback forms, emails
Whom to trust? Reviews can be often contradictory/biased across channels
Difficult to maintain parity in customer experience across channels seamlessly
Offering personalized service and offers is not always possible
SOLUTION
Adopt a 360 degree approach
Read and understand the reviews
– internal or external
Extract actionable insights for
operational improvements
Unify your internal feedback with
POS transactionsUnderstand what sets your
competition ahead
Personalized offers & services for
each guest
Own an intelligent restaurant
management system (RMS)
Increase footfalls and repeat visits for you !
RATINGS – HOW MUCH IS BETTER ?
Increasing scope of differentiating operational improvements
Decreasing scope of customer loyalty
BRIEF TERMINOLOGY
You build an algorithm, machine learns patterns, machine predicts, rinse & repeat.
MACHINE LEARNING
TEXT ANALYTICS
Analyzing unstructured text, assign structure, load into a BI/program to visualize
CASE STUDY
HOW WE HELPED A RESTAURANT SERVE THEIR CUSTOMERS BETTER
PROBLEM STATEMENT
The client had thousands of customer reviews which they wanted to analyse - to
understand customer feedback and identify improvement opportunities.
The broad questions we focused on;
What did they say about the restaurant?
Keywords & topics of discussion across the
comments
What elements of the restaurant would they
want improved? – service, staff behaviour,
ambience etc.
When did the customer visit the store?
How is client’s traffic distributed over time?
Ticket sizes across multiple customer
dimensions – age, gender, ratings, location,
time of visit etc.
Overall customer sentiments & views about
UCH
PRIMARY FOCUS AREAS SECONDARY FOCUS AREAS
FOCUS AREAS
TOPICS KEYWORDS SENTIMENT POINT OF SALES
APPROACH
Extract data and
validate
Corpus from
social media
Tokenise and
remove stop
words
Initiate ML models ,
NER , parsers & topic
algorithms
Initiate detection rules for
topics, keywords, gender,
sentiment and multi-word
concept detection
Final Output
PRE - PROCESSING PARSING & ANALYSIS OUTPUT
Part of Speech
(POS) Tagger
DATA SNAPSHOT
Bill No. Net Amount Membership No. Gender Profession Marital Status Date Rating Comment
SL-0220 678 EXXXXXX FEMALE SALARIED UNMARRIED 02-02-2013 5
This is a fantastic, inexpensive
casual place to have delicious……
SL-0221 1202 EXXXXXX MALE SALARIED MARRIED 15-02-2013 4
Great shakes and burgers. The
sandwiches…
SL-0222 707 EXXXXXX MALE SALARIED MARRIED 18-02-2013 3
Very good food but the service is
slow.
SL-0223 791 EXXXXXX FEMALE SALARIED MARRIED 21-02-2013 4
A friend steered me here for the
…..
SL-0224 619 EXXXXXX FEMALE SALARIED UNMARRIED 27-02-2013 3
Bah! Below is my outdated review.
…..
TOPICS – OUR GENERIC MODEL
TOPICS
TOPIC PERCENTAGE NUMBER OF RECORDS
Overall Visit Experience 47.0% 22,320
Service 24.7% 11,730
Taste/Quality 18.9% 8960
Recommendation 2.7% 1300
Referral/Loyalty 1.9% 920
Temperature (too hot/cold) 1.1% 530
Quantity 0.9% 420
Music 0.8% 400
Pricing (Too low/high) 0.6% 290
Drinks 0.6% 260
Options/Menu Choices 0.5% 250
Ambience 0.3% 150
TOPICS VS SENTIMENT
Negative Neutral Positive
Topic % # % # % #
Overall Visit Experience 10.5% 270 34.0% 2360 50.8% 19,310
Service 11.3% 290 20.6% 1430 27.6% 10,510
Taste/Quality 47.1% 1210 28.8% 2000 14.8% 5630
Recommendation 3.7% 260 2.7% 1040
Referral/Loyalty 1.2% 30 1.0% 70 2.2% 820
Temperature (too hot/cold) 10.5% 270 1.9% 130 0.3% 130
Quantity 3.9% 100 2.0% 140 0.5% 180
Music 8.9% 230 1.6% 110 0.2% 60
Pricing (Too low/high) 1.9% 50 1.7% 120 0.3% 120
Drinks 1.2% 30 2.2% 150 0.2% 80
Options/Menu Choices 2.3% 60 2.0% 140 0.1% 50
Ambience 1.2% 30 0.4% 30 0.2% 90
TOPICS – AC TEMPERATURE
TOPICS – AC TEMPERATURE
Some of the randomly picked negative reviews on temperature were –
- A Remarks
- The Ac Was Too Cold
- Your Restaurant Is Too Cold
- Too Cold We Were Shivering
- Change The Music Style AC A Bit Too Cold
- Temperature Of The Restaurant Too Cold Air Conditioned
TOPICS – AC TEMPERATURE
RATINGS ARE NOT HOLY
It’s not recommended to rely on the ratings alone– they tend to paint a different story than is.
A customer might give a rating 5, but deplore you in his review.
A quick look at reviews vs the actual sentiment of the text.
A sample review with rating of 4 ;
“Desserts Very Bad”
Rating (out of 5) Negative Neutral Positive
4 123 412 2,609
3 77 208 972
2 41 55 109
1 8 6 11
FINAL RECOMMENDATIONS
Improve speed of service
Redesign menu for easy read
Decrease portion size
Use ACs at ambient temperature
Hire more female staff
Expand beer selection
DEMO
LUNCHBOX – AN INTELLIGENT RESTAURANT APP
HOW WE DO IT ?
Single platform to
analyse customer
reviews – from
internal or social
channels
Actionable
intelligence on
competitors and
upcoming threats
Unified feedback
management system
– real time analysis
of internal & social
feedback
Target customers with
hyper-personalized
offers – both real-time
and app-based
campaigns
OUR PLATFORM
10.7 Mn 92.6 K 62.6 K16
reviews restaurants user profilestopics
As on 31st October, 2015
QUESTIONS ?
MANAS RANJAN KAR
manas@jsm.email
+91-9971 420 188
www.unlocktext.com

Mining customer reviews to decode businesses

  • 1.
    A talk deliveredat Understanding Consumers in Digital Era IIM Lucknow, Noida Campus DECODING RATINGS FOR SUPERIOR SERVICE IN RESTAURANTS Using text to understand customers
  • 2.
    BROAD AGENDA • CHALLENGESOF DATA • HOW TO TACKLE IT • SOME TERMINOLOGY • CASE STUDY : RESTAURANT + TEXT ANALYTICS • WHY RATINGS ARE NOT HOLY • LUNCHBOX – AN INTELLIGENT RESTAURANT APP
  • 3.
    Did you know?Total traffic to restaurant review sites exceeds 100 million per month in India.
  • 4.
    CHALLENGES You are drowningin too much data – social channels, feedback forms, emails Whom to trust? Reviews can be often contradictory/biased across channels Difficult to maintain parity in customer experience across channels seamlessly Offering personalized service and offers is not always possible
  • 5.
    SOLUTION Adopt a 360degree approach Read and understand the reviews – internal or external Extract actionable insights for operational improvements Unify your internal feedback with POS transactionsUnderstand what sets your competition ahead Personalized offers & services for each guest Own an intelligent restaurant management system (RMS)
  • 6.
    Increase footfalls andrepeat visits for you !
  • 7.
    RATINGS – HOWMUCH IS BETTER ? Increasing scope of differentiating operational improvements Decreasing scope of customer loyalty
  • 8.
    BRIEF TERMINOLOGY You buildan algorithm, machine learns patterns, machine predicts, rinse & repeat. MACHINE LEARNING TEXT ANALYTICS Analyzing unstructured text, assign structure, load into a BI/program to visualize
  • 9.
    CASE STUDY HOW WEHELPED A RESTAURANT SERVE THEIR CUSTOMERS BETTER
  • 10.
    PROBLEM STATEMENT The clienthad thousands of customer reviews which they wanted to analyse - to understand customer feedback and identify improvement opportunities. The broad questions we focused on; What did they say about the restaurant? Keywords & topics of discussion across the comments What elements of the restaurant would they want improved? – service, staff behaviour, ambience etc. When did the customer visit the store? How is client’s traffic distributed over time? Ticket sizes across multiple customer dimensions – age, gender, ratings, location, time of visit etc. Overall customer sentiments & views about UCH PRIMARY FOCUS AREAS SECONDARY FOCUS AREAS
  • 11.
    FOCUS AREAS TOPICS KEYWORDSSENTIMENT POINT OF SALES
  • 12.
    APPROACH Extract data and validate Corpusfrom social media Tokenise and remove stop words Initiate ML models , NER , parsers & topic algorithms Initiate detection rules for topics, keywords, gender, sentiment and multi-word concept detection Final Output PRE - PROCESSING PARSING & ANALYSIS OUTPUT Part of Speech (POS) Tagger
  • 13.
    DATA SNAPSHOT Bill No.Net Amount Membership No. Gender Profession Marital Status Date Rating Comment SL-0220 678 EXXXXXX FEMALE SALARIED UNMARRIED 02-02-2013 5 This is a fantastic, inexpensive casual place to have delicious…… SL-0221 1202 EXXXXXX MALE SALARIED MARRIED 15-02-2013 4 Great shakes and burgers. The sandwiches… SL-0222 707 EXXXXXX MALE SALARIED MARRIED 18-02-2013 3 Very good food but the service is slow. SL-0223 791 EXXXXXX FEMALE SALARIED MARRIED 21-02-2013 4 A friend steered me here for the ….. SL-0224 619 EXXXXXX FEMALE SALARIED UNMARRIED 27-02-2013 3 Bah! Below is my outdated review. …..
  • 14.
    TOPICS – OURGENERIC MODEL
  • 15.
    TOPICS TOPIC PERCENTAGE NUMBEROF RECORDS Overall Visit Experience 47.0% 22,320 Service 24.7% 11,730 Taste/Quality 18.9% 8960 Recommendation 2.7% 1300 Referral/Loyalty 1.9% 920 Temperature (too hot/cold) 1.1% 530 Quantity 0.9% 420 Music 0.8% 400 Pricing (Too low/high) 0.6% 290 Drinks 0.6% 260 Options/Menu Choices 0.5% 250 Ambience 0.3% 150
  • 16.
    TOPICS VS SENTIMENT NegativeNeutral Positive Topic % # % # % # Overall Visit Experience 10.5% 270 34.0% 2360 50.8% 19,310 Service 11.3% 290 20.6% 1430 27.6% 10,510 Taste/Quality 47.1% 1210 28.8% 2000 14.8% 5630 Recommendation 3.7% 260 2.7% 1040 Referral/Loyalty 1.2% 30 1.0% 70 2.2% 820 Temperature (too hot/cold) 10.5% 270 1.9% 130 0.3% 130 Quantity 3.9% 100 2.0% 140 0.5% 180 Music 8.9% 230 1.6% 110 0.2% 60 Pricing (Too low/high) 1.9% 50 1.7% 120 0.3% 120 Drinks 1.2% 30 2.2% 150 0.2% 80 Options/Menu Choices 2.3% 60 2.0% 140 0.1% 50 Ambience 1.2% 30 0.4% 30 0.2% 90
  • 17.
    TOPICS – ACTEMPERATURE
  • 18.
    TOPICS – ACTEMPERATURE Some of the randomly picked negative reviews on temperature were – - A Remarks - The Ac Was Too Cold - Your Restaurant Is Too Cold - Too Cold We Were Shivering - Change The Music Style AC A Bit Too Cold - Temperature Of The Restaurant Too Cold Air Conditioned
  • 19.
    TOPICS – ACTEMPERATURE
  • 20.
    RATINGS ARE NOTHOLY It’s not recommended to rely on the ratings alone– they tend to paint a different story than is. A customer might give a rating 5, but deplore you in his review. A quick look at reviews vs the actual sentiment of the text. A sample review with rating of 4 ; “Desserts Very Bad” Rating (out of 5) Negative Neutral Positive 4 123 412 2,609 3 77 208 972 2 41 55 109 1 8 6 11
  • 21.
    FINAL RECOMMENDATIONS Improve speedof service Redesign menu for easy read Decrease portion size Use ACs at ambient temperature Hire more female staff Expand beer selection
  • 22.
    DEMO LUNCHBOX – ANINTELLIGENT RESTAURANT APP
  • 23.
    HOW WE DOIT ? Single platform to analyse customer reviews – from internal or social channels Actionable intelligence on competitors and upcoming threats Unified feedback management system – real time analysis of internal & social feedback Target customers with hyper-personalized offers – both real-time and app-based campaigns
  • 24.
    OUR PLATFORM 10.7 Mn92.6 K 62.6 K16 reviews restaurants user profilestopics As on 31st October, 2015
  • 25.
  • 26.