基于AI的个性化推荐引擎打造一个忠诚的客户群

了解在工作场所、社交媒体或真实社交圈中你不太熟悉的人的个性总是一个有趣的想法。这可以帮助企业了解客户、员工和合作伙伴的心理,这反过来又有助于建立成功的合作伙伴和忠诚的客户。一个推荐引擎,可以提供一个客户的个性洞察力可以非常有效地保持忠诚的客户基础,通过调整他们的需求和行为模式,同时提出一个新的产品/服务。然而,创造这样的发动机,并保持它的最新变化与人性的行为方面,是一项艰巨的任务。
展开查看详情

1.Create a Loyal Customer Base by Knowing Their Personality Sourav Mazumder, IBM Analytics (smazumder@us.ibm.com) Aradhna Tiwari, University of South Florida (aradhna@mail.usf.edu) #AI9SAIS

2. Agenda v Team Introduction v Big 5 Personality Model v Prior Work and Our approach v The Datasets we used v The Technologies v Modeling techniques used v Findings v How to Use the Customer Loyalty Indicators for Business Use cases ? v Demonstration v Q and A #AI9SAIS 2

3.Team Introduction Kaushik Dutta Sourav Mazumder Associate Professor Data Science Thought Leader Muma College of Business, IBM Analytics University of South Florida Aradhna Tiwari Sai Seetha Ram Stacey Ronaghan Nomula Business Sr. Data Scientist Analytics Business Analytics Graduate Student Graduate Student IBM Analytics Muma College of Muma College of Business, Business, University of University of South South Florida Florida #AI9SAIS 3

4.Big 5 model for Personality Traits Big Five Model : Describes how a person engages with the world. #AI9SAIS 4

5.We based our work on two (recent) prior works Aspects Prior Work 1 Prior Work 2 Hypothesis Used Big 5 Traits -> Customer Empowerment -> Big 5 Traits -> Brand Identity (Excitement, Customer Satisfaction/Loyalty Competence, Sincerity, Ruggedness) -> Brand Loyalty Year 2017 2014 Industry Retail Automobile Sample Retail customers across the world above 21 Customers for a particular brand (150) yrs age (278) Big 5 Traits used for All 5 Only Consciousness, Extroversion and Emotional comparison Range are used Key Conclusion “Therefore, it can be inferred that companies’ “The results indicate that there is a positive and strategies to promote loyalty and satisfaction direct relationship between extroversion and among consumers should consider focusing excitement, conscientiousness and excitement, more on consumers related to conscientiousness and competence, excitement Conscientiousness and Agreeableness” and loyalty, competence and loyalty, sincerity and loyalty and ruggedness and loyalty and other hypotheses were rejected” #AI9SAIS 5

6.Our Hypothesis and Approach Hypothesis - Every individual has certain innate Personality traits those contribute to his/her propensity to become a loyal customer for any product/brand with varied degree Approach – ü Identify Personality traits using Data across various Brands and Products ü Verify the consistency of those traits across various product types ü Use of Data Science techniques – less Time and Money and more reliable model compared to Survey based approach #AI9SAIS 6

7.About the Datasets used v The dataset comprises of ratings and reviews from an Online Retailer. v The datasets are selected for 5 types of products namely – v Electronics v Book v Grocery v Pet Supplies v Baby v Dataset Citation: v R. He, J. McAuley. Modeling the visual evolution of fashion trends with one-class collaborative filtering. WWW, 2016 v J. McAuley, C. Targett, J. Shi, A. van den Hengel. Image-based recommendations on styles and substitutes. SIGIR, 2015 #AI9SAIS 7

8.Key Attributes from the Datasets v Reviews: v Ratings: v Customer ID v Customer ID v Product ID v Product ID v Rating v Review Text #AI9SAIS 8

9.Technology Stack Technologies Usage Watson API for Personality Insights For generating Personality Traits associated with the Big 5 based model Watson Studio For development of overall Model using Notebooks in a Collaborative way REST Data source for Spark Used the library to parallelize calling Watson API through GitHub Link Spark for multiple sets of input and collating the result in single Dataframe Python 3.5 Used as the programming language. PySpark ML 2.0 Used this library to identify important features using Random Forest algorithm #AI9SAIS 9

10.Diving Deep into Model v First, the output of Watson Personality Insights API is used to get Big 5 Personality Traits - Conscientiousness, Extraversion, Agreeableness, Openness, and Emotional Range. v Correlation analysis between the Big 5 Personality Traits and the Average Rating (separately for each Big 5 Personality Trait) v Next Random Forest algorithm is used to create a model where v Big 5 Personality Traits are used as Independent Variables and v Average Rating (signifying Loyalty) are used as Dependent variable v Feature Importance estimation using Random Forest to identify the Big 5 traits those are most important #AI9SAIS 10

11. Findings – The Personality Traits that can indicate Loyal Customers (1/2) Person Coefficient relating Big 5 Traits to Rating (Loyalty) 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 Electronics Pet_Supplies Book Grocery Baby -0.05 -0.1 Conscientiousness Agreeableness Extraversion Emotional range Openness #AI9SAIS 11

12. Findings – The Personality Traits that can indicate Loyal Customers (2/2) 0.6 Random Forest's Feature Importance across 5 different Product Types for Big 5 Traits 0.5 0.4 0.3 0.2 0.1 0 Electronics Pet Supplies Book Grocery Baby Conscientiousness Agreeableness Extraversion Emotional Range Openness #AI9SAIS 12

13. Creating a Customer Loyalty Database Spark Environment -------- Calling -------- Watson Data Merging Wrangling Text -------- API Loading Preparation -------- Data Sources - Emails, Speech to Text from IVR, Loyalty Review comments, Indicator etc. based on the Important important Personality Big 5 Personality Traits Traits Traits Loyal Applications Customer base #AI9SAIS 13

14.Using Customer Loyalty Indicators for new or potential Customers Spark Streaming What are u looking for Qs’ in Calling -------- Watson online site, Chat texts, etc. -------- API -------- Merging -------- Cleaning Text Preparation Loyalty Indicator based on the Important Big 5 Applications Personality traits Traits #AI9SAIS 14

15.Quick Demonstration Demo #AI9SAIS 15

16.#AI9SAIS 16

17.References The following papers were referred that infer a strong relationship between personality and loyalty. v Jan 2017, Journal of Business and Retail Management Research (JBRMR), Vol. 11 Issue 2 , Javier Castillo v Jan 2014,Case Study: product group of Isfahan Iran Khodro,Dr. Hassan Ghorbani, Seyede Maryam Mousavi v John, O. P., & Srivastava, S. (1999). The Big Five trait taxonomy: History, measurement, and theoretical perspectives. In L. A. Pervin, & O. P. John (Eds.), Handbook of personality: Theory and research (pp. 102–138). New York: Guilford Press. #AI9SAIS 17