Research work - My Datafication

Research work

Master Thesis

Identifying customer satisfaction patterns via Data Mining. The case of Greek online shops
K. Kalaidopoulou

Thesis Supervisors: Ass. Prof. Katerina Pramatari, Prof. Georgios Doukidis
Assistant supervisor: PhD Candidate Anastasia Griva

March 2017
Grade: 10.00 / 10.00

Abstract: It is a marketplace reality that more e-retailers promise their customers that online experiences will be satisfying, and therefore it becomes crucial to understand what creates a satisfying customer experience. In a competitive setting, such as the Internet market, where competition may be only one click away, it is even more important to satisfy the customers, as switching costs have been diminished. Therefore, many organizations have identified the need to not only understand customer purchase behavior, but also the customer satisfaction through the online purchase journey. Additionally, technological advances enable direct communication with customers by sending surveys and/or promotional information. This new method of collecting customer responses has led to large volumes of satisfaction data, available for knowledge extraction. Many researches, in both academia and the industry have focused on these data in order to calculate the satisfaction levels or identify those factors which affect satisfaction. However, to the best of my knowledge there is no research which is focused on extracting customer satisfaction patterns from survey data in order to identify different satisfaction types. 

To address that issue, this thesis presents a Data Mining based framework, which can be used to identify customer satisfaction types from satisfaction survey data in an e-shop or in e-commerce in general. A customer satisfaction type describes how satisfied the customer is in terms of specific attributes, and how the satisfaction is changed between them, i.e. attributes. For instance, some customers may be highly satisfied with the website characteristics such as navigation or security, but less satisfied with product characteristics, such as price or variety. Such an information can be extracted by applying the clustering data mining technique to customer satisfaction data, and more specifically by applying the k-means algorithm to satisfaction surveys collected after the customer checkout and after the order delivery.

This thesis is based on the Design Science approach, which means that it first explores the relevant researches for the research problem and then an artifact is developed. The artifact is a Data Mining based framework, which draws on CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology. It describes an explicit way to handle the data through the following five phases. (A) Business and Data Understanding, with sub-tasks the Data Acquisition and Data Exploration, (B) Data Preparation with sub-tasks the Data modeling and Data Sampling, (C) Data Mining Modeling, (D) Evaluation, and (E) Deployment. As part of the Data Modeling task, the framework proposes the following sub-tasks: Data Integration, Data Cleansing, Data Transformation, and Data Validation.

Then the proposed framework is applied to real data and more specifically to survey data from 83 e-shops and 11 industries of the Greek market. The results, evoked from the customer web-based surveys in two phases, indicate that there exist different types of customer satisfaction. Next, this thesis uses sales data from one of the 83 e-shops in order to extract customer segments based on their purchase history and combine the results with the satisfaction types. It is validated that different customer segments have different prominent satisfaction types, indicating that e-shops should address each segment’s need separately.

Lastly, the proposed framework has some limitations, but it contributes a lot to both academia and the industry. It identifies for the first time the customer satisfaction types from satisfaction surveys, using a clustering algorithm. In light of the major findings, the thesis sets forth strategic implications for customer loyalty in the setting of e-commerce. Moreover, it combines the results with customer segments and thus enriches the results with a combination of knowledge from two sources. Such finding can support e-shops’ decision making and customer-oriented strategies. Lastly, further research could be applied using a different Data Mining technique and evaluate if different ways of identifying satisfaction types exist, e.g. by applying text analytics on customer reviews.

Read full research here.


How shoppers buy a specific product category in retail stores? The case of face care products
K. Kalaidopoulou, A Griva, P Sarantopoulos,13th Student Conference of Management Science and Technology, May 2016

Abstract: Customer satisfaction is the key of the success of any retail store or business and has an effect on their profitability. Companies have identified the importance to gain insights on their customer’s behavior, so as to better satisfy their needs. Both retailers and suppliers have understood the need to cooperate in order to combine their data, gain knowledge and improve the offered services and customer satisfaction. Additionally, they have realized the potential mutual benefits if they combine their individual knowledge and expertise. Nowadays, the advent of business analytics, aids this cooperation. Thus, data mining techniques could be utilized to analyze the vast amount of data both retailers and suppliers have, to extract knowledge and support data-driven decision-making. However, not enough research has been conducted to analyze such data in order to investigate consumers’ behavior regarding a specific product category. This study presents an effort to fill this gap by introducing a Data Mining-based framework, which could be used to discover sales affinities in customers' visits in a supermarket related to a specific product category, and extract behavioral insights. The utility of this framework has been evaluated by applying it in real data of two representative supermarket stores of a Greek retailer. The proposed approach is useful for both an academic and business perspective. It gives retailers the opportunity to extract how shoppers buy specific categories when they visit their stores, while it also enriches the suppliers’ knowledge about their shoppers. This knowledge can be used to support the decision-making process for all stakeholders in the retail domain, and improve their relationships with shoppers.

Read full publication here.


Extracting Greek elections tweet’s characteristics
K. Kalaidopoulou, A. Romanou, G. Lekakos, 9th Mediterranean Conference on Information Systems, October 2015

Abstract: Social media offer platforms that anyone can use, giving the opportunity to share information among networks in an easy and interactive way. It is not a surprise that social media marketing has become a primary focus on both digital and traditional revenue models of businesses. In this work, information sharing by users in the context of Twitter is studied, by modeling message’s characteristics and users’ behavior about Greek 2015 January elections. A detailed data set about tweets’ characteristics such as length, existence of URLs or hashtags and mentioning of other users, is collected after the elections day, and the relationships between related users and network’s responses on the shared tweets, are examined. An unsupervised clustering model is implemented on tweets’ characteristics using CRISP-DM methodology. The empirical results suggest the existence of different content groups, such as tweets with extensive text, URLs and hashtags which can be characterized as “Linked” type of shared content.

Read full publication here.


Identification of Customer Segments via Data Mining
K. Kalaidopoulou, G. Kanellopoulos, A.Griva, 12th Student Conference of Management Science and Technology, May 2014

Abstract: In the ever-changing digital world, attracting new customers and retaining the older ones is a major challenge for modern businesses. Consumers' behavior and expectations for service have changed dramatically in recent years, as they have become far more demanding. Many digital businesses have identified the need to become more customer-centric to respond to the ever increasing demands of consumers, and cope with the global competition. At the same time, the advent of Big Data analytics has created new ways for businesses to analyze the vast amount of data that have been stored and remain unutilized. Thus, in order to understand, and satisfy their shoppers, modern businesses seek to exploit the new technologies to analyze any customer related data, generated by their interaction with their shoppers. Leveraging new technological trends, this paper utilizes transactional data and Google Analytics data of a modern e-commerce shop that sells technological equipment, in order to identify customers’ behavior; mine the distinct customer segments visited this online store, their unique characteristics, and their purchase patterns. As well, this study aims to evaluate the effect in shopper’s perception caused by the diversification of the company in household categories. The steps of CRISP- DM (Cross Industry Standard Process for Data Mining) methodology have been adapted to structure research’s data analytics approach. The results of this research approach could be useful for the online shops, as it gives them the opportunity to understand their shoppers and support several decisions, ranging from personalized promotions and recommendations, to modifying the online catalogue’s layout.

Read full publication here.


Digital Leadership in Greece
K. Kalaidopoulou, I. Rompou, K. Fraidaki, 12th Student Conference of Management Science and Technology

Abstract: In recent years, Digital Marketing becomes more and more important in the business promotion plan, while businesses are increasing the annual budget available. The Digital Marketing Tools available to the companies are a commercial or non-commercial site, advertising on search engines, especially on Google and Social Media, Search Engine Optimization, applications for smartphones, e-mail marketing, sending newsletters etc. Each of these tools serve different objectives of a promotion strategy and thus more than one should be used for an efficient 360-degree marketing strategy. 
The aim of this research is to identify contemporary Digital Marketing practices of large companies operating in Greece. In particular, it includes a thorough study of the tools' usage and an evaluation of their effectiveness. Then, we analyze the trends and best practices in the field of digital marketing, which are suggested by the respondents. To achieve the above objectives, we built a structured questionnaire, which was completed during interviews with 32 Greek companies from 11 different sectors.

Read full publication here.


Investigating shopping visits patterns across different store types: The case of a grocery retail chain
K. Kalaidopoulou, L. Koutsokera, V. Stavrou, A. Griva, 13th Student Conference of Management Science and Technology, May 2016

Abstract: Consumers' behavior and expectations of service have changed dramatically in recent years, as they have become more demanding, regarding the quality of services and the needs to fill in. Many organizations have identified the importance to understand their shoppers’ preferences, and satisfy their needs. As any other business, so do retailers, understanding their shoppers, and more specifically the reasons they enter stores has always been among their greatest aspirations. Many retailers face the questions of (a) how shoppers visit their different retail stores, (b) what needs they need to satisfy in each store visit, and (c) how they can sense their shopping behavior and thus their visiting patterns, with a view to offer them more efficient shopper oriented services. At the same time, business analytics techniques that have been developed can combine and process large datasets to enable broader and deeper analysis than previously possible. Therefore, data-driven decision-making is widely adopted by managers, following the enthusiasm for the notion of Big Data. Big Data analytics now drive near every aspect of our modern society, including retail industry, financial services etc. Retailers have realized the importance of applying these new technological trends to sense their shoppers, support decision-making and offer them appropriate services to satisfy their needs. In essence, retailers seek to infer a wealth of consumer behavioral insights by extracting the knowledge hidden in the vast amount point-of-sale (POS) data they already have, and until recently have only been used for accounting purposes. The investigation of related literature pointed out a major gap regarding the examination of retail chains based on the store dimension. So far, most studies handle retail chains as a bulk of stores, or they solely focus on a specific store. They do not examine different shopper behaviors per store type within the same retail chain. Furthermore, the existing studies focus on perceived data which do not always indicate the actual behavior of a shopper during his visits. Motivated by the aforementioned, in our research we aim to examine how a grocery retail chain could benefit via analyzing its shoppers’ visits patterns derived from actual data, across two different store types and extract useful insights by exploring the aforementioned questions. More specifically, we propose a data mining-based approach which investigates shopping visits patterns from POS data. The utility of this approach has been evaluated by applying it in real data provided by a major Greek retail chain. The proposed approach is useful from both an academic and business perspective. Our results indicate that shoppers’ behavior diverges across different store types even within the same retail chain. The findings highlight the need of applying customized marketing actions to the different store types, in order to offer shoppers more efficient shopper oriented services.

Read abstract here.

No comments:

Post a Comment