Intelligent Marketing Decision Model Based on Customer Behavior Using Integrated Possibility Theory and K-Means Algorithm

E-commerce has been welcomed by many human societies over the last few years. Today, fewer people can be found that the term e-commerce is unfamiliar for them [1]. The development of information and communication technology has created an opportunity for institutions to carry out their business activities quickly. Today, most middle and large organizations in the world have a website. Websites are the tool for marketing [2]. The importance of marketing in today's world is very valuable that misleading marketing decisions may lead to the destruction or failure of the organization, especially in today's global marketplace, which is constantly evolving and rivals offer new products and services every day. On an ecommerce site, e-marketing should help to the customers for buying different goods, which requires customer information [3]. Marketers believe that, in the online shopping activity, customers disclose information that reflects their behavior such as the customer's email address and phone number, occupation, age, education, address, and customer orders, order history, methodologies for paying, favorites, and their requirements that can be used to understand customer behavior [4]. The reason for collecting information is we need to get a sense of customer behavior, how they decide on their needs, and what internet marketing methods are suitable and successful for them. In fact, the availability of customer data and transaction records provide a better understanding of customer behavior and their willingness to market [5]. By recognizing its customers, the marketer can expand relationships with customers and expand the various fields of work as well as improve the decision-making process in the marketing process. The volume of data gathered from the customer data is huge, and analysis of them is one of the problems, but one way to overcome this problem is to use the data mining technique. This tool uses from customer information to help marketers and customers find their products and services more efficiently [6].

Applying data for marketing as to make decisions intelligent was used profoundly [7]. For agricultural products where the lifecycle is significant it is necessary to make use of data driven approaches to increase the value chain. Recommender systems are useful tool in intelligent marketing based on customer suggestions. Li et al. [8] employed the attention allocation index concept to propose an intelligent prediction algorithm to join users of the same brand products. Martínez-López and Casillas [9] introduces knowledge discovery in database to marketing modeling as a decision support for consumer behavior. They hybridized artificial intelligence methods as a knowledge base marketing management approach. A new paradigm of industry 4.0 related products influenced marketing efforts and decisions. Khan et al. [10] investigated the impact of artificial intelligent and industry 4.0 on consumer behavior using a meta-analysis. They found that intention to buy, acceptance and need for recognition are significantly impact the consumer behavior. Business models are important in forming marketing models and decisions. Hence, Moradi and Dass [11] studied the artificial intelligence use in B2B marketing and listed several pros and cons. The study emphasized the role of machine learning methods throughout the customer life cycle. The analysis employed technology acceptance model to suggest recommendations for marketers and business policy makers. Considering user experience in developing marketing recommendation system is significant. Due to large amount of data, it is necessary to make use of user data into product design [12]. Rahim et al. [13] examined the purchasing behavior of Muslims towards Halal products. They collected a dataset and used structural equation modeling partial least square for statistical analysis. Outsourcing marketing decisions for different purposed were investigated in several researches [14], [15]. Eco-tourism infrastructure effectiveness, demand segmentation, personal relationships, knowledge sharing, and size of the company were the most emphasized factors justified outsourcing marketing decisions. Detecting frauds in market transactions was investigated in the study of Eshghi and Kargari [16] to analyze the trend and developing scenarios for marketers.

In this paper, we design an e-marketing model using the recommendation system. The customer's attitude in the online shopping process is not constant and is dynamic and variable based on gender, age, geographical location and tastes. So, we cannot definitely discuss the behavior of customers as a matter of fact. First, collect information from customers, and after preprocessing data and separating inappropriate data, the relevant data is being prepared for the proposed login. The output of the K-Mean algorithm includes strategies and recommendations for marketing, and finally, with helping of the theory, the possibility and the requirement of the proposal are calculated. The related tools: To use the marketing model, we need to introduce the tools and explain the reasons for using each one. In fact, the proposed system of the research, by analyzing the behavior of its users, proposes the most suitable items for business executives.

Among the mathematical theories in terms of ambiguity, the Possibility Theory can be considered the most appropriate and coherent theory in the analysis of the uncertainties of the decision environment. In summary, the content of this theory can be expressed in such a way that in the analysis of environmental events and conditions that we are not only looking for possible events, and in uncertain structures we are looking for all possible events that are introduced with the degree of possible events and the degree of possible contradictory events (see Table 1). In the theory of probability, we try to construe it with the probability of occurrence of the event with the probability that it does not occur, while in more general theory, the uncertainty of an event can be determined by two numbers [2], [6]. The degree of the possibility of self-occurrence and the degree of necessity of the event (the degree of necessity of the event = the degree of probability of contradiction - 1).

In this paper k-means technique is applied for clustering the customers based on their views and purchase behavior.

The K-means algorithm:

Clustering is the process of grouping an item set into similar item classes. A popular clustering algorithm proposed by MacQueen is named K-means. The main idea of K-means is to discover k clusters that the records in each cluster are similar to each other and differ from other cluster records. The K-means algorithm is often used to preprocess data or to cluster because of its simplicity and efficiency [1].

The architecture of our proposed marketing model consists of several steps as shown in Figure 1:

Step 1: Data Collection: Customer data is collected after customers have made their online shopping.

Step 2: Classify Customers: The K-Mean algorithm receives customers' data as inputs, and classifies customers.

Step 3: The marketer provides advice and recommendations.

Step 4: the numbers in the name of the possibility and the requirement for any recommendation in the system output are provided.

In this section, we will implement the system for an online store. From the virtual stores, we consider the DigiKala store. DigiKala store is the largest online store in the country, with a range of diverse products for all segments of society and in every age group. The range of goods of this site is in the range of the following goods: digital goods, home appliances, beauty and health, culture and art, sports and entertainment. The steps of implementing the marketing Model with the K-Mean Algorithm:

Step 1: Get the necessary information from actual customers:

The customer purchases his/her product from the DigiKala website, the following information is received from the customer: email address, mobile phone number, the code of product, date of birth, gender, the place of residence, purchased goods. Sample data collected from customers from DigiKala site is as follows (see Table 2):

Step 2: Data clustering using the K-Means algorithm:

At this stage, the data are categorized using the K-Means clustering algorithm. Output of the K-Means algorithm for the two characteristics of the age and goods (see Figure 2 and Figure 3):

Suggested strategy for K-Means algorithm output:

Vertical Line: desired product: Household appliances are in range from 1 to 3. Cultural and sports goods are in range from 3 to 6. Digital goods are in range from 6 to 10.

Horizontal Line: Customer's age range: The customer's age is divided into 10.

Number of clusters: k = 3.

For more detail information, see Table 3.

Output of K-Means algorithm for gender and goods are shown in Figure 4 and Table 4.

Suggested strategy for K-Means algorithm output:

Vertical Line: desired product: Household appliances are in range from 1 to 3. Cultural and sports goods are in range from 3 to 6. Digital goods are in range from 6 to 10.

Horizontal Line: Customer gender: male (1), female (2).

Number of clusters: k = 3

For more detail information, see Table 4.

From the two tables above one can extract the strategy suggested in Table 5:

Step 3: Using theory of possibility:

At this stage, data analysis is done using the theory of Possibility. In this way, after classifying and analyzing information, we must examine the degree of probability and necessity of the event. The following table, based on the latest reports and customer information from the DigiKala website in the last 6 months, estimates the average amount of purchases of goods based on customers' age ranges as possible in Table 6.

Now we would like to answer the following questions:

What is the chance for customers to buy digital goods from 11 to 38 years old? What is the Possible and necessity of this event?

Customers are between the age range of 11 and 38: customers are from 11 to 18 and 18 to 38 years old, is equivalent to the following:

The Necessity of the above-mentioned event is calculated according to the following formula:

Therefore, based on the necessity and possibility of purchasing digital goods at the age range of 11 to 38 years, we can conclude about 99% of digital sales are in this age range, so new entering digital goods will be sent to people in this age range either by email or SMS. The purpose of this paper is to design a marketing model that covers the disadvantages of past models. In order to improve the marketing planning process, a marketing model has been provided that benefits from the previous models and covers its disadvantages. In this proposed model, the method of computer model marketing and group decision making have been used in combination.

In this paper, an electronic marketing system was proposed by using the recommendation system and the possibility theory based on the customer's random behavior. From the perspective of technology, the proposed marketing system has been designed to provide the necessary synchronization to operate with existing technologies. The purpose of this system is the development of e-mail marketing, which is needed in today's market, customer analysis and reporting marketers for marketing. This marketing system will create a competition that will be effective in the marketplace. In this system, the marketer can find out, at any time, the number of sales of different types of goods in different areas, based on the gender and age of the customers and provide marketing recommendations. Then, using the theory, it is possible to examine the possibility and necessity of recommendations and to assure the marketer of the recommendations.