Article Preview
TopIntroduction
According to statistics (Kemp, 2020), 3.8 billion people worldwide used social media in January 2020, comprising about 50% of the total global population (7.75 billion). Another interesting fact is that out of all social media users, 2.5 billion people, or 66% of users, employed e-commerce platforms. Liu, Q., et al. (Liu, Shao, Tang, & Fan, 2019) analyzed and summarized the main reasons for social media’s ability to attract large numbers of users. The first reason was the opportunity to exchange and share user information, while the second was the ability for users to respond to each other in real time. Furthermore, social media platforms have become important tools for communication and business between purchasers and vendors, leading to the birth of social media marketing (Chen & Lin, 2019). In the sales negotiation process, the vendor can rapidly respond to the buyer via various social media platforms such as Facebook, Twitter, Instagram and YouTube. In addition, use of social media platforms can be used to analyze and process large volumes of data (Big Data) (AI-Ibrahim & Alzamil, 2019; Sarkar, Roy, Giri, & Kole, 2019; Xu, Jiang, Wang, Yuan, & Ren, 2014; Zhu, Gong, Zhang, Zhao, & Zhou, 2018) through analyzing and processing interactions between vendors and buyers. This leads to understand patterns and relationships of social media users for improving performance of user interactions on an online business in the future (Ali, Mohammed, & Rajamani, 2014; J. Pan et al., 2007; Jansen B. J., Sobel K., & Cook, 2011; Liao & Tasi, 2019; Long, 2010; Palmisano, Tuzhilin, & Gorgoglione, 2008; Singh, Chauhan, & Dhir, 2019; Tanantong, Sanglerdsinlapachai, & Donkhampai, 2020; Turban, 2010; Wakita, Oku, Huang, & Kawagoe, 2015).
As mentioned above, the analysis and processing of Big Data, i.e., social media data, is important in order to support business decision making. One popular analysis and processing technique for Big Data is association rule mining (Si H. et al., 2019). This is a data mining technique in the form of unsupervised learning. Unsupervised learning is a learning method without a human trainer, instead relying on input data which is automatically analyzed and processed by an association rule mining algorithm to produce association rules as output. The research studied in (Khongtuk T., 2016) compared the efficiency of association rule mining algorithms, concluding that the most suitable algorithms for working with Big Data were FP-Growth, Apriori and PrePost+. In (Chunmatcha, 2016; Hongkun, 2017; Worachotekamjorn & Sanrach, 2014), the researchers demonstrated applications of use of association rule mining techniques in analysis and processing of data related to the business sector.
In (Worachotekamjorn & Sanrach, 2014), association rule mining were applied to analyze health food, cosmetics and consumer products in the MLM business. The data consisted of 24,879 orders for 108 products. Significant data attributes i.e. order reference numbers, catalogue numbers of products ordered and number of products ordered, were selected for analysis. The data underwent cleansing to eliminate incomplete data items (such as null values) and then was transformed into crosstab report format. Analysis was then undertaken by the WEKA program using the Apriori algorithm. The conditions set were that support was 5% and the confidence value was 50%. 6 association rules met these conditions. One such rule, which had a confidence value of 61%, stated that “Customers who buy washing powder will buy herbal toothpaste.”
Nattayaporn Chunmatcha et al. (Chunmatcha, 2016) presented an association rule mining method for purchase order data using the FP-Growth technique with the Rapidminer Studio 6 tool. The point of sale (POS) dataset from the University of California, Irvine (UCI) database, which contained data from a total of 108,131 sales, was employed. Each entry contained information on goods purchased by customers. Data was processed as follows: goods were classified; then the tool was instructed to find frequent itemsets. The condition was that support had to be equal to 0.1 and 0.2, with confidence values equal to 0.7, 0.8 and 0.9. The result was that for conditions where support was equal to 0.1 and confidence value was 0.7, 15 association rules were found, while in conditions with support equal to 0.2 and confidence values of 0.7 and 0.8, 6 association rules each were found. In experimental conditions where support was equal to 0.1, with a confidence value of 0.8, 11 association rules were found. And in conditions where the confidence value was 0.9, no association rules were found under any conditions of support (0.1 and 0.2). The obtained rules can be applied to analyze and plan a business strategies for selling goods and services.