استخراج تغییر رفتار مشتری در الگوهای متوالی فاصله زمانی فازی

Comprehending changes of customer behavior is an essential problem that must be faced for survival in a fast-changing business environment. Particularly in the management of electronic commerce (EC), many companies have developed on-line shopping stores to serve customers and immediately collect buying logs in databases. This trend has led to the development of data-mining applications. Fuzzy time-interval sequential pattern mining is one type of serviceable data-mining technique that discovers customer behavioral patterns over time. To take a shopping example, (Bread, Short, Milk, Long, Jam), means that Bread is bought before Milk in a Short period, and Jam is bought after Milk in a Long period, where Short and Long are predetermined linguistic terms given by managers. This information shown in this example reveals more general and concise knowledge for managers, allowing them to make quick-response decisions, especially in business. However, no studies, to our knowledge, have yet to address the issue of changes in fuzzy time-interval sequential patterns. The fuzzy time-interval sequential pattern, (Bread, Short, Milk, Long, Jam), became available in last year; however, is not a trend this year, and has been substituted by (Bread, Short, Yogurt, Short, Jam). Without updating this knowledge, managers might map out inappropriate marketing plans for products or services and dated inventory strategies with respect to time-intervals. To deal with this problem, we propose a novel change mining model, MineFuzzChange, to detect the change in fuzzy time-interval sequential patterns. Using a brick-and-mortar transactional dataset collected from a retail chain in Taiwan and a B2C EC dataset, experiments are carried out to evaluate the proposed model. We empirically demonstrate how the model helps managers to understand the changing behaviors of their customers and to formulate timely marketing and inventory strategies.

In today's world, where the market is competitive and products are multitudinous, various marketing strategies have been developed to attract customers to endorse products or services [26]. In such a situation, customer switching behavior [23], [24] and [40] takes place all the time. Thus, managers often have to make changes in their marketing plans. In many cases, they might have no idea about how and where to start comprehending these changes and how they come about. To address this problem, they often adopt traditional approaches, such as market surveys; yet, such approaches time-consuming and costly and they are unable react to changes immediately. As a result, comprehending changes of customer behavior accurately and responding to customer need in a timely manner has become an essential problem. There are two main explanations for the significance of this problem [30]: (1) to follow trends based on changes, managers like to know in which direction trends are proceeding so as not to be left behind. They analyze changes of customer behavior to provide products and services corresponding to the changing needs of the customers. (2) To stop or to delay undesirable changes, managers also like to about undesirable changes as soon as possible so that they can develop remedial measures to stop or delay the pace of such changes. As the problem states, we investigate two research questions. First, how managers can be responsive to changes of customer behavior in a dynamic market? Second, how managers can detect and utilize the changes they indentify in customer behavioral patterns in order to respond in an accurate and timely manner? In the management of electronic commerce (EC), many companies have developed on-line shopping stores to serve customers and immediately collect buying logs in databases. Therefore, managers can segment databases in accordance with different time-periods to analyze what is changing and how it has been changed [5]. According to changes in customer behavior, managers can provide the right products or services over different time-periods. When it comes to the topic of changing behaviors, the primary task is to adopt practical tools to discover customer behavioral patterns. As a result, many data-mining techniques have been proposed to discover useful information, such as product bundling [50], RFM (Recency, Frequency, and Monetary) sequential patterns [13], product recommendation [22] and [25], cross-selling [31], customer profiling [36] and [49], personalized marketing [8], personal moving profiles in wireless networks [21], intrusion detection [47], mental health [16], and fuzzy time-interval sequential patterns [10]. Among the numerous data-mining techniques applied in EC, mining sequential patterns plays an important role in helping managers discover customer behavior over time. The problem of mining sequential patterns was first introduced in the mid-1990s. It consists of discovering a set of subsequences that occur frequently in a sequence database [1]. An Amazon.com shopping example of a sequential pattern is described as follows: having bought the book Alice's Adventures in Wonderland, a customer returns to buy the book Introduction to Logic and then the book Queen Victoria: A Personal History. A mapping of letters and items’ names is given as: (a, Alice's Adventures in Wonderland), (b, Introduction to Logic), (c, Queen Victoria: A Personal History), and (d, The Lord of the Rings: the Fellowship of the Ring). Moreover, Chen et al. [9] proposed a generalization of sequential patterns, called time-interval sequential patterns, which reveals not only the order of items, but also the time-intervals between successive items. A time-interval sequential pattern in a database can be described as follows: having bought a, a customer returns to buy b in two months and then c in a half month. The Amazon.com manager, therefore, can refer to this type of pattern and develop his/her inventory or marketing plans. Such an approach, however, could cause a sharp boundary problem. That is, when a time-interval is near the boundary of two adjacent ranges, we either ignore or overemphasize it. For instance, let a time-interval of ti1 be 1 ≤ tg < 4 and that of ti2 be 4 ≤ tg < 7, where tg is the time gap between two successive items. Then, if the time gap between items a and b is near 4, either a little larger or smaller, it is difficult to say whether the time gap is in ti1 or in ti2. Hence, the case can only be one hundred percent in ti1 or in ti2. This difficulty can be adequately tackled by using fuzzy techniques, since fuzzy set theory allows the time gap to be 50% in ti1 and 50% in ti2 at the same time. Additionally, other patterns might be discovered from the same database, such as: having bought a, a customer returns to buy b in three months and then c in one month and having bought a, a customer returns to buy b in two and a half months and then c in one month. These detailed but similar patterns make it cumbersome for managers to make decisions. Therefore, a fuzzy extension, called fuzzy time-interval sequential patterns (FTSP), was proposed by Chen and Huang [10] to find the fuzzy time-interval between items in sequential patterns. A fuzzy time-interval sequential pattern takes the following form. Having bought a, a customer returns to buy b in a Long period and then c in a Short period. The pattern can be represented by (a, Long, b, Short, c) as well. This simple example indicates that the fuzzy concept is better than the partition method because fuzzy sets provide smooth transitions between members and non-members of a set. As mentioned in the above introduction, we know that there are several advantages to applying fuzzy time-interval sequential patterns in EC. First, the knowledge of managers in decision making can be represented more naturally and appropriately by fuzzy logic, and partitioning and representing time information are types of manager knowledge. Second, it is widely acknowledged that many real-life situations are intrinsically fuzzy, and partition of time information is one such situation. Third, using linguistic terms is simple and easy for managers. As part of executive work activities, managers often take care of strategic issues and long-term trends [48]. Therefore, verbal communication with linguistic terms is preferred for the exchange of soft knowledge. Moreover, since customer requirements in EC markets are often changing, managers need to gain more general and concise knowledge, not precise knowledge in the form of figures, for making quick-response decisions. Accordingly, any manager can first define the linguistic terms that are meaningful and understandable to them and then finally discover specific knowledge. Although applying this technique to mining FTSP is workable, it still fails to take into account customer behavioral changes in fast-changing EC environments. For example, a fuzzy time-interval sequential pattern (a, Long, b, Short, c) may be available for the previous year. The pattern, however, is not necessary a trend this year, and it may be substituted by (a, Short, b, Long, d). If managers cannot capture the customer behavioral change in time, two failed beliefs between the previous year and the present year still exist in their minds, including: (1) They still believe that a customer will buy b after buying a in a Long period. In fact, the period has been changed from Long to Short. (2) After buying a and b, a customer will buy c in a Short period; however, the item has been changed from c to d and the period is not Short but Long. Without updating this knowledge, managers fail to provide appropriate products or services to customers and may adopt inappropriate inventory strategies with respect to time-intervals. No studies, to our knowledge, have yet to address the issue of changes in fuzzy time-interval sequential patterns so that managers can detect and utilize these types of changes. Therefore, our research goal is to propose a novel change mining model, MineFuzzChange, to detect FTSP change. The remainder of this paper is organized as follows. Section 2 reviews related works. Section 3 defines a similarity measurement for fuzzy time-interval sequential patterns. Section 4 presents the MineFuzzChange model for mining changes in fuzzy time-interval sequential patterns. Section 5 shows the experimental results of the proposed model. Conclusions are drawn in Section 6.

Fuzzy time-interval sequential pattern mining is a useful method for discovering customer purchasing patterns through time from transactional databases. It is beneficial to managers for quick and easy decision making. Nevertheless, the issue of changes in this type of pattern has not been discussed by the past studies. By not updating their knowledge based on time trends, managers fail to formulate appropriate marketing or inventory strategies. Accordingly, this problem inspired us to investigate a change mining model, MineFuzzChange, in order to address more real business circumstances. We proposed three types of change patterns, emerging patterns (time trend change), unexpected changes, and added/perished patterns, to detect changes in fuzzy time-interval sequential patterns. Brick-and-mortar transactional and B2C EC datasets are used to evaluate the proposed model. We show some interesting results about the three types of change patterns. This study reveals valuable findings in regard to a number of issues, including seasonal, daily, and low-price products. Managers can refer to these findings to handle the marketing and production of their products in the future. In summary, the proposed model informs managers that change patterns discovered from the past databases can be used as lessons about the future in management.The change model for mining fuzzy time-interval sequential patterns represents a new and promising research area in data mining. For business practitioners, it provides rough, quick, and valuable references to managers when making decisions, especially in EC management. In an EC market, customers can search for online information quickly and change their preferences anytime; therefore, managers have to utilize this type of model to follow trends and provide suitable products or services to their customers. Managers can find the reason for changes over time and make the right reaction to changes. Furthermore, since the environment of the EC market is more competitive than that of the brick-and-mortar market, managers understand that expanding new niche markets will be important as well. Summarily, managers responding to these changes in a timely and accurate manner can increase their return on their investment. For academic researchers, future research can expand upon this topic in several possible directions. In crisp cases, we can develop other pattern detection methods to discover changes, such as quantitative sequential patterns, multi-level sequential patterns, RFM sequential patterns, and sequential patterns with different minimum supports. In fuzzy cases, we can also devise new models to mine fuzzy quantitative sequential patterns, fuzzy multi-level sequential patterns, etc. In addition, other kinds of EC datasets, such as B2B and C2C, or brick-and-mortar datasets should be used to test the proposed model's performance. Ultimately, we discuss the segmentation issue of datasets. Since the time-point to divide a dataset into two sub-datasets will influence outcomes, managers should have prior knowledge to carry out this model. A good time-point to derive and mine sub-datasets can reveal the significant fuzzy time-interval sequential patterns and help them to find more interesting changes.