Research on the implementation of teaching consumer online behavior pattern recognition technology in higher vocational college e-commerce education

Higher education is an important part of the national education system, is to realize the modernization of education, the construction of education, labor education as a form of education closely related to social production, bear the socialist modernization with Chinese characteristics to cultivate “big country craftsmen” an important historical mission [1-3]. With the continuous change of social production, the country’s demand for innovative talents and application-oriented talents is increasing, and the training of talents in full-time general undergraduate colleges and universities focuses on theoretical teaching and cultivation of scientific research literacy, neglecting the exercise of practical skills and so on [4-6]. Therefore, colleges and universities to strengthen the transformation and change of labor education, in line with the changes in social production and labor, catering to the development of science and technology, helps to integrate high-quality educational resources both inside and outside the school, focuses the limited educational resources on the central task of educational work, improves the quality of labor education, and highlights the status of labor education in the education system of colleges and universities [7-9]. As the main position and main channel for cultivating workers in the new era, innovating the traditional labor education mode and promoting the transformation of labor education is not only an important way for colleges and universities to seek their own development and enhance their schooling strength, but also a manifestation of colleges and universities attaching great importance to labor education and effectively assuming the important task of cultivating applied talents and highly sophisticated labor talents for the construction of socialist modernization with Chinese characteristics [10-12]. With the continuous updating and upgrading of digital technology, digital technology is widely used in the production and labor process of society, and colleges and universities should fully grasp the development opportunities of the digital era, promote the digital transformation of labor education, promote the transformation of labor education, create a good social atmosphere of “loving labor, respecting labor, and advocating skills”, and realize the high-quality development of labor education in colleges and universities [13-15]. And the use of big data analysis methods to study the digital transformation and quality monitoring mechanism of labor education in colleges and universities will help to cultivate students’ digital literacy and digital application ability, enhance college students’ adaptability to the future digital society, and help them to firmly establish the value of helping the great rejuvenation of the Chinese nation with labor [16-18].

This paper describes the related theories of customer segmentation, RFM, and consumer purchase decision, and in this way leads to a segmentation method for e-commerce consumer online behavior. The RFM model based on entropy weight method and the customer segmentation method based on clustering are combined to solve the problem of unclear segmentation boundary of traditional RFM model. Firstly, the MAF and MAM indexes of each consumer are calculated by using the sliding window method, and the consumers are dynamically labeled weekly, so as to determine the consumer groups to which the consumers belong well, and get the consumer state matrix. Then the one-step transfer probability matrix between consumer groups is calculated, and finally a preliminary portrait of e-commerce consumer characteristics is obtained, based on which the model is applied to study the consumer behavior characteristics of e-commerce consumers, and the model in this paper is applied to the teaching of e-commerce education to verify the feasibility of the model.

2

Consumer online behavior pattern recognition

2.1

Customer segmentation concept

Customer segmentation [19] is a crucial part of customer relationship management, and is gradually emphasized by enterprises, becoming one of the core of customer relationship management. Through the realization of customer segmentation, the enterprise can more effectively communicate with the customer and understand, to provide customers with better service.

There is no uniform specification of customer segmentation methods, in various industries, enterprises will be based on their own needs to carry out customer segmentation, in the academic community, according to the purpose of the study to determine the method of customer segmentation. Overall, there are four main types of customer segmentation methods based on different indicators: 1)

Customer segmentation based on statistical characteristics of customers.

2)

Customer segmentation based on customer behavior.

3)

Customer segmentation based on customer life cycle.

4)

Customer segmentation based on customer value-related indicators.

This study adopts customer segmentation based on RFM model, i.e., customer behavior from three aspects of purchase frequency, purchase amount, and proximity for segmentation research.

2.2

Introduction to the RFM model

R is the degree of proximity, i.e., the time interval between the time of the last purchase and today’s time; the longer the period, the greater the R. Once a threshold is exceeded, the customer becomes a churn customer. F is frequency, i.e., the number of purchases made during the study period; the more often a customer makes a purchase, the more likely he or she is to enter into a transaction with the business again. M refers to the amount, i.e., the amount of money spent by the customer during the study period; the larger M is, the more likely it is that the firm’s products and services will be purchased again.

RFM analysis is based on the following assumptions: 1)

Customers who have recently made a purchase with a company are more likely to buy again than customers who have not made a purchase.

2)

Customers who purchase more frequently have a higher preference for the company’s products and services and are therefore more likely to purchase again.

3)

Customers with higher total purchase amounts are more likely to have buying behavior again and are higher value customers.

In the enterprise, RFM is usually used to calculate the value of the customer, through the calculation of RFM three variables in the customer’s score, and then with the weight of each variable weighted sum to get the customer’s score, that is, the value of the customer, through the calculation of the score, but also can be similar to the score of the customer for clustering, to achieve the role of customer segmentation. After that, the RFM was modified to use the average purchase amount and the number of purchases to build a customer value matrix, so that the results were more concise and clear, and the customer groups were divided into four categories through the matrix analysis of customer value, and the classification results are shown in Figure 1.

According to the Customer Value Matrix, the best customers are the most valuable customers of the company, and the company should pay attention to this part of the customers, who are the main source of the company’s revenue. For Happy Spenders and Recurring Customers, they are the potential group to become the best customers, by increasing the purchase frequency of Happy Spenders and the average purchase amount of Recurring Customers, thus facilitating their conversion to the best customers. For uncertain customers, carefully identify potential customers and cultivate them into high-value customers, and for customers with no development potential, do not need to invest too much cost to maintain customer relationships.

2.3

Theories Related to Consumer Purchasing Decisions

Consumer purchase decision refers to the consumer in the face of a variety of behavioral choices, through the evaluation of products or services to choose a certain behavioral process, summed up to include the emergence of demand - the formation of motivation - behavioral choices –Late evaluation. After the research of related scholars, five characteristics of consumer purchase decision-making are summarized. 1)

The purpose of consumer purchasing decision. The whole decision-making process of consumers is to realize a certain demand, and satisfy the demand is the goal of decision-making, all the decision-making process is around this goal to complete and realize.

2)

The process of consumer purchasing decision. Consumer purchasing decision-making begins with the stimulation of consumer senses from internal and external factors, on the basis of which, the formation of purchase motives, followed by the choice of options to determine, and then the implementation of the purchase plan, each post-purchase experience will be accumulated in the brain, affecting the next consumer purchasing decision, which is a complete cycle of the process of experience accumulation.

3)

The performance of the characteristics of the subject of consumer purchasing decision-making. The consumer’s behavior of purchasing goods is the objective display of his subjective needs, is the external embodiment of the subjective will, but the occurrence of this behavior is accompanied by the influence of objective factors, the effect of this influence is becoming more and more obvious.

4)

The complexity of consumer purchasing decision. In the process of consumer purchase decision-making, full of psychological activities of the continuous change process, and the process is also complex. The overall view is mainly manifested in two aspects, one is the complexity of the content involved in the decision-making process. The second is the complexity of the factors affecting the purchase decision. Therefore, synthesizing the above analysis, we can see the complexity of consumer purchase decision-making and the embodied aspects.

5)

The situational nature of consumer purchasing decisions. First of all, the factors affecting decision-making are not static, they will change constantly with the changes of time, place and environment, so consumer buying decision-making has obvious situational nature. By understanding the characteristics of consumer purchase decision, we can better understand the consumer purchase decision model. In this paper, on the basis of the general pattern of consumer purchase decision-making, combined with the characteristics of consumer behavior, its purchase decision-making behavior is analyzed. The purchase decision process is shown in Figure 2.

3

Customer Segmentation Methods

3.1

Traditional RFM model

RFM model [20] distinguishes valuable users from a large amount of user data, is an important tool and means of evaluating the value of users, and is often used to study user loyalty and activity. In the RFM model, R indicates the time of the user’s most recent consumption, F indicates the frequency of the user’s consumption in the most recent period of time, and M indicates the amount of the user’s consumption in the most recent period of time. The three RFM indicators are used to describe the value status of customers.

E-commerce enterprises have a large amount of consumer behavior data, R, F, M three indicators are easy to obtain, reflecting intuitive. By analyzing the consumption behavior and obtaining the R, F and M indicators of users, we can explore the consumption habits of customers, and then complete the effective segmentation.

However, there are some shortcomings in the traditional RFM model, the traditional RFM model by sorting the value of R, F, M, the RFM indicators 5 equal points, and ultimately the RFM value of each customer with a numerical rating, the final value of the customer is the sum of the five levels. The disadvantages of this approach are that the weighting of the metrics is too subjective and the results of customer segmentation are too numerous.

3.2

Improvement of RFM model based on entropy weight method

Influenced by the individual differentiation of customers, the degree of influence of each indicator on customer value is different, and it is difficult to directly determine the degree of contribution of each indicator to customer value on the actual customer segmentation. Therefore, in response to traditional thinking, it is an objective approach to empowerment. It is mainly based on the degree of discrete data of each indicator, using information entropy to calculate the weight of each indicator, which can avoid the interference of subjective factors to a certain extent. The specific steps are as follows: 1)

Raw data processing: For the data with m consumers, the three indicators R, F and M are calculated for each consumer.

2)

Data standardization and elimination of the effect of the scale: Since the scales of the three indicators R, F and M are different, it is necessary to carry out standardization in order to eliminate the scales.

3)

Calculate the characteristic weights: Calculate the characteristic weights of the jth indicator for the ith customer: (1) $y_{i j} = \frac{x_{i j}}{\sum_{i = 1}^{m} x_{i j}}$ $${y_{ij}} = \frac{{{x_{ij}}}}{{\sum\limits_{i = 1}^m {{x_{ij}}} }}$$

Where m is the number of consumers. 4)

Calculate the information entropy: calculate the information entropy of each indicator, the information equation of indicator j: (2) $e_{j} = - k \sum_{i = 1}^{m} y_{i j} \ln y_{i j}$ $${e_j} = - k\sum\limits_{i = 1}^m {{y_{ij}}} \ln {y_{ij}}$$

where k = 1/lnm and m are the number of consumers. 5)

Calculate the information utility of the indicator: for a given indicator, the smaller the difference in x_ij and the larger the value of e_j, the less useful the indicator is, and vice versa. Therefore, define the information utility of the indicator (3) $d_{j} = 1 - e_{j}$ $${d_j} = 1 - {e_j}$$

6)

The weighted calculation of the various types of customers R, F, M indicators can be used to analyze the tendency of consumer behavior of customers. The weight of indicator j: (4) $w_{j} = \frac{d_{j}}{\sum_{j = 1}^{3} d_{j}}$ $${w_j} = \frac{{{d_j}}}{{\sum\limits_{j = 1}^3 {{d_j}} }}$$

7)

Is weighted for the original dataset to obtain the weighted metrics.

3.3

Dynamic RFM customer segmentation model based on K-Means clustering

Aiming at the shortcomings of the existing customer segmentation methods, this paper proposes a dynamic RFM customer segmentation model based on K-Means clustering [21] to obtain the user group that each user belongs to in each week, and accordingly derive the transfer probability matrix between user groups in order to accurately portray the change of user loyalty and the transfer characteristics between user groups.

The model is divided into four steps: 1)

The RFM model based on entropy weight method and the customer segmentation model based on K-Means are combined to determine the user group division standard.

The index calculation of RFM needs to utilize survival weighting, which is calculated as follows:

First, the total F (purchase frequency) and M (purchase amount) of each user during the observation period are converted to MAF (average weekly purchase frequency) and MAM (average weekly purchase amount) and R (number of weeks since the last purchase) using survival weighting to obtain the MAF, MAM, R indicators for each user during the observation period.

Second, due to the different units, the above MAF, MAM, R indicators are normalized to eliminate the effect of magnitude. Here, min-max normalization is used to normalize the data into the $[0, 1]$ $$\left[ {0,1} \right]$$ interval to obtain the normalized indicators, i.e., N-MAF, N-MAM, and N-R indicators.

Then, based on the entropy weight method to determine the weights for the above normalized indicators, to get the weighted indicators, namely, W-MAF, W-MAM, W-R indicators.

Finally, K-Means clustering algorithm is used to cluster the weighted indicators, and CHI, contour coefficient and inertia indicators are used as the evaluation standard of good or bad clustering to get the number of optimal user group divisions and the corresponding clustering center.

2)

Dynamic calculation of consumer MAF, MAM and R metrics

Calculate the MAF, MAM, and R metrics per user per week using a sliding window, for user i in week t the metrics are calculated logically as:

Calculate MAF (average weekly purchase frequency), MAM (average weekly purchase amount), and R (number of weeks since last purchase in week t) for user i in the $[t - 3, t]$ $$\left[ {t - 3,t} \right]$$-week window and min-max normalization, according to which the N-MAF, N-MAM, and N-R metrics per user per week can be obtained, and then using the metrics weights determined in step (1), the weighted metrics are obtained , i.e., W-MAF, W-MAM, W-R.

3)

Dynamic consumer clustering

Based on the weekly W-MAF, W-MAM, and W-R metrics of each user obtained in step (2), and the clustering center determined in step (1), adopt the Euclidean distance as a metric to dynamically mark the users on a weekly basis, determine the user group to which the user belongs on a weekly basis, and obtain the user state matrix.

4)

Consumer group migration probability matrix

Based on the user state matrix obtained in step (3), calculate the one-step transfer probability matrix between user groups. The specific calculation logic is: (1)

First calculate the probability of each state a_i at moment t: (5) $P {X_{t} = a_{i}}$ $$P\left\{ {{X_t} = {a_i}} \right\}$$

(2)

Calculate the probability of state a_i at moment t and state a_j at moment t + 1: (6) $P {X_{t} = a_{i}, X_{t + 1} = a_{j}}$ $$P\left\{ {{X_t} = {a_i},{X_{t + 1}} = {a_j}} \right\}$$

(3)

Calculate the transfer probability: (7) $\begin{array}{rcl} P_{i j} & = & P_{i j} (l) = P {X_{t + 1} = a_{j} | X_{t} = a_{i}} \\ = & P {X_{t} = a_{i}, X_{t + 1} = a_{j}} / P {X_{t} = a_{i}} \end{array}$ $$\begin{array}{rcl} {P_{ij}} &=& {P_{ij}}(l) = P\left\{ {{X_{t + 1}} = {a_j}|{X_t} = {a_i}} \right\} \\ &=& P\left\{ {{X_t} = {a_i},{X_{t + 1}} = {a_j}} \right\}/P\left\{ {{X_t} = {a_i}} \right\} \\ \end{array}$$

where l is the state space.

The design steps of the dynamic RFM customer segmentation model based on K-Means clustering involved in this paper are shown in Fig. 3.

4

User profiling based on RFM modeling

Based on the above RMF model analysis, clustering analysis and further variable correlation verification of the consumption data of A e-commerce product, a preliminary portrait of the user characteristics of consumers of A e-commerce product can be obtained, and consumers are sorted according to their high or low value, and the basic attributes of the users of each category are sorted out to make a basis for the construction of the subsequent user portrait. Consumers are categorized into the following four categories according to the order of customer value from high to low: important value users, general value users, key development users, and key retention users, and the categorization results are shown in Table 1. The platform can adopt precise marketing strategies according to the user profiles of different categories, provide differentiated services for e-commerce consumers, and improve the efficiency and capability of platform services.

Table 1.

Consumer feature portrait

		Important value user	General value user	Key development user	Important user
Active (R)		Very high	Higher	Very high	Very low
Loyalty (F)		Very high	Higher	Lower	Lower
Contribution (M)		Very high	Higher	Lower	Lower
Gender characteristics		They are very masculine	Comparative men	Comparative men	Comparative men
Age characteristics	Average	40.6	38.9	36.3	36.1
	Median	39	37	36	35
	Structural characteristics	Middle and young, middle-aged and young	Middle and young, young people are more young than middle age	Middle and young, young people are far from middle age	Middle and young, young people are far from middle age
Product purchase	Loyalty	A lot of	More	Less	Less
Product purchase	Activity	Be very active	More active	More active	Inactivity

The first category of important value users with the highest customer value. E-commerce activity and loyalty are extremely high, and the consumption level is high. The second category of general value users with higher customer value. A more male-oriented consumer group, with an age distribution dominated by middle-aged and young people. Higher consumption level, higher e-commerce loyalty, higher activity and product interest. The third category focuses on developing users with medium customer value. More skewed towards male consumer groups, age distribution is dominated by young and middle-aged people. Very high e-commerce activity, low loyalty and consumption levels. The fourth category of important retention users, with low customer value. Biased towards male consumer groups, with the highest proportion of females among the four categories, and an age distribution dominated by middle-aged and middle-aged people. With very low e-commerce activity, loyalty and consumption levels, they are “new users”.

As this paper mainly carries out the study of user image on the user behavior data of e-commerce consumers, which has certain limitations, in order to further improve the exploration of the application of user image for e-commerce consumption, the relevant data of another e-commerce consumption product B is introduced for the study to make up for the limitations in terms of data representativeness.

Three metrics of RFM model are selected, and there are three other browsing data metrics of user e-commerce behavior, namely R (consumption proximity), F (consumption frequency), and M (consumption amount). Due to the different units of measurement, there are large deviations in the values of the indicators, and the deviations will lead to inaccurate clustering results, which cannot accurately analyze the user characteristics. The original data is standardized, and the processed data meets the standard normal distribution, which can be used as the clustering variable of the K-Means clustering algorithm.

As mentioned earlier, the construction of RFM model is mainly divided into three stages: standardized processing, hierarchical analysis method (AHP), and K-means clustering analysis. The operation method of hierarchical analysis method is consistent with the previous section, in order to increase the consistency and facilitate the comparison and analysis of user profiles of two products. The matrix after normalization processing is shown in Table 2.

Table 2.

The post-processing matrix

	R	M	F
R	0.1114	0.0765	0.1306
M	0.3335	0.2306	0.2176
F	0.5559	0.6954	0.6523

The RFM model’s standardized indicators were used as clustering variables, and the K-Means clustering algorithm was used for clustering analysis. After testing, K was assigned to 3, 4 and 5 respectively, and the results showed that the clustering results had a more stable performance when K=4, so the consumers were classified into 4 classes. The categorization results are shown in Table 3, and according to the weights obtained from the hierarchical analysis method (AHP), the values and rankings of the four categories of consumers can be obtained.

Table 3.

Final clustering center and consumer value rankings

Categories	User ratio/%	R	F	M	SCORE_RFM	Customer value ranking
1	0.13	-0.86445	-0.12419	19.952232	12.69574	1
2	2.21	-0.59949	-0.03741	3.38315	2.19645	2
3	62.52	-0.6495	0.0259	-0.05578	0.04068	3
4	35.14	1.20759	-0.0428	-0.17985	-0.25364	4

The number of users in category 1 with the highest value is the least, 0.13%, the number of users in category 2 with the second highest value ranking is less, 2.21%, and the number of users in the last two categories in the ranking is more, 62.52% and 35.14%, which is basically in line with the expectation, but the clustering result is still somewhat different from that of the A product. On the sales network platform of product B, the value of high-value users is relatively higher.

Combined with the above RFM model for A product e-commerce consumer users according to the clustering results of consumption behavior, combined with the value of the clustering results of the sorting, the study under the classification for the analysis of user browsing behavior data, mainly from the number of times browsing analysis. Under the RFM model classification system, the number of times users browse before the last purchase is analyzed, and the results of the analysis of the number of times browsed are shown in Table 4. The total number of browsing reached 6341 times.

Table 4.

Analysis of user browsing times in RFM model

Browsing frequency	1-10	11-20	21-40	41-60	61-80	81-100	101-200	201-300	>301	Total
Categories1		2	2		3		2	1		10
Categories2	33	23	36	29	15	8	5	2	3	154
Categories3	2025	978	605	170	70	28	59	15	12	3962
Categories4	1426	405	245	72	36	16	13		2	2215
Total	3884	1408	888	271	124	52	79	18	17	6341

Category 1 and category 2 users, the proportion of high-frequency word browsing is higher, combined with the relative data of the proportion of the analysis is more reasonable, Figure 4 and Table 5 show the analysis of the proportion of the chart and the analysis of the proportion of the table respectively. According to the results of the relative percentage, it can be seen that along with the increase in the value of user behavior, the number of user browsing is relatively high, high value users in the consumption of the user before, there are more times of browsing, category 1 users browsing more than 100 times of the percentage of 30%, much higher than the category 4 users, the number of times of its browsing more than 100 times of the percentage is only 0.69%, indicating that as a more mature consumer users, their Consumption behavior is after more careful consideration or product comparison after the decision. Low-value users generally only browse fewer times before consumption, indicating that users make decisions more quickly.

Table 5.

The number of user browsing in RFM model is the score

Browsing frequency	1-10/%	11-20/%	21-40/%	41-60/%	61-80/%	81-100/%	101-200/%	201-300/%	>301/%
Categories1	0.00	20	20	0.00	30	0.00	20	10	0.00
Categories2	21.43	14.94	23.38	18.83	15	9.74	3.25	1.30	1.95
Categories3	51.11	24.68	15.27	4.29	1.77	0.71	1.49	0.38	0.30
Categories4	64.38	18.28	11.06	3.25	1.63	0.72	0.59	0.00	0.09

Combined with the above results of data analysis related to the e-commerce purchase behavior of Product B, the user profile is presented as shown in Table 6. Based on the sorting and analysis of the e-commerce purchase user portrait of Product B, it is found that there is a big difference with the e-commerce purchase user portrait of Product A collated in the previous section. First, from the viewpoint of activity, loyalty and contribution, the three indicators of the high-value user group of Product A are all very high, basically the same as its final value distribution, with a more neat law, while in the user distribution of Product B, although the activity and contribution of the important value users are very high, the loyalty (F-value) is, on the contrary, very low, which indicates that among the high-value users, there is a group purchasing behavior, with a larger number of single purchasing number of purchase orders, such customers should be the key service object of product B of the e-commerce platform.

Table 6.

User portraits of product B

		Important value user	General value user	Key development user	Important user
Active (R)		Very high	Higher	Higher	Very low
Loyalty (F)		Very low	Higher	Higher	Lower
Contribution (M)		Very high	Higher	Lower	Lower
Browsing behavior	Loyalty	Very high	Higher	Lower	Very low
Browsing behavior	Activity	Lower	Higher	Lower	Higher

5

Analysis of the effectiveness of teaching implementation

In testing the learning effect of higher vocational students, the total score of the knowledge point test of the 19th grade e-commerce A, B and C classes was used as a data support as a way to check the effect of learning. The research object of the teaching experiment is the students of a higher vocational school in the class of 2022, majoring in e-commerce, and the situation of the students in three teaching classes: the control experimental class A (50), the control experimental class B (50) and the experimental class C (50), and the male and female ratios of the three teaching classes are comparable.

In this study, one-way ANOVA was used to determine the overall level differences among the three classes after the teaching experiment, and the results of one-way ANOVA and post hoc multiple comparisons are shown in Tables 7 and 8, respectively. The results show that there are significant differences in the achievement indicators between the three classes. The LSD test results show that there is no significant difference between class A and class B, but there is a significant difference between class A and class C, and there is a significant difference between class B and class C. The results of the one-way ANOVA and post-hoc multiple comparisons are shown in Tables 7 and 8, respectively. Overall, class C as the experimental class and with the control experimental class (class A and class B) are significantly different in the post-test scores. The significance level of *mean difference is 0.05.

Table 7.

Variance analysis of total score

	Sum of squares	freedom	Mean square	F	Significance
Intergroup	528.456	3	265.156	3.939	0.021
Within group	8222.546	125	66.854
Total	8742.139	126

Table 8.

Post-mortem

	Class A	Class B	Mean difference(A-B)	Std.e	Sig.	95%CI upper limit	95%CI lower limit
LSD	1	2	-1.28245	1.80954	0.440	-4.8633	2.30077
		3	-4.93522^*	1.87127	0.007	-8.6424	-1.2354
	2	1	-1.28141	1.80948	0.440	-2.3008	4.8645
		3	-3.65644^*	1.71945	0.034	-7.0554	-0.2513
	3	1	-4.938189^*	1.871255	0.007	1.2334	8.6422
		2	-3.65652^*	1.71952	0.034	0.2554	7.0530
Tamhane	1	2	-1.28154	1.87375	0.871	-5.8654	3.3051
		3	-4.93821^*	1.92207	0.035	-9.6543	-0.2154
	2	1	-1.28545	1.87329	0.871	-3.3058	5.8684
		3	-3.65278	1.66212	0.085	-7.7047	0.3954
	3	1	-4.93568^*	1.92812	0.035	0.2155	9.6544
		2	-3.652896	1.66455	0.085	0.3915	7.7045

The mean and variance analysis of the three classes are shown in Table 9, the six achievement factors of classes A and B are smaller than C. From the analysis of variance, the data of class C relative to classes A and B in the four areas of group work performance, number of responses, practice, and attendance are the most concentrated in the mean, with variances of 7.33, 1.05, 1.39, and 0.42, respectively, and the data are the most stable. The data in the two areas of test scores and presentations are more concentrated in the mean and the data are generally stable.

Table 9.

The mean and variance of the three classes of the ais

Achievement part	Class	N	Mean	Variance
Test scores at (25%)	A	50	19.41	12.22
	B	50	19.74	5.85
	C	50	21.15	6.66
Team homework (20%)	A	50	14.25	9.16
	B	50	14.63	8.17
	C	50	15.91	7.33
Number of responses (15%)	A	50	11.88	1.1
	B	50	12.02	1.09
	C	50	12.38	1.05
Practice (15%)	A	50	11.91	1.61
	B	50	11.98	1.65
	C	50	12.51	1.39
Speaking at (15%)	A	50	11.97	1.65
	B	50	11.96	1.61
	C	50	12.26	1.69
Attendance(10%)	A	50	8.94	0.48
	B	50	8.98	0.46
	C	50	9.08	0.42

Table 10 shows the results of independent t-tests for the three classes. The significance of class A and B in test scores, group work scores, and practice is less than class C. The p-values of the tests for class C and class B are 0.014, 0.011, and 0.027, respectively, which are less than the p-values of the tests for class A and class B. The p-values of the tests for class C and class B are 0.014, 0.011, and 0.027, respectively. The non-significance in number of responses, presentations, and attendance are greater than class C. That is to say, in the three aspects of “thinking”, “experiencing” and “expressing”, both in the separate assessment and the overall satisfaction are control experiment A<control experiment B<experiment C. Experiment C classes are significantly more effective than control experiment A and B classes. The teaching effect of experimental class C is significantly better than that of control experimental classes A and B. Experimental class C can wait for equal realization of “teaching thinking”, “teaching experience” and “teaching expression”, and the teaching effect between all three is better than that of control experimental class A and B.

Table 10.

The f value and the p value of the t test between the three classes

Achievement part	Class	N	F	P
Test scores at (25%)	A&B	50	7.446	0.611
	A&C	50	5.333	0.017
	B&C	50	0.064	0.014
Team homework (20%)	A&B	50	0.071	0.629
	A&C	50	0.386	0.018
	B&C	50	0.135	0.011
Number of responses (15%)	A&B	50	0.088	0.563
	A&C	50	0.23	0.055
	B&C	50	0.685	0.125
Practice (15%)	A&B	50	0.014	0.669
	A&C	50	0.116	0.055
	B&C	50	0.042	0.027
Speaking at (15%)	A&B	50	0.031	0.879
	A&C	50	0.024	0.24
	B&C	50	0.123	0.267
Attendance(10%)	A&B	50	0.111	0.707
	A&C	50	0.162	0.403
	B&C	50	0.003	0.606

6

Conclusion

This paper proposes a dynamic RFM customer segmentation model based on K-Means clustering to achieve the purpose of accurately identifying e-commerce consumers’ online behavioral patterns, and applies the model to e-commerce teaching with a view to improving the teaching effect.

The model divides the users of Product A and Product B into four categories of important value, general value, key development, and key maintenance users, and as the value of the user’s behavior increases, the number of their browsing times will correspondingly increase, and the number of users with browsing times greater than 100 times in Class 1 accounts for 30%, which is much higher than the rest of the three categories of users.

Among the users of Product B, the activity and contribution of important value users are high, while the loyalty is very low, suggesting that there may be group purchasing behaviors among high-value users, and e-commerce operators can take this kind of customers as the key service objects of Product B. The above results illustrate that the model in this paper can accurately identify consumer online behavior patterns, which provides a direction for e-commerce operators to improve their service quality and develop high-value customers.

After the implementation of the teaching, class C, as the experimental class, has a significant difference between its performance and that of the control class in the post-test, and the significance of class A and B in the test scores, group work scores, and practice is smaller than that of class C. The p-values of the test for class C and class B are 0.014, 0.011, and 0.027, respectively, which are smaller than those of the test for class A and class B. The non-significance of the test in the number of responses, presentations, and attendance for class A and class B is greater than that for C class. In any aspect, the teaching effect of experimental class C is significantly greater than that of control experimental classes A and B. The results of experimental class C in the six achievement factors are not significant. That is to say, the teaching effect of experimental class C is better than that of classes A and B among the six achievement factors, and the application of consumer online behavior pattern recognition technology helps to improve the teaching effect in e-commerce education.

Language:: English

Publication timeframe:: 1 times per year
Journal Subjects:: Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other

Journal RSS Feed

Research on the implementation of teaching consumer online behavior pattern recognition technology in higher vocational college e-commerce education

Yi Yang

Qiang Li

Published Online: Jun 05, 2025

Received: Jan 12, 2025

Accepted: May 08, 2025

DOI: https://doi.org/10.2478/amns-2025-1114

KeywordsCustomer segmentation, RFM model, e-commerce education, K-Means clustering

© 2025 Yi Yang and Qiang Li, published by Sciendo.

This work is licensed under the Creative Commons Attribution 4.0 International License.

Keywords
Customer segmentation, RFM model, e-commerce education, K-Means clustering