Online shopping consumer perception analysis and future network security service technology using logistic regression model

Research theory and establishment of consumer perception model

DEA model design

DEA is a well-established tool within the field of economic research, serving as a pivotal instrument for extracting substantial managerial insights. This method, initially formulated by Charnes and Cooper in 1978, encompasses several notable models. Among these models, the Charnes-Cooper-Rhodes model is primarily applied to the estimation of frontiers, while the Banker-Charnes-Cooper model focuses on assessing variable scale benefits (Charnes, Cooper & Rhodes, 1978) .

Suppose s is the decision unit, and X_ns the input vector of the decision unit with dimension n.Y_sm is the output vector produced by the decision unit with dimension m, and Z is the set of all possible outputs of Y when the input is X. Then, the effectiveness H_s of the decision unit can be calculated according to Eq. (1). (1)${\displaystyle {\mathit{H}}_{\mathit{s}}=\frac{{\mathit{U}}^{\mathit{Z}}{\mathit{Y}}_{\mathit{sm}}}{{\mathit{V}}^{\mathit{Z}}{\mathit{X}}_{\mathit{sn}}}.}$

In Eq. (1), V is a measure of input, and U is a measure of output.

SOR model design

The SOR model is a foundational framework employed for examining consumer behavior patterns. It constitutes a study of how individuals process information from the vantage point of cognitive psychology (Hussain et al., 2023). The SOR model posits that human behavior is a consequence of stimuli arising from the interplay between consumers’ internal psychological processes and external interactions, which subsequently instigate consumer motivation. This motivation, in turn, leads to the formulation of consumer consumption decisions, culminating in shopping behavior. Following the shopping experience, consumers engage in assessments based on their perceptions (Grădinaru et al., 2022; Alanadoly & Salem, 2022). The SOR model is grounded in three key elements: stimuli, organism, and response, which serve as the focal research variables and components (Karim et al., 2021). A detailed explanation of the SOR model structure and its specific meaning is shown in Fig. 4. The SOR model structure and model variable definitions based on negative evaluation are shown in Fig. 5.

The SOR model, when applied to negative evaluations, is formally established. In order to investigate the relationship between negative evaluations, consumer perception of risk, and consumer behavior, statistical analyses, including the Pearson correlation coefficient (PCC) and factor analyses, are conducted.

Suppose there are two variables, n and m.Cov is the covariance, and Q is the standard deviation. Then, the PCC P_n,m is calculated as shown in Eq. (2). (2)${\displaystyle {\mathit{P}}_{\mathit{n},\mathit{m}}=\frac{\mathit{Cov}\left(\mathit{n},\mathit{m}\right)}{{\mathit{Q}}_{\mathit{n}}{\mathit{Q}}_{\mathit{m}}}.}$

When two variables are independent of each other, P_n,m = 0. However, P_n,m = 0 does not confirm that the two are independent. When P_n,m < 0, there is a negative correlation between the two. When P_n,m > 0, there is a positive correlation between the two.

Factor analysis is a statistical method to study the common factors of variables. Suppose there are x samples, and the sample dimension is unified to y. When x is small, large likelihood estimation is used to estimate the mean U and variance.

(3)${\displaystyle \mathit{U}=\frac{1}{\mathit{x}}\sum _{\mathit{i}=1}^{\mathit{x}}{\mathit{n}}^{\mathit{i}}}$ (4)${\displaystyle \mathit{E}=\frac{1}{\mathit{x}}\sum _{\mathit{i}=1}^{\mathit{x}}\left({\mathit{n}}^{\mathit{i}}-\mathit{U}\right).}$

In Eqs. (3)–(4), E is a covariance matrix, and E is irreversible.

LR model design

The main reason for choosing a logistic regression model is to explain or predict specific types of consumer behavior, thereby gaining a deeper understanding of the events that occur in the online shopping environment. In addition to focusing on consumer decisions to buy or not, this study emphasizes the potential correlation between these behaviors and the online shopping network’s security. It captures and quantifies the influence of various factors on consumer behavior, and subsequently deduces the impact of these behaviors on the security of online shopping networks through the logistic regression model.

The LR model represents a nonlinear regression model extensively employed for the assessment of risk factors, forecasting event occurrence probabilities, and sample classification. The applicable conditions, advantages, and disadvantages of the logistic regression model are shown in Fig. 6.

Based on the above content, this study employs a logistic regression model for a comprehensive consumer behavior analysis to better understand and predict consumer shopping decisions. In the model, after careful consideration, this study mainly focuses on consumer behavior as the dependent variable. The following outlines the process of constructing the model:

Suppose there is a series of factors that affect the result with M, N₁, N₂, …, N_s as a result. Then, the multivariate linear relationship between the dependent and independent variables is calculated, as shown in Eq. (5). (5)${\displaystyle \mathit{M}={\vartheta }_0+{\vartheta }_1{\mathit{N}}_1+{\vartheta }_2{\mathit{N}}_2+\dots +{\vartheta }_{\mathit{s}}{\mathit{N}}_{\mathit{s}}.}$

In Eq. (5), ϑ₀, ϑ₁, ϑ₂, …, ϑ_s are the parameters of the model. When it is translated into a logarithmic equation, the relationship between the dependent and independent variables will be clear. (6)${\displaystyle \mathbf{log}\mathit{M}={\vartheta }_0+{\vartheta }_1{\mathit{N}}_1+{\vartheta }_2{\mathit{N}}_2+\dots +{\vartheta }_{\mathit{s}}{\mathit{N}}_{\mathit{s}}.}$

If the probability of an event occurring is R and the probability of not occurring is 1 − R, the probability of an event A is shown in Eq. (7):

(7)${\displaystyle \mathit{T}\left(\mathit{A}\right)=\frac{\mathit{R}}{1-\mathit{R}}}$ (8)${\displaystyle \mathit{R}=\frac{\mathit{exp}\left({\vartheta }_0+{\vartheta }_1{\mathit{N}}_1+{\vartheta }_2{\mathit{N}}_2+\dots +{\vartheta }_{\mathit{s}}{\mathit{N}}_{\mathit{s}}\right)}{1+\mathit{exp}\left({\vartheta }_0+{\vartheta }_1{\mathit{N}}_1+{\vartheta }_2{\mathit{N}}_2+\dots +{\vartheta }_{\mathit{s}}{\mathit{N}}_{\mathit{s}}\right)}}$ (9)${\displaystyle 1-\mathit{R}=\frac{1}{1+\mathit{exp}\left({\vartheta }_0+{\vartheta }_1{\mathit{N}}_1+{\vartheta }_2{\mathit{N}}_2+\dots +{\vartheta }_{\mathit{s}}{\mathit{N}}_{\mathit{s}}\right)}.}$

The LR model can be expressed as shown in Eq. (10): (10)${\displaystyle \mathbf{log}\frac{\mathit{R}}{1-\mathit{R}}={\vartheta }_0+{\vartheta }_1{\mathit{N}}_1+{\vartheta }_2{\mathit{N}}_2+\dots +{\vartheta }_{\mathit{s}}{\mathit{N}}_{\mathit{s}}.}$

The feature selection aims to ensure that the selected feature set can best explain and predict network security risks while enhancing the model’s interpretability and reliability. Firstly, this study consolidates previous research to identify the key factors affecting online shopping network security, including consumer capacity, motivational, and contextual factors. In order to facilitate practical analysis, the study utilizes a Taobao user shopping behavior dataset from the Tianchi platform, encompassing extensive information related to e-commerce shopping, such as user shopping behavior, product details, transaction specifics, and more. Data is subjected to cleaning and preprocessing during the feature selection process, including handling missing data and removing outliers to ensure data quality and accuracy. Subsequently, statistical analysis is performed using the Statistical Package for the Social Sciences (SPSS) 26.0, calculating the correlations between various factors and online shopping network security. This process culminates in the selection of a feature set, identifying the most influential factors on network security. Based on the results of feature selection, a logistic regression model is established. The regression equation derived under the logistic regression model is presented in Eq. (11): (11)${\displaystyle \mathbf{ln}\frac{\mathit{R}}{1-\mathit{R}}={\vartheta }_0+{\vartheta }_1{\mathit{N}}_1+{\vartheta }_2{\mathit{N}}_2+\dots +{\vartheta }_{\mathit{s}}{\mathit{N}}_{\mathit{s}}.}$

In Eq. (11), R is the probability of online shopping network risk. N₁, N₂, …, N_S are the variables affecting the network security of online shopping. ϑ₀ is the constant term. ϑ₀, ϑ₁, ϑ₂, …, ϑ_s is the regression coefficient. The specific variable settings of the Logistic regression model based on online shopping network security risks are shown in Fig. 7.

In Fig. 7, the factors influencing online shopping network security primarily encompass consumer capability, motivational, and contextual factors. The reasons for selecting these variables are as follows: 1. Weak consumer perception capabilities may lead to consumers not fully recognizing the potential risks of online shopping. If consumers fail to perceive the risks of information leakage adequately, they might not take appropriate preventive measures, such as using strong passwords and secure payment methods, making their information more vulnerable to threats. Additionally, weak consumer perception capabilities could result in consumers overlooking network security factors in their shopping decisions, as they may prioritize price, convenience, or product features. This might lead them to choose less secure shopping channels or engage in transactions with untrustworthy sellers, thereby increasing network security risks. 2. In the pursuit of ultimate speed, consumers may not pay special attention to the choice of logistics companies. If the information systems of the logistics companies chosen by consumers have vulnerabilities, data leakage and other issues may occur, leading to network security risks. 3. The allure of low prices may prompt consumers to seek lower-priced goods and services when shopping. However, this could also lead consumers to shop on cheaper platforms with fewer security measures. In some cases, this could lead to occurrences of online fraud, such as infringement, false advertising, or unfair competition. 4. User reviews are typically related to product quality and consumer satisfaction. However, with the advancement of technology, cases of fake or abused reviews are gradually increasing, which can also affect network security in online shopping. 5. Brands are usually associated with product reputation and market position. Some well-known brands may become targets of cyberattacks, leading to the leakage of consumers’ personal information and posing network security risks. 6. The convenience of the shopping experience has a certain relationship with network security. The smoothness and simplicity of the shopping process can reduce security risks caused by shoppers’ negligence.

Standard error is an indicator used to measure the dispersion of data points from the sample mean in statistical samples. In the logistic regression model, standard error is used to assess the precision and statistical significance of regression coefficients. A smaller standard error indicates more precise estimates of the regression coefficients. The calculation of standard error is shown in Eq. (12): (12)${\displaystyle SE=\text{sqrt}\frac{\sum {\left(x-\check{x}\right)}^2}{n-p}.}$

In Eq. (12), x represents the observed dependent variable value, $\check{x}$ represents the dependent variable value predicted by the Logistic regression model, n represents the sample size, and p represents the number of independent variables in the model.

Experimental data design

The DEA model, the SOR model of consumer perception under negative evaluation, and the Logistic regression model for online shopping network information security based on consumer perception risk all use the Tianchi database. The Tianchi dataset is a scientific research data platform open to the public by Alibaba Group, primarily consisting of three categories: “Tianchi competition dataset”, “Alibaba research paper dataset”, and “Alibaba business dataset”. The Tianchi dataset mainly features ‘industry-scarce’ datasets that cover most of Alibaba’s business areas, including e-commerce, finance, healthcare, industry, transportation, and more. The e-commerce domain contains a dataset of Taobao user shopping behavior, including data from approximately one million user behaviors and a dataset of approximately 30,000 negative comments on Taobao e-commerce. The data selected for this research is from 2017 to 2021 and includes publicly available user behavior and negative comment databases on the Taobao platform, with no involvement of human experiments or participants. It does not collect any data that can identify personal identities and is not associated with any potential limitations or biases related to data sources.

The models used in this study primarily rely on Python 3.6.13 and SPSS 26.0 software. The operating system is Windows 10, 64-bit, with 16GB of installed memory and an Intel i7 processor. Python is a popular high-level programming language known for its simple and readable syntax, cross-platform support, and versatility. It is widely used in web development, data analysis, scientific computing, artificial intelligence, and automation tasks. Python 3.6.13 is a specific version within the Python 3.x series, introducing significant language and library updates. SPSS is a powerful statistical analysis software widely used in social science research, business analysis, and data mining. SPSS provides a wide range of statistical analysis tools, including descriptive statistics, inferential statistics, regression analysis, factor analysis, cluster analysis, and multidimensional data analysis, enabling users to easily process and analyze various data types. It features a user-friendly interface that allows users to perform data visualization and report generation, making it accessible to non-statisticians. SPSS offers powerful tools for data analysis and decision support, serving as an invaluable tool for researchers and analysts.

This study conducted data preprocessing using Python 3.6.13. Initially, relevant data processing and analysis libraries were imported in Python, and the original data was loaded using the panda’s library. Subsequently, data cleansing procedures were applied, which included removing duplicate records using deduplication functions, converting date data into the proper date format using time series transformation functions, rectifying format errors, filling missing data using the fillna function, and addressing missing values through the use of the value function. Following this, outlier detection was performed using Z-scores, and outliers were removed based on Z-score criteria employing boolean indexing. Subsequently, feature selection was carried out in alignment with the research requirements, and data dimensionality was reduced using principal component analysis (PCA) while ensuring the preservation of user privacy through anonymization. Finally, the dataset was partitioned into training and testing sets, with the training dataset comprising 1,052 data points and the testing dataset comprising 263 data points.

SPSS 26.0 was used for statistical analysis of the data in this study. The data was first imported into SPSS, and the variable types, data distribution, and potential outliers were checked. Descriptive statistical analysis was then performed to generate a statistical summary of the dataset, including mean, median, and standard deviation, to understand the data’s characteristics. Finally, a logistic regression model was built using SPSS. The network security analysis of online shopping consumer perception using the logistic regression model and the future research roadmap for network security services are presented in Fig. 8.

Analysis of online shopping consumer perception and network security results under negative reviews

Analysis of SOR model results based on negative reviews

In this study, the SOR model, specifically tailored to negative reviews, is employed to examine the impact of unfavorable feedback on consumer perception and behavior. The primary research variables of interest are consumer perception and behavior within the context of online shopping. The comprehensive analysis of online shopping consumer perception is conducted through an exploration of negative reviews on the Taobao platform and the associated product’s consumer behavior data spanning the period from 2017 to 2021. Figure 9 reveals the results of the SOR model analysis based on negative reviews.

In Fig. 9, the findings illustrate a discernible correlation between negative reviews and three distinct consumer behaviors. Notably, the quantity of negative reviews significantly influences the tendency to delay consumption, as indicated by a correlation coefficient of 0.41. Meanwhile, the intensity of negative reviews exerts a pronounced impact on the inclination to reject consumption, with a correlation coefficient of 0.38. Furthermore, the length of negative reviews significantly affects the proclivity for opposing consumption, displaying a correlation coefficient of 0.38. In sum, it is evident that the length, intensity, and quantity of negative reviews collectively manifest a detrimental effect on consumption, thereby diminishing consumer purchase intent. The correlation analysis results of PR reveal associations between PR, consumer behavior, and negative reviews. Specifically, PR exhibits a strong correlation between delayed shopping behavior and the number of negative reviews, featuring correlation coefficients of 0.41 and 0.4, respectively. Conversely, its impact on opposing consumption behavior is relatively modest, with a correlation coefficient of only 0.27. Overall, it is evident that negative reviews contribute positively to the perception of consumer risk. These findings underscore the presence of a causal relationship between PR and negative consumer behavior.

Results analysis of LR model for NS

This study has devised a LR model with the primary aim of investigating the network security aspects of online shopping. It centers its focus on network security as the principal research variable while considering consumers’ individual characteristics, aptitude factors, motivational elements, and conditional factors as independent variables. The in-depth analysis of online shopping NS, contingent upon consumer perception risk, is conducted by scrutinizing negative reviews on Taobao platforms and the associated product consumer behavior data spanning the years 2017 to 2021. Figure 10 plots the analysis results of the online shopping NS-LR model based on consumer perception risk.

In Fig. 10, the convenience of consumer purchases has a relatively minor impact on online shopping security. The relationship between the two is not particularly significant. Enhancing or reducing the convenience of consumer purchases does not yield consequential effects on online shopping security. However, delivery speed exhibits a notable association with online shopping security, which may be attributed to the considerable proportion of food items in online shopping. The regression coefficient for buyer reviews stands at 0.12, with a standard error of 0.135 and an approximate p-value of 0.05. This data indicates a significant relationship between buyer reviews and online shopping security. Enhancing the authenticity of reviews may promote online shopping security and reduce associated risks. Brand and price demonstrate a significant relationship with online shopping security, but both variables exhibit regression coefficients less than zero. The data suggests that these factors do not completely guarantee online shopping security but decrease the probability of online shopping risks. There is a substantial relationship between consumer perception and online shopping security. Strengthening consumer perception can enhance the capacity to process risk-related information, subsequently bolstering resistance to online shopping risks and elevating online shopping security. Based on the results of statistical significance, it is evident that delivery speed, reviews, brand, price, and consumer perception all exert a significant influence on online shopping security, with consumer perception having the most substantial impact. This underscores the importance of enhancing consumer perception to effectively heighten online shopping security. However, the impact of purchase convenience on online shopping security lacks statistical significance. This discovery offers valuable insights into the factors influencing online shopping security and contributes to the ongoing efforts to enhance and maintain online shopping safety. In summary, online shopping security is influenced by a multitude of factors, and among them, consumer perception holds considerable sway compared to other variables.

Correlation analysis

In order to further explore the relationships between various factors influencing online shopping security, the PCC is applied to conduct a correlation analysis between brand, price, delivery speed, and online shopping security. This coefficient is employed to measure the strength and direction of linear relationships between variables. Table 1 presents the experimental results.

Table 1 indicates that there is a positive correlation between brand and online shopping security, with a correlation coefficient of 0.25. This suggests that an enhancement in brand reputation may positively impact online shopping security. There is a negative correlation between price and online shopping security, with a correlation coefficient of −0.18, indicating that a price reduction may decrease the potential risks associated with online shopping. Delivery speed shows a positive correlation with online shopping security, with a correlation coefficient of 0.30, implying that faster delivery speeds may contribute to an improvement in the security of online shopping.