Accumulated local total effect graph-based passenger loyalty influence mechanism analysis method

Abstract

The invention provides a passenger loyalty influence mechanism analysis method based on a cumulative local total effect graph. The method comprises the following steps: collecting multi-dimensional public transport service element quality and passenger loyalty data; verifying the reliability and validity of the data, and researching the correlation between the quality attributes of the public transport service elements; preprocessing the data, constructing a causal analysis model between the public transport service element quality and the passenger loyalty, and training the model by using the data; an accumulated local total effect analysis method is used to deeply discuss the influence of the service quality of each dimension element in the public transport service on the loyalty of passengers. The method is applied to deep analysis of the passenger loyalty influence mechanism, can be independent of the used influence mechanism analysis model, and explores and objectively reveals the essential relationship between the perception service quality of public transport service elements and the passenger loyalty; and a firmer scientific basis is provided for formulating a public transport service quality optimization strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method for analyzing a passenger loyalty influence mechanism based on a cumulative local total effect diagram. Background Art

[0002] The evaluation of urban public transportation service quality involves multiple key dimensions, such as line and site layout, vehicle operation status, riding environment, service attitude and fare policy. The service quality of these elements directly determines the travel experience and satisfaction of passengers, and profoundly affects the loyalty of passengers to public transportation. High loyalty not only reflects the high recognition and continuous choice of public transportation services by passengers, but also builds a stable and reliable passenger flow foundation for the public transportation system, providing a strong guarantee for its long-term operation and development.

[0003] It is worth noting that the relationship between the quality of public transportation service elements and passenger loyalty is not a simple linear correspondence, but presents complex nonlinear characteristics. That is, a slight change in the quality of a service element may cause a significant fluctuation in passenger loyalty. Therefore, it is of great theoretical and practical significance to deeply explore how the service quality of various dimensional elements in urban public transportation services affects passengers' public transportation loyalty. On the theoretical level, this study helps to reveal the inherent mechanism between public transportation service quality and passenger loyalty, and further enrich and improve the theoretical system of public transportation service management; on the practical level, this study can provide scientific decision-making basis for public transportation service providers, guide them to optimize service processes and improve service quality, thereby effectively enhancing passengers' public transportation loyalty and promoting the sustainable development of the public transportation system.

[0004] When exploring the nonlinear relationship between the quality of public transportation service elements and passenger loyalty, machine learning methods such as gradient boosting decision trees and random forests play a key role. These methods can not only handle complex nonlinear data relationships, but also reveal the degree of influence of different service elements on passenger loyalty through their internal mechanisms. At the same time, model-independent interpretation methods such as partial dependency plots and accumulated local effects plots are also widely used to visualize and analyze these relationships, providing intuitive decision support for public transportation service providers.

[0005] As an intuitive visualization tool, partial dependence plots can show how one or two features affect the prediction results of machine learning models. Its advantages are that it is intuitive and easy to understand, applicable to a variety of models, and can reveal the linear and nonlinear relationship between features and prediction results. However, when there is a strong correlation between the quality characteristics of public transportation service elements, partial dependence plots may not accurately reflect the independent impact of these elements on passenger loyalty because it assumes that the features for which partial dependence is calculated are uncorrelated with other features, which is often not true in practical applications.

[0006] In contrast, the cumulative local effects plot is a visualization tool that can quantify the average impact of explanatory variables on the model response variable. Its advantage is that it can solve the correlation problem between features while quantifying the impact, and it is also a model-independent interpretation method. However, the disadvantage of the cumulative local effects plot is that it does not directly consider the impact of the value distribution of the explanatory variable on the result, which may lead to ignoring data sparse areas, failing to accurately reflect the global effect, making it difficult to explain the impact of extreme values, and possibly giving misleading conclusions, which also limits its integrity as a model interpretation tool. Summary of the invention

[0007] According to the technical problems raised above, a method for analyzing the mechanism of passenger loyalty based on cumulative local total effect diagram is provided. Based on the advantage of cumulative local effect diagram in processing strongly correlated variables, the present invention further innovatively incorporates the sample score distribution consideration of each explanatory variable, thereby being able to more accurately and comprehensively depict the actual relationship between the quality of public transportation service elements and passenger loyalty.

[0008] The technical means adopted by the present invention are as follows:

[0009] A passenger loyalty influence mechanism analysis method based on cumulative local total effect diagram, comprising:

[0010] S1. Collect multi-dimensional data on the quality of public transport service elements and passenger loyalty;

[0011] S2, test the reliability and validity of the data in S1, and study the correlation between the quality attributes of public transport service elements;

[0012] S3. Preprocess the data, calculate the passenger loyalty score based on the passenger loyalty measurement model, build a causal analysis model between the quality of public transportation service elements and passenger loyalty, capture the relationship between the quality of service elements and passenger loyalty, and use the data to train the model;

[0013] S4. Use the cumulative local total effect to analyze the impact of the service quality of each dimensional element in public transportation services on passenger loyalty.

[0014] Furthermore, the step S1 specifically includes:

[0015] The passenger loyalty refers to the passenger's continued willingness to choose public transportation in the future based on their previous travel experience. Passenger loyalty is quantified using explicit variables, including: will continue to use public transportation in the future, public transportation is still the preferred mode of travel, willing to recommend public transportation services to others, and public transportation is still the preferred mode of travel even if they have the ability to buy a car;

[0016] The quality of public transportation service elements includes two aspects: the service quality perceived by passengers in terms of transportation-related technical attributes and the service quality perceived by passengers in terms of comfort and safety. The service quality perceived by passengers in terms of transportation-related technical attributes refers to the quality of service elements perceived by passengers to meet their basic needs, including: public transportation service time, transfer convenience, walking distance for passengers to and from stations, and waiting time for passengers at stations; the service quality perceived by passengers in terms of comfort and safety refers to the service quality related to transportation comfort and safety perceived by passengers, including: station facility layout and waiting environment, station sanitary conditions, operating smoothness, temperature inside the vehicle, seat comfort and safety.

[0017] Furthermore, the reliability is used to test the internal consistency of the data. The Cronbach coefficient is used to test the reliability of the passenger loyalty data in S1. When the calculated Cronbach coefficient is greater than 0.7, it indicates that the passenger loyalty data is reliable.

[0018] The validity is used to detect whether the manifest variables under each structure can reflect the content represented by passenger loyalty. During the verification process, the KMO statistic of the manifest variables, the significance probability of the Bartlett sphericity test, the average variance extracted and the combined reliability value are calculated respectively. When the calculation results are all within the preset range and the factor loading is greater than 0.5, it indicates that the passenger loyalty data is valid.

[0019] Furthermore, the correlation analysis is used to measure the degree of correlation between two different variables in the quality of public transportation service elements. When the correlation coefficient between the variables is greater than 0.7, it indicates that there is a strong correlation between the variables; when the correlation coefficient is between 0.4 and 0.7, it indicates that the correlation between the variables is strong; and when it is less than 0.4, it indicates that the correlation is weak;

[0020] The positive correlation between variables in the quality of public transportation service elements is verified through the calculation results of correlation.

[0021] Furthermore, the step S3 specifically includes:

[0022] The preprocessing is used to clean, organize and process the collected data;

[0023] The passenger loyalty measurement model constructs a measurement model of passenger loyalty and corresponding explicit variables, and calculates a passenger loyalty score based on a path coefficient;

[0024] The causal analytical model between the quality of public transportation service elements and passenger loyalty takes the quality of public transportation service elements as input and the passenger loyalty score as output. During the model training process, the grid search method and cross-validation method are used to tune the parameters, and the mean absolute percentage error is used to evaluate the prediction results of the model.

[0025] Furthermore, step S4 specifically includes:

[0026] S41. Calculate the average local effect of the variables in the quality of public transport service elements on passenger loyalty in the kth interval:

[0027] k=1,2,3,4,…K, where x s represents the quality variable of the currently selected partially dependent public transport service element; x c Represents division by x s The vector of other explanatory variables is composed of ; the value range of the public transportation service element quality variable is divided into a series of intervals, K represents the number of intervals; Z k+1 It represents the lower bound value of the k+1th interval and the upper bound value of the kth interval; represents the set of data samples in the kth interval, Indicates the number of samples in the data sample set; i represents The i-th data sample in Representative Set Other explanatory variables x for the i-th data sample in c Scoring situation; Representative Set The partial dependent variable x of the i-th data sample in s The alternative score is is a differentiable function used to represent a machine learning model. The input of the machine learning model is the service quality of each dimensional element in public transportation services, and the output is the loyalty of passengers. X represents the vector composed of the service quality of each dimensional element in public transportation services. Represents x s The average local impact on passenger loyalty in the kth interval;

[0028] S42, calculate x sThe cumulative local total effect on the response variable is to calculate the cumulative local effect of each data sample, accumulate it in intervals, and form the cumulative local total effect:

[0029]

[0030] S43, normalize the cumulative local effect in S42 and display the x in each interval s The proportion of the cumulative local total effect on the response variable:

[0031]

[0032] in, Represents partial dependence on variable x S The unstandardized cumulative local total effect on the response variable, f xs,k,TALE (x) represents the value of m xs,k (x) Cumulative percentage curve of the results after normalization.

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] The present invention provides a passenger loyalty influencing mechanism analysis method based on the cumulative local total effect diagram. In the research field of exploring the passenger loyalty influencing mechanism, the academic community has widely adopted a variety of analysis methods, which include traditional linear regression techniques, such as multiple linear regression, path analysis and structural equation model, as well as nonlinear methods, such as three-factor theory, gradient boosting decision tree and random forest. In order to objectively and accurately reveal the relationship between the explanatory variables and the response variables, an analysis method that does not depend on a specific model is urgently needed. The cumulative local total effect diagram (TALE) proposed in the present invention is a model-independent interpretation method. It does not depend on a specific model type. It only needs to derive a differentiable function that meets the conditions from the model to accurately and efficiently reveal the intrinsic connection between the variables. It is suitable for analyzing the relationship between the explanatory variables and the response variables in various models. In addition, compared with methods such as partial dependence diagram (PDP), the cumulative local total effect diagram (TALE) inherits the advantages of the cumulative local effect diagram (ALE), which can effectively solve the correlation problem between variables, thereby more accurately describing the independent influence between variables.

[0035] The present invention provides a passenger loyalty influencing mechanism analysis method based on a cumulative local total effect diagram. The cumulative local total effect diagram (TALE) used in the present invention shows significant advantages when processing data samples with uneven score distribution, and can accurately reveal the true relationship between the explanatory variable and the response variable. In the study of the passenger loyalty influencing mechanism, the cumulative local total effect diagram (TALE) adopts standardized processing in data processing to reflect the proportion of the cumulative local total effect of the explanatory variable on the response variable in each interval. In contrast, the cumulative local effect diagram (ALE) calculates the cumulative local effect by centering, which is achieved by subtracting the mean of each eigenvalue from its mean, so that the mean of the processed data is zero. However, this centralization process weakens the scale information and the influence of extreme values ​​in the original data, which may be an obstacle to a comprehensive understanding of the influencing mechanism of passenger loyalty. In fact, the sample score distribution of passenger loyalty has a great influence on the formulation of the final transportation policy. In contrast, the TALE method using standardized processing shows unique advantages. Standardization not only maintains the relative scale information of the data between different features, but also allows the existence of extreme values, so that the complex dynamics of passenger loyalty can be more accurately reflected. Therefore, in the study of the influencing mechanism of passenger loyalty, the standardized TALE method can better retain and analyze the subtle differences and extreme cases in the data, more comprehensively consider the diversity and differences of passenger behavior, and has strong applicability and explanatory power.

[0036] The passenger loyalty influencing mechanism analysis method based on the cumulative local total effect diagram provided by the present invention reveals how the service quality of each dimensional element in urban public transportation services affects the public transportation loyalty of passengers in a nonlinear manner from a more objective and in-depth perspective. This method not only retains the advantages of the traditional cumulative local effect diagram in dealing with strongly correlated variables, but also innovatively incorporates the sample score distribution considerations of each explanatory variable. This innovation solves a series of problems caused by the traditional method that does not directly consider the impact of the characteristic value distribution of the explanatory variable on the results, including the possibility of ignoring data sparse areas, failing to accurately reflect the global effect, difficulty in explaining the impact of extreme values, and possible misleading conclusions. This not only improves the integrity of the model as an explanatory tool, but also can more accurately and comprehensively depict the actual causal relationship between the explanatory variables and the response variables. By applying this method to the in-depth analysis of the passenger loyalty influencing mechanism, the present invention can not only explore and objectively reveal the essential relationship between the perceived service quality of each public transportation service element and passenger loyalty independently of the influencing mechanism analysis model used, but also provide a more solid scientific basis for the formulation of public transportation service quality optimization strategies.

[0037] Based on the above reasons, the present invention can be widely promoted in fields such as data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0039] Figure 1 It is a flow chart of the method for analyzing the passenger loyalty influence mechanism based on the cumulative local total effect diagram in the present invention.

[0040] Figure 2 This is a cumulative local total effect diagram of the passenger loyalty explanatory variables based on the GBDT model using the TALE method in an embodiment of the present invention.

[0041] Figure 3 It is a cumulative local total effect diagram of the passenger loyalty explanatory variables based on the GBDT model using the ALE method in an embodiment of the present invention.

[0042] Figure 4 Graph showing the relationship between passenger loyalty and its explicit variable in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0044] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0045] like Figure 1As shown, the present invention provides a method for analyzing the passenger loyalty influence mechanism based on a cumulative local total effect diagram, comprising:

[0046] S1. Collect multi-dimensional data on the quality of public transport service elements and passenger loyalty;

[0047] In specific implementation, as a preferred embodiment of the present invention, the step S1 specifically includes:

[0048] The passenger loyalty refers to the passenger's continued willingness to choose public transportation in the future based on their previous travel experience. Passenger loyalty is quantified using explicit variables, including: will continue to use public transportation in the future, public transportation is still the preferred mode of travel, willing to recommend public transportation services to others, and public transportation is still the preferred mode of travel even if they have the ability to buy a car;

[0049] The quality of public transportation service elements includes two aspects: the service quality perceived by passengers in terms of transportation-related technical attributes and the service quality perceived by passengers in terms of comfort and safety. The service quality perceived by passengers in terms of transportation-related technical attributes refers to the quality of service elements perceived by passengers to meet their basic needs, including: public transportation service time, transfer convenience, walking distance for passengers to and from stations, and waiting time for passengers at stations; the service quality perceived by passengers in terms of comfort and safety refers to the service quality related to transportation comfort and safety perceived by passengers, including: station facility layout and waiting environment, station sanitary conditions, operating smoothness, temperature inside the vehicle, seat comfort and safety.

[0050] The quality of public transportation service elements in multiple dimensions is taken as the key explanatory variables, and passenger loyalty is taken as the response variable, and a mapping relationship between the two is established.

[0051] Table 1 Passenger loyalty impact variables and corresponding explicit variable settings

[0052]

[0053] Based on the research hypothesis, the present invention considers the perceived service quality of passenger transportation-related technical attributes and the perceived service quality of passenger comfort and safety, and carefully selects 11 public transportation service elements as explanatory variables (i.e., input variables) of the GBDT model. Among them, the perceived service quality of passenger transportation-related technical attributes mainly refers to the quality of service elements perceived by passengers to meet their basic needs, which is specifically measured from five aspects (TRA1, TRA2, TRA3, TRA4, TRA5), covering key elements such as public transportation service time, transfer convenience, walking distance of passengers to and from public transportation stations, and waiting time of passengers at public transportation stations. The perceived service quality of passenger comfort and safety refers to the service quality related to transportation comfort and safety perceived by passengers, which aims to measure whether the equipment and surrounding environment related to transportation truly meet the actual needs of passengers for comfort and safety. It is specifically measured from six aspects (CS1, CS2, CS3, CS4, CS5, CS6), and the detailed indicators are shown in Table 2. In order to collect data on these 11 attributes, the present invention adopts a questionnaire survey method and strictly ensures the authenticity and representativeness of the data. In the questionnaire design, the Likert five-point scale method was adopted, setting a scoring range of 1 to 5 points (where 1 represents very dissatisfied and 5 represents very satisfied) to facilitate subsequent data analysis and model construction.

[0054] Table 2 Explanatory variables for public transport service quality

[0055]

[0056]

[0057] During the implementation, an on-site questionnaire survey was conducted in Xiamen, Fujian Province, China. To this end, a special questionnaire was designed to directly collect citizens' feedback on the quality of public transportation service elements in various dimensions, and quantitatively evaluate them using the Likert five-level scale, so as to accurately identify the advantages and disadvantages of public transportation service elements in various dimensions, aiming to provide a scientific basis for public transportation companies and related departments to formulate more accurate and effective improvement measures. In order to ensure the representativeness of the sample and the reliability of the data, random sampling was carried out in areas with dense traffic and high frequency of public transportation use in the central business district of Xiamen. A total of 900 questionnaires were collected by distributing questionnaires on site, and abnormal data, missing data and duplicate data were processed and cleaned. Finally, 625 valid samples were obtained, ensuring the adequacy of the sample size and the accuracy and validity of the data. It is worth mentioning that this survey was conducted before the Xiamen Metro system began to operate, which enabled the research results to focus on the quality evaluation of traditional public transportation services and effectively avoid the potential interference of emerging transportation modes such as subways on the evaluation of public transportation services.

[0058] S2, test the reliability and validity of the data in S1, and study the correlation between the quality attributes of public transport service elements;

[0059] In specific implementation, as a preferred embodiment of the present invention, the reliability is used to test the internal consistency of the data, and the Cronbach's coefficient is used to test the reliability of the passenger loyalty data in S1. When the calculated Cronbach's coefficient is greater than 0.7, it indicates that the passenger loyalty data is reliable. Table 3 shows the results of the passenger loyalty reliability analysis, and the results show that all explicit variables can well reflect the content of the latent variables, proving that the survey data has good reliability.

[0060] Table 3 Reliability analysis results

[0061] Latent variables Explicit variables Number of items Cronbach's alpha LOY <![CDATA[LOY1,LOY2,LOY3,LOY4]]> 4 0.771

[0062] The validity is used to detect whether the manifest variables under each structure can reflect the content represented by passenger loyalty. During the verification process, the KMO statistic of the manifest variables, the significance probability of the Bartlett sphericity test, the average variance extracted and the combined reliability value are calculated respectively. When the calculation results are all within the preset range and the factor loading is greater than 0.5, it indicates that the passenger loyalty data is valid.

[0063] When implemented, the KMO statistic value is greater than 0.7, the significance probability value of the Bartlett sphericity test is less than 0.01, the average variance extraction is greater than 50%, and the combined reliability value is greater than 0.7. In addition, the factor loadings of all explicit variables are greater than 0.5. Therefore, it can be considered that the survey data used in the present invention can effectively measure the public transportation loyalty of passengers and calculate its score.

[0064] Table 4 Validity analysis results

[0065]

[0066] S3. Preprocess the data, calculate the passenger loyalty score based on the passenger loyalty measurement model, build a causal analysis model between the quality of public transportation service elements and passenger loyalty, capture the relationship between the quality of service elements and passenger loyalty, and use the data to train the model; in implementation, take the gradient boosting decision tree model GBDT as an example;

[0067] In specific implementation, as a preferred embodiment of the present invention, the correlation analysis is used to measure the degree of correlation between two different variables in the quality of public transportation service elements. When the correlation coefficient between the variables is greater than 0.7, it indicates that there is a strong correlation between the variables; when the correlation coefficient is between 0.4 and 0.7, it indicates that the correlation between the variables is strong; and when it is less than 0.4, it indicates that the correlation is weak;

[0068] Pearson correlation analysis was used to analyze the correlation between service quality attributes (as shown in Table 5). The results showed that there was a strong positive correlation between some service factor variables. Therefore, there was no way to use partial dependence diagrams to analyze the impact of multidimensional explanatory variables on response variables.

[0069] Table 5 Results of correlation analysis of explicit variables

[0070] <![CDATA[TRA1]]> <![CDATA[TRA2]]> <![CDATA[TRA3]]> <![CDATA[TRA4]]> <![CDATA[TRA5]]> <![CDATA[CS1]]> <![CDATA[CS2]]> <![CDATA[CS3]]> <![CDATA[CS4]]> <![CDATA[CS5]]> <![CDATA[CS6]]> <![CDATA[TRA1]]> 1.000 0.423 0.349 0.485 0.398 0.323 0.274 0.234 0.177 0.239 0.270 <![CDATA[TRA2]]> 0.423 1.000 0.520 0.484 0.510 0.407 0.407 0.370 0.296 0.391 0.385 <![CDATA[TRA3]]> 0.349 0.520 1.000 0.407 0.346 0.283 0.263 0.277 0.289 0.304 0.322 <![CDATA[TRA4]]> 0.485 0.484 0.407 1.000 0.618 0.424 0.358 0.342 0.343 0.301 0.310 <![CDATA[TRA5]]> 0.398 0.510 0.346 0.618 1.000 0.458 0.424 0.366 0.336 0.333 0.341 <![CDATA[CS1]]> 0.323 0.407 0.283 0.424 0.458 1.000 0.612 0.395 0.421 0.488 0.372 <![CDATA[CS2]]> 0.274 0.407 0.263 0.358 0.424 0.612 1.000 0.467 0.471 0.526 0.428 <![CDATA[CS3]]> 0.234 0.370 0.277 0.342 0.366 0.395 0.467 1.000 0.269 0.428 0.433 <![CDATA[CS4]]> 0.177 0.296 0.289 0.343 0.336 0.421 0.471 0.269 1.000 0.464 0.359 <![CDATA[CS5]]> 0.239 0.391 0.304 0.301 0.333 0.488 0.526 0.428 0.464 1.000 0.463 <![CDATA[CS6]]> 0.270 0.385 0.322 0.310 0.341 0.372 0.428 0.433 0.359 0.463 1.000

[0071] The positive correlation between variables in the quality of public transportation service elements is verified through the calculation results of correlation.

[0072] S4. Use the cumulative local total effect to analyze the impact of the service quality of each dimensional element in public transportation services on passenger loyalty.

[0073] In specific implementation, as a preferred embodiment of the present invention, step S3 specifically includes:

[0074] In the implementation, the GBDT model is used as a nonlinear model, aiming to deeply explore the essential relationship between the quality of public transportation service elements in various dimensions and the public transportation loyalty of passengers. The construction process of the GBDT model mainly includes four key links: data preprocessing, feature construction, model training, model optimization and evaluation.

[0075] The preprocessing is used to clean, organize and process the collected data;

[0076] The passenger loyalty measurement model uses Amos software to build a measurement model of passenger loyalty and its corresponding explicit variables, and calculates the passenger loyalty score based on the path coefficient;

[0077] The GBDT model takes the quality of public transportation service elements as input and the passenger loyalty score as output. During the model training process, the grid search method and cross-validation method are used to tune the parameters, and the mean absolute percentage error is used to evaluate the model's prediction results.

[0078] During the model training stage, the MAPE value of the GBDT model fitting was 12.2%, indicating that the fitting results were good and within an acceptable range.

[0079] In specific implementation, as a preferred embodiment of the present invention, step S4 specifically includes:

[0080] S41. Calculate the average local effect of the variables in the quality of public transport service elements on passenger loyalty in the kth interval:

[0081] k=1,2,3,4,…K, where x s represents the quality variable of the currently selected partially dependent public transport service element; x c Represents division by x s The vector of other explanatory variables is composed of ; the value range of the public transportation service element quality variable is divided into a series of intervals, K represents the number of intervals; Z k+1 It represents the lower bound value of the k+1th interval and the upper bound value of the kth interval; represents the set of data samples in the kth interval, Indicates the number of samples in the data sample set; i represents The i-th data sample in Representative Set Other explanatory variables x for the i-th data sample in c Scoring situation; Representative Set The partial dependent variable x of the i-th data sample in s The alternative score is is a differentiable function used to represent a machine learning model. The input of the machine learning model is the service quality of each dimensional element in public transportation services, and the output is the loyalty of passengers. X represents the vector composed of the service quality of each dimensional element in public transportation services. Represents x s The average local impact on passenger loyalty in the kth interval;

[0082] S42, calculate x s The cumulative local total effect on the response variable is to calculate the cumulative local effect of each data sample, accumulate it in intervals, and form the cumulative local total effect:

[0083]

[0084] S43, normalize the cumulative local effect in S42 and display the x in each interval s The proportion of the cumulative local total effect on the response variable:

[0085]

[0086] in, Represents partial dependence on variable x S The unstandardized cumulative local total effect on the response variable, Express Cumulative percentage curves of normalized results were performed.

[0087] Example 1

[0088] like Figure 1 As shown, the present invention provides a method for analyzing the mechanism of passenger loyalty influence based on the cumulative local total effect diagram. This embodiment uses the proposed cumulative local total effect diagram (TALE) to explore the causal relationship between the explanatory variable and passenger loyalty. The explanatory variable, that is, the service quality attribute perceived by the passenger, is measured by a 5-point Likert scale, so the score of the variable is divided into four intervals (k = 1, 2, 3, 4). On this basis, the standardized TALE of each explanatory variable (service element quality) to the response variable (passenger loyalty) can be obtained by calculation. Figure 2 The nonlinear relationship between the explanatory variables and the response variables is shown. This nonlinear relationship is not only due to the nonlinear growth of the accumulated local effects in each interval in the gradient boosting decision tree (GBDT) model, but also because the uneven distribution of sample scores also leads to nonlinear results. Specifically, the slope of the curve of most variables in the second interval has a significant increase, which indicates that the sample distribution also has a significant impact on the results of the GBDT model. Further observation shows that the accumulated standardized TALE percentage curves of most variables based on the GBDT model show obvious inconsistency in the third and fourth intervals. For example, the slopes of variables TRA3, TRA5, CS5 and CS6 increase significantly in the third and fourth intervals, while the slopes of variables TRA4, CS1, CS3 and CS4 decrease. This result further proves that in the GBDT model, in addition to the influence of the actual score distribution of the sample, the accumulated local effects in each interval also play an important role in the results of passenger loyalty, resulting in the standardized TALE of service attributes showing a nonlinear trend of change. At the same time, the cumulative percentage of standardized TALE in the third interval shows that TRA4, CS1, CS3, and CS4 can contribute more than 60% of the loyalty percentage, and improving the service quality of these attributes can further improve passenger loyalty. However, the passenger loyalty contributed by TRA5 is less than 40%, and improving this service attribute has a lower impact on passenger loyalty.

[0089] In order to demonstrate the superiority of the proposed cumulative local total effect diagram (TALE) over the traditional cumulative local effect diagram (ALE), we also apply the ALE method to the GBDT model for performance evaluation while keeping the explanatory variables and the response variables completely consistent. Figure 3As shown in Figure 2, the figure shows the cumulative local effects of the explanatory variables of passenger loyalty, revealing that there is a complex nonlinear relationship between the explanatory variables and the response variable in the GBDT model. However, since the ALE method evaluates the relationship between variables by calculating the average of the predicted changes of samples in the grid, when the samples are unevenly distributed in the interval, the results of ALE may not be able to fully capture the impact of the sample distribution on the response variable, resulting in a relatively small degree of change in the curve. This makes it challenging to discern the trend of ALE in the GBDT model under the same circumstances. In contrast, when the sample score distribution is taken into account, TALE is more stable in reflecting the impact of service attributes on passenger loyalty.

[0090] like Figure 2 As shown in the figure, the slope of almost every service factor variable in the second interval (between 2 and 3 points) has changed significantly, showing a clear growth trend. This actually reflects that the number of samples in the first and second intervals is small, and the scores of most samples are concentrated in the interval between "3 points" and "5 points". At the same time, the percentage value of TALE in the third interval (3 points to 4 points) accurately quantifies the specific contribution percentage of the improvement in loyalty when the passenger changes from negative (very dissatisfied to average) to positive (satisfied), which is crucial to understanding how service quality improvements are transformed into growth in passenger loyalty. TALE can keenly identify and quantify such changes, providing researchers with in-depth insights into how service attributes affect passenger satisfaction. In contrast, the cumulative local effects map (ALE) evaluates the relationship between variables by calculating the average of the predicted changes of samples in the grid. This method may perform well when dealing with uniformly distributed samples, but when faced with uneven sample distribution, it ignores the sparse data area, cannot accurately reflect the global effect, and has difficulty explaining the impact of extreme values, such as Figure 3 As shown in the figure. The explanatory variable curve of passenger loyalty has a small degree of change in each interval, almost presenting a smooth straight line, which ignores the significant differences that may exist in the actual sample distribution, making it difficult to truly reflect the degree and direction of the influence of the explanatory variable on the response variable. Therefore, in order to more accurately analyze the impact mechanism and take into account the actual situation of uneven distribution of sample scores, it is more appropriate to use the TALE method. Through the TALE method, researchers can go deep into each sample interval, analyze the relationship between variables in different intervals, and then draw more comprehensive and accurate conclusions, which is of great significance to the formulation of relevant transportation policies and the optimization of public transportation company management systems and service levels.

[0091] Figure 4Describes how the response variable is constructed around passenger loyalty in the gradient boosted decision tree (GBDT) model. The quantification of this loyalty is not given directly, but relies on the deep and complex relationship between the four explicit variables (LOY1, LOY2, LOY3, LOY4) and passenger loyalty. These four explicit variables represent the specific manifestations of passenger loyalty in different aspects, and the strength of their association with passenger loyalty is measured by the path coefficient. In order to calculate the passenger loyalty score, we use the measurement model equation to calculate it.

[0092] Example 2

[0093] This example analyzes the application and results of TALE in GBDT and compares them with the ALE results.

[0094] 1. Application of TALE in GBDT and results analysis

[0095] By applying the GBDT model, the impact of each explanatory variable on the response variable was deeply analyzed, and the cumulative standardized TALE was calculated to quantify this impact. Figure 2 As shown in Figure 2, the results show that there is a significant nonlinear relationship between the explanatory variables and the response variable. This finding provides a new perspective for understanding the complex mechanism of the service attributes on passenger loyalty. There are two factors for the nonlinear relationship between the explanatory variables and the response variable. First, in each prediction interval, the GBDT model shows a nonlinear growth trend through the cumulative local effect. This nonlinear growth not only reflects the complex learning mechanism within the model, but also highlights the differences in the local effects of different service attributes on passenger loyalty in different value ranges. Secondly, the score distribution of the sample data also significantly affects the results of TALE. In particular, in some intervals, such as the second interval, the difference in the distribution of samples with a score of "2" and a score of "3" leads to a significant change in the slope of the curve, further strengthening the nonlinear characteristics of the results. At the same time, according to the cumulative percentage value of the standardized TALE in the third interval, the service attribute that contributes most to passenger loyalty is clearly identified.

[0096] In summary, the application of TALE in the GBDT model not only reveals the nonlinear relationship between service attributes and passenger loyalty, but also provides a scientific basis for formulating targeted loyalty improvement strategies. By accurately identifying high-contribution attributes and optimizing their service quality, passenger loyalty can be more effectively improved.

[0097] 2. Comparative analysis of ALE and TALE results of GBDT model

[0098] In order to highlight the superiority of TALE, a detailed comparison is made between it and the traditional ALE method. The specific practice is to apply the ALE method to the results of the GBDT model and observe the results (such as Figure 3 shown). Figure 3 It is clearly shown that in the GBDT model, although the ALE value of each explanatory variable shows nonlinear changes, the degree of change is relatively small. This is because ALE calculates the average value of the predicted changes of samples in the grid. When the samples are unevenly distributed in the interval, the results of ALE may not fully capture the impact of the sample distribution on the response variable. However, when comprehensively and truly evaluating the impact of explanatory variables on passenger loyalty, the sample distribution that reflects the passenger's current satisfaction with service quality is particularly important. Therefore, in comparison, when considering the sample score distribution, the standardized TALE shows higher stability in reflecting the impact of service attributes on passenger loyalty.

[0099] The advantages of using cumulative standardized TALE to explain model results and explore the relationship between variables are mainly reflected in the following three aspects:

[0100] First, the cumulative standardized TALE is a model-independent explanatory method that does not depend on a specific model type, so it can be widely used to analyze the complex relationship between explanatory variables and response variables in different models. This feature allows researchers to explore passenger loyalty and its influencing factors without being restricted by the selection of models, thus ensuring the wide applicability and depth of the analysis.

[0101] Secondly, compared with ALE, standardized TALE can not only explain the relationship between variables with strong correlation, but also fully consider the impact of the actual score distribution of the sample. This makes it more suitable for reflecting the true overall effect of the explanatory variables on the response variable. This feature is particularly important for passenger loyalty research, because it can more accurately capture the comprehensive and true overall effect of the explanatory variables on the response variable of passenger loyalty, providing a more solid empirical basis for formulating targeted transportation policies.

[0102] The accumulated standardized TALE percentage value plays an important role in evaluating the impact of service attributes on passenger loyalty. While quantifying the specific contribution of each service attribute to passenger loyalty, it can also accurately identify the contribution of different service attributes to improving passenger loyalty in the same stage, accurately reflecting the changes in loyalty brought about by improving specific service attributes, and providing strong data support for the optimization of service quality. Therefore, the TALE method has good applicability in studying the relationship between service attributes and passenger loyalty.

[0103] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0104] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0105] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0106] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0107] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0108] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A passenger loyalty influence mechanism analysis method based on cumulative local total effect diagram, characterized in that: include: S1. Collect multi-dimensional data on the quality of public transport service elements and passenger loyalty; S2, test the reliability and validity of the data in S1, and study the correlation between the quality attributes of public transport service elements; S3. Preprocess the data, calculate the passenger loyalty score based on the passenger loyalty measurement model, build a causal analysis model between the quality of public transportation service elements and passenger loyalty, capture the relationship between the quality of service elements and passenger loyalty, and use the data to train the model; S4. Use the cumulative local total effect to analyze the impact of the service quality of each dimensional element in public transportation services on passenger loyalty.

2. The passenger loyalty influence mechanism analysis method based on cumulative local total effect diagram according to claim 1, characterized in that: The step S1 specifically includes: The passenger loyalty refers to the passenger's continued willingness to choose public transportation in the future based on their previous travel experience. Passenger loyalty is quantified using explicit variables, including: will continue to use public transportation in the future, public transportation is still the preferred mode of travel, willing to recommend public transportation services to others, and public transportation is still the preferred mode of travel even if they have the ability to buy a car; The quality of public transportation service elements includes two aspects: the service quality perceived by passengers in terms of transportation-related technical attributes and the service quality perceived by passengers in terms of comfort and safety. The service quality perceived by passengers in terms of transportation-related technical attributes refers to the quality of service elements perceived by passengers to meet their basic needs, including: public transportation service time, transfer convenience, walking distance for passengers to and from stations, and waiting time for passengers at stations; the service quality perceived by passengers in terms of comfort and safety refers to the service quality related to transportation comfort and safety perceived by passengers, including: station facility layout and waiting environment, station sanitary conditions, operating smoothness, temperature inside the vehicle, seat comfort and safety.

3. The passenger loyalty influence mechanism analysis method based on cumulative local total effect diagram according to claim 1, characterized in that: The reliability is used to test the internal consistency of the data. The Cronbach coefficient is used to test the reliability of the passenger loyalty data in S1. When the calculated Cronbach coefficient is greater than 0.7, it indicates that the passenger loyalty data is reliable; The validity is used to detect whether the manifest variables under each structure can reflect the content represented by passenger loyalty. During the verification process, the KMO statistic of the manifest variables, the significance probability of the Bartlett sphericity test, the average variance extracted and the combined reliability value are calculated respectively. When the calculation results are all within the preset range and the factor loading is greater than 0.5, it indicates that the passenger loyalty data is valid.

4. The passenger loyalty influence mechanism analysis method based on cumulative local total effect diagram according to claim 1, characterized in that: The correlation analysis is used to measure the degree of correlation between two different variables in the quality of public transportation service elements. When the correlation coefficient between the variables is greater than 0.7, it indicates that there is a strong correlation between the variables; when the correlation coefficient is between 0.4 and 0.7, it indicates that the correlation between the variables is strong; and when it is less than 0.4, it indicates that the correlation is weak; The positive correlation between variables in the quality of public transportation service elements is verified through the calculation results of correlation.

5. The passenger loyalty influence mechanism analysis method based on cumulative local total effect diagram according to claim 1, characterized in that: The step S3 specifically includes: The preprocessing is used to clean, organize and process the collected data; The passenger loyalty measurement model constructs a measurement model of passenger loyalty and corresponding explicit variables, and calculates a passenger loyalty score based on a path coefficient; The causal analytical model between the quality of public transportation service elements and passenger loyalty takes the quality of public transportation service elements as input and the passenger loyalty score as output. During the model training process, the grid search method and cross-validation method are used to tune the parameters, and the mean absolute percentage error is used to evaluate the prediction results of the model.

6. The passenger loyalty influence mechanism analysis method based on cumulative local total effect diagram according to claim 1, characterized in that: Step S4 specifically includes: S41. Calculate the average local effect of the variables in the quality of public transport service elements on passenger loyalty in the kth interval: Among them, x s represents the quality variable of the currently selected partially dependent public transport service element; x c Represents division by x s The vector of other explanatory variables is composed of ; the value range of the public transportation service element quality variable is divided into a series of intervals, K represents the number of intervals; Z k+1 It represents the lower bound value of the k+1th interval and the upper bound value of the kth interval; represents the set of data samples in the kth interval, Indicates the number of samples in the data sample set; i represents The i-th data sample in Representative Set Other explanatory variables x for the i-th data sample in c Scoring situation; Representative Set The partial dependent variable x of the i-th data sample in s The alternative score is is a differentiable function used to represent a machine learning model. The input of the machine learning model is the service quality of each dimensional element in public transportation services, and the output is the loyalty of passengers. X represents the vector composed of the service quality of each dimensional element in public transportation services. Represents x s The average local impact on passenger loyalty in the kth interval; S42, calculate x s The cumulative local total effect on the response variable is to calculate the cumulative local effect of each data sample, accumulate it in intervals, and form the cumulative local total effect: S43, normalize the cumulative local effect in S42 and display the x in each interval s The proportion of the cumulative local total effect on the response variable: in, Represents partial dependence on variable x S The unstandardized cumulative local total effect on the response variable, Express Cumulative percentage curves of normalized results were performed.

Images