Merchant operation mode identification method and device based on Fourier transform

Abstract

The invention discloses a merchant operation mode identification method and device based on Fourier transform, and the method comprises the steps: carrying out the preprocessing of collected transaction flow data of a target merchant, so as to obtain a time domain signal of the target merchant; determining a corresponding time domain feature based on the time domain signal; based on the time domain signal, combining Fourier transform to obtain a corresponding frequency domain feature; based on the time domain feature and the frequency domain feature, determining a multi-dimensional feature of the target merchant; obtaining a corresponding similarity evaluation result based on the multi-dimensional features of the target merchant and the preset multi-dimensional features of the reference merchant; and determining an operation mode of the target merchant based on a similarity evaluation result. According to the method, a multi-dimensional evaluation system of a time domain and a frequency domain is constructed, the matching degree of a target merchant and a reference merchant is quantified from multi-dimensional features, and accurate identification of the operation mode is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of small and micro merchant credit management, and in particular to a merchant operation mode recognition method and device based on Fourier transform. BACKGROUND

[0002] In the field of small and micro merchant credit management, traditional risk control assessment mainly relies on static total indicators such as total transaction amount and daily average number of transactions (hereinafter referred to as traditional indicators), but such indicators have significant limitations: unable to capture the regularity of operating time: the transaction flow of the merchant has typical periodic characteristics (for example, the early peak of a breakfast restaurant, the lunch peak of a lunch restaurant), and the traditional indicators only count the total amount, and cannot quantify the stability of the peak period (for example, whether the transaction proportion at noon each day is stable) and the strength of the periodicity (for example, whether the daily cycle transaction fluctuation is significantly higher than the weekly cycle). Insufficient coverage for low-transaction-volume merchants: small and micro merchants lacking standardized data (for example, with fewer daily transactions, which may be due to incomplete flow data acquisition) result in the traditional risk control model being unable to effectively assess their credit risk, and even being directly excluded from the credit range.

[0003] At present, the most similar existing implementation schemes to the present application mainly include the following four types.

[0004] 1. Statistical feature-based evaluation method: This method evaluates the operation of the merchant by calculating statistical quantities such as mean, variance, and peak of the merchant's transaction data. However, this method can only reflect the surface characteristics of the data and cannot deeply mine the periodicity and pattern characteristics of the transaction data.

[0005] 2. Machine learning-based classification method: Some companies attempt to use machine learning algorithms (such as decision trees, logistic regression, etc.) to classify and evaluate merchants. These methods usually require a large amount of labeled data for model training, and the model has poor interpretability. In addition, for small and micro merchants with low-frequency transactions, the model evaluation effect is often not ideal due to insufficient data samples.

[0006] 3. Time series analysis-based method: Some studies use time series analysis techniques (such as autoregressive models, moving average models, etc.) to analyze merchant transaction data. These methods can capture the time correlation of the data, but have limited effectiveness in dealing with non-linear, multi-period complex transaction patterns, making it difficult to accurately extract the core operating characteristics of the merchant.

[0007] 4. Rule-based pattern recognition method: Some financial institutions develop a series of rules to identify the operating patterns of merchants based on industry experience, such as classifying catering merchants into breakfast restaurants, lunch restaurants, and dinner restaurants according to their peak periods. This method relies on human experience and lacks objectivity and universality, making it difficult to adapt to the characteristics of merchants in different industries and different regions.

[0008] Therefore, an effective and reliable business operation mode recognition method is urgently needed to assist in risk assessment of small and micro business credit. SUMMARY

[0009] The application provides a business operation mode recognition method and device based on Fourier transform, aiming to effectively recognize the operation mode of a business.

[0010] To achieve the above-mentioned purpose, the application provides the following technical solutions:

[0011] A business operation mode recognition method based on Fourier transform, comprising:

[0012] Preprocessing the target business transaction flow data collected to obtain the time domain signal of the target business; the time domain signal comprises a plurality of data points and corresponding time sequences; the data points represent the number of transactions occurring within the time sequence;

[0013] Determining the corresponding time domain features based on the time domain signal;

[0014] Based on the time domain signal, the corresponding frequency domain features are obtained by combining Fourier transform;

[0015] Based on the time domain features and the frequency domain features, the multi-dimensional features of the target business are determined;

[0016] Based on the multi-dimensional features of the target business and the preset multi-dimensional features of the reference business, the corresponding similarity evaluation results are obtained; the reference business includes sample businesses with credit risks in operation mode;

[0017] Based on the similarity evaluation results, the operation mode of the target business is determined.

[0018] Optionally, based on the time domain signal, the corresponding time domain features are determined, comprising:

[0019] Determining a plurality of wave peaks in the time domain signal;

[0020] Filtering target wave peaks meeting a first condition from each of the wave peaks, and determining the time period corresponding to the target wave peaks as a wave peak time period; the first condition is that the wave peak height is greater than a target threshold, and the corresponding time period meets a preset time;

[0021] Calculating the arithmetic mean of the wave peak heights of each of the wave peaks to obtain the average wave peak height;

[0022] determine a peak intensity ratio based on a ratio of a peak height between a second peak and a first peak; the second peak is a peak with a second highest peak height among the peaks, and the first peak is a peak with a first highest peak height among the peaks;

[0023] determine a time interval between adjacent peaks among the peaks, and determine a minimum time interval and a maximum time interval from the time intervals;

[0024] determine a peak interval ratio based on a ratio of the minimum time interval and the maximum time interval;

[0025] determine a corresponding time domain feature based on the peak period, the peak average height, the peak intensity ratio, and the peak interval ratio.

[0026] Optionally, based on the time domain signal, a corresponding frequency domain feature is obtained by combining a Fourier transform method, including:

[0027] convert the time domain signal into a frequency domain signal by using the Fourier transform method; the frequency domain signal includes a plurality of complex numbers and corresponding orders; the complex number includes a frequency, an amplitude, and a phase;

[0028] extract a target amplitude and a target phase corresponding to a target frequency from the frequency domain signal; the target frequency represents a core period of the business operation;

[0029] determine a daily cycle intensity ratio based on a ratio between the target amplitude and a total amplitude; the total amplitude is a sum of each amplitude in the frequency domain signal;

[0030] determine a daily cycle phase based on the target phase;

[0031] determine a corresponding frequency domain feature based on the complex number, the daily cycle intensity ratio, and the daily cycle phase.

[0032] Optionally, based on the time domain feature and the frequency domain feature, a multi-dimensional feature of the target business is determined, including:

[0033] analyze the target business transaction flow data to obtain an average transaction number at different time nodes; the average transaction number is a mean value of transaction numbers at the same time node on different dates;

[0034] determine a global feature based on the average transaction number at the different time nodes;

[0035] determine the multi-dimensional feature of the target business based on the time domain feature, the frequency domain feature, and the global feature.

[0036] Optionally, based on the multi-dimensional features of the target merchant and the preset multi-dimensional features of the reference merchant, a corresponding similarity evaluation result is obtained, including:

[0037] The multi-dimensional features of the target merchant are compared with the preset multi-dimensional features of the reference merchant to obtain a multi-dimensional feature comparison result.

[0038] Based on the multi-dimensional feature comparison result, a corresponding similarity evaluation result is determined in combination with the preset weight of the multi-dimensional features.

[0039] Optionally, the multi-dimensional feature comparison result at least includes a frequency spectrum similarity, and a determination process of the frequency spectrum similarity is as follows:

[0040] Based on the frequency domain features of the target merchant, a corresponding first vector is determined, and based on the frequency domain features in the preset multi-dimensional features of the reference merchant, a corresponding second vector is determined.

[0041] The dot product of the first vector and the second vector is calculated.

[0042] Based on the ratio of the real part of the dot product to a target product, a frequency spectrum similarity is determined; the target product includes the product of the length of the first vector and the length of the second vector.

[0043] Optionally, based on the similarity evaluation result, an operating mode of the target merchant is determined, including:

[0044] The target merchant transaction flow data is parsed to obtain a transaction frequency of the target merchant.

[0045] Based on the similarity evaluation result and the transaction frequency, an operating mode of the target merchant is determined.

[0046] A merchant operating mode recognition device based on Fourier transform, comprising:

[0047] A transaction data processing unit is configured to preprocess the target merchant transaction flow data collected to obtain a time domain signal of the target merchant; the time domain signal includes a plurality of data points and corresponding time sequences; the data points represent the number of transactions occurring within the time sequences.

[0048] A time domain feature determination unit is configured to determine a corresponding time domain feature based on the time domain signal.

[0049] A frequency domain feature determination unit is configured to obtain a corresponding frequency domain feature based on the time domain signal and in combination with Fourier transform.

[0050] A multi-dimensional feature combination unit is configured to determine the multi-dimensional features of the target merchant based on the time domain features and the frequency domain features.

[0051] The similarity evaluation unit is configured to obtain a corresponding similarity evaluation result based on the multi-dimensional features of the target merchant and preset multi-dimensional features of reference merchants, wherein the reference merchants include sample merchants with credit risk in the business mode.

[0052] The business mode determination unit is configured to determine the business mode of the target merchant based on the similarity evaluation result.

[0053] A storage medium includes a stored program, wherein the program is executed by a processor to perform the Fourier transform-based business mode identification method.

[0054] An electronic device includes a processor, a memory and a bus, wherein the processor is connected to the memory through the bus.

[0055] The memory is configured to store a program, and the processor is configured to execute the program, wherein the program is executed by the processor to perform the Fourier transform-based business mode identification method.

[0056] The technical solution provided by the present application pre-processes the target merchant transaction flow data collected to obtain the time domain signal of the target merchant. Based on the time domain signal, the corresponding time domain features are determined. Based on the time domain signal, the corresponding frequency domain features are obtained by combining the Fourier transform. Based on the time domain features and the frequency domain features, the multi-dimensional features of the target merchant are determined. Based on the multi-dimensional features of the target merchant and the preset multi-dimensional features of the reference merchants, the corresponding similarity evaluation result is obtained. Based on the similarity evaluation result, the business mode of the target merchant is determined. The present application quantifies the matching degree of the target merchant and the reference merchant from multiple dimensions by constructing a multi-dimensional evaluation system of time domain and frequency domain, and realizes accurate identification of the business mode. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0058] Figure 1 A flowchart of a Fourier transform-based business mode identification method provided by an embodiment of the present application is shown in the figure;

[0059] Figure 2 A flowchart of another Fourier transform-based business mode identification method provided by an embodiment of the present application is shown in the figure;

[0060] Figure 3 FIG. 1 shows a flowchart of another embodiment of the method for identifying a business operation mode based on Fourier transform provided in the present application.

[0061] Figure 4 FIG. 1 shows a flowchart of another embodiment of the method for identifying a business operation mode based on Fourier transform provided in the present application.

[0062] Figure 5 FIG. 1 shows a flowchart of another embodiment of the method for identifying a business operation mode based on Fourier transform provided in the present application.

[0063] Figure 6 FIG. 1 shows a flowchart of another embodiment of the method for identifying a business operation mode based on Fourier transform provided in the present application.

[0064] Figure 7 FIG. 1 shows a flowchart of another embodiment of the method for identifying a business operation mode based on Fourier transform provided in the present application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0066] In the present application, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between the entities or operations. The term “comprises”, “comprising” or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the statement “comprises a” does not exclude the presence of additional identical elements in the process, method, article or equipment including the element.

[0067] As shown in FIG. 1, a flowchart of a method for identifying a business operation mode based on Fourier transform provided in an embodiment of the present application is shown, which includes the following steps. Figure 1

[0068] S101: Preprocessing the target business transaction flow data collected to obtain the time domain signal of the target business. ​

[0069] The time-domain signal includes multiple data points and their corresponding time series; the data points represent the number of transactions that occur within the time series.

[0070] In some examples, the target merchants include, but are not limited to, small and micro-sized merchants in various industries, including but not limited to the catering industry, retail industry, and life service industry.

[0071] In some examples, the target merchant transaction data includes the target merchant's transaction volume and the number of transactions.

[0072] In some examples, the collection period for the target merchant's transaction data can be a continuous N days, where N is a positive integer. Specifically, N can be set to 30 to ensure that the merchant's periodic operating characteristics can be captured.

[0073] In some examples, the granularity of the target merchant's transaction data can be summarized according to time nodes. The time nodes can be set by technical personnel according to the actual situation. For example, the time node can be set to an hour, that is, the target merchant's transaction data is summarized by hour to form an hourly transaction sequence. For example, the number of transactions in 24 hours of a day is summarized into 24 data points, and a sequence of 30×24=720 data points is formed in 30 days. This sequence is recorded as the target merchant's transaction data.

[0074] It should be noted that the purpose of preprocessing the transaction data of the target merchants is to remove noise and interference from the original data, in order to prepare for subsequent Fourier transform and feature extraction.

[0075] In some examples, preprocessing includes, but is not limited to, operations such as filling in missing values, zero-mean normalization, and denoising. Filling in missing values ​​means filling missing hourly transaction data with zeros to ensure the continuity of the data sequence. Zero-mean normalization means removing the DC component from the signal, making the mean of the processed signal zero. Denoising means filtering high-frequency noise and frequencies with extremely low amplitudes.

[0076] In some examples, the transaction data of the target merchant can be regarded as a typical discrete time series signal, that is, a time domain signal. Specifically, the time domain signal can be denoted as x[n], n=0,1,…,N−1, where N is the number of sampling points (which can be understood as the total number of hours, N=24×number of days), and x[n] can represent the number of transactions in the nth hour.

[0077] S102: Determine the corresponding time-domain characteristics based on the time-domain signal.

[0078] Among them, time-domain features can be used to reflect the peak position and intensity distribution of time-domain signals.

[0079] Optionally, the process of determining the corresponding time-domain features based on the time-domain signal can be found in [reference needed]. Figure 2 The steps shown are accompanied by corresponding explanations.

[0080] S103: Based on the time-domain signal, combined with Fourier transform, the corresponding frequency-domain characteristics are obtained.

[0081] Among them, frequency domain features can be used to quantify the periodicity of the target merchant's operations.

[0082] Optionally, the process of obtaining the corresponding frequency domain features based on the time domain signal and combined with Fourier transform can be found in [reference needed]. Figure 3 The steps shown are accompanied by corresponding explanations.

[0083] S104: Determine the multi-dimensional characteristics of the target merchant based on time-domain and frequency-domain features.

[0084] Among them, determining the multi-dimensional characteristics of the target merchant based on time domain and frequency domain features can effectively represent the business model of the target merchant.

[0085] Optionally, the process of determining the multi-dimensional characteristics of the target merchant based on time-domain and frequency-domain features can be found in [reference needed]. Figure 4 The steps shown are accompanied by corresponding explanations.

[0086] S105: Based on the multi-dimensional features of the target merchant and the preset multi-dimensional features of the reference merchant, obtain the corresponding similarity assessment results.

[0087] Among them, the reference merchants include sample merchants whose business models pose credit risks.

[0088] In some examples, the type of reference merchant can also be set by technicians according to the actual situation. This application embodiment does not impose any restrictions, such as sample merchants without credit risk.

[0089] Optionally, the implementation process for obtaining the corresponding similarity assessment results based on the multi-dimensional features of the target merchant and the preset multi-dimensional features of the reference merchant can be found in [reference missing]. Figure 5 The steps shown are accompanied by corresponding explanations.

[0090] S106: Based on the similarity assessment results, determine the business model of the target merchant.

[0091] The similarity assessment results can be divided into similarity levels, such as high similarity, medium similarity, and low similarity.

[0092] It should be noted that, to further improve the accuracy of identifying the business model of target merchants, transaction frequency can also be introduced as a reference for business model identification. Optionally, the process of determining the business model of target merchants based on similarity assessment results can be as follows: analyze the transaction flow data of target merchants to obtain the transaction frequency of target merchants; and determine the business model of target merchants based on similarity assessment results and transaction frequency.

[0093] In some examples, transaction frequency can be categorized as high-frequency or low-frequency.

[0094] In possible implementations, the target merchant's business model may include one of the following: high frequency and high similarity, high frequency and low similarity, low frequency and high similarity, and low frequency and low similarity.

[0095] In some examples, different credit granting logics and credit limit coefficients can be used to assess the creditworthiness of target merchants with different business models. For details, please refer to Table 1.

[0096] Table 1

[0097] Business Model Credit granting logic Credit coefficient High frequency - high similarity Model premium + stability reward 1.4-1.8X High frequency - low similarity Benchmark credit x similarity 0.8-1.2X Low frequency - high similarity Transaction count compensation 1.5-2.0X

[0098] For the items shown in Table 1, low-frequency, high-similarity target merchants are selected for credit compensation because: in traditional merchant credit granting, low-frequency merchants often have lower credit coverage and lower credit limits. After compensation through the number of transactions, credit can be granted according to the credit granting scheme for high-frequency merchants.

[0099] Combination Figures 2-6 The method shown in this application has the following four innovative features.

[0100] (1) Innovation Point 1: Applying Fourier Transform to the analysis of merchant transaction flow sequences. Specifically, the hourly transaction flow of merchants is regarded as a time-domain signal that changes with time. The Fourier Transform is used to transform it to the frequency domain, thereby extracting the periodic features in the transaction data. The Fourier Transform can represent any periodic function as a linear combination of sine and cosine functions with different frequencies and amplitudes. This characteristic enables it to effectively capture the periodic patterns in merchant transaction data.

[0101] The innovation point is that the solution is to extract three core parameters from the transaction data through Fourier transform: frequency (representing the speed of periodic changes), amplitude (representing the strength of the contribution of the frequency to the signal), and phase (representing the starting position of a specific frequency component). This accurately depicts the time regularity and periodicity of merchants' operations and solves the problem that existing methods are unable to capture the periodicity of operations.

[0102] (2) Innovation Point Two: Constructing a dual-dimensional feature system of "time domain + frequency domain". Specifically, based on Fourier transform, a multi-dimensional feature system including time domain features, frequency domain features and global features is constructed. Time domain features reflect the location, intensity distribution and balance of transaction peaks; frequency domain features quantify the periodicity and stability of operations; global features eliminate the impact of daily fluctuations and reflect the long-term business model of merchants.

[0103] The second innovation addresses this issue by fusing multi-dimensional features to comprehensively characterize a merchant's business operations. Even for low-frequency merchants, valuable features can be extracted from their limited transaction data, solving the problem of insufficient evaluation based on sparse data. Furthermore, multi-dimensional features provide richer evaluation criteria, improving the objectivity and accuracy of the evaluation results.

[0104] (3) Innovation Point 3: Transaction Compensation Mechanism for Low-Frequency Merchants. Specifically, in response to the problem of sparse data for low-frequency merchants, a transaction compensation mechanism based on similarity is proposed. By calculating the "low-frequency compensation coefficient", the transaction number of low-frequency merchants is approximately amplified by the coefficient, so that they can be evaluated with high-frequency merchants at a similar level of data volume.

[0105] The third innovation is addressed by introducing a low-frequency compensation coefficient, which solves the problem of low-frequency merchants being underestimated due to insufficient data, improves their access to credit, and expands the coverage of financial services.

[0106] (4) Innovation Point 4: Based on a multi-dimensional similarity assessment method, specifically, based on industry operating rules, multiple typical transaction models can be predefined as assessment benchmarks, or typical merchants can be selected as evaluation benchmarks. Through a custom similarity function and complex dot product method, the matching degree between merchants and typical models / merchants can be assessed in multiple dimensions, thereby realizing the automatic identification and classification of merchant operating models.

[0107] The fourth innovation is addressed by: defining typical patterns by combining industry knowledge and data characteristics, avoiding the subjectivity of relying solely on manual rules; the similarity assessment method enables automatic matching between merchants and typical patterns / merchants, improving assessment efficiency and adapting to the assessment needs of large-scale merchants; and the quantitative calculation of similarity provides an objective basis for merchant classification, improving the accuracy of classification results.

[0108] Compared to existing technologies, the advantages achieved by the embodiments of this application include: 1. More targeted feature extraction: The frequency domain features extracted by this invention based on Fourier transform can directly reflect the periodicity and regularity of merchants' operations. These features are closely related to the merchants' business models and have clear business meanings. In contrast, machine learning methods usually rely on data-driven feature selection, and the extracted features may lack clear business interpretations; 2. Higher interpretability: The evaluation results of this invention are based on explicit features and similarity calculations, which can clearly explain the source and basis of the evaluation results. In contrast, many machine learning models (such as deep learning models) are called "black box" models, and their decision-making processes are difficult to interpret. This is a significant disadvantage in fields such as credit assessment where interpretability is highly demanding; 3. Processing sparse data 4. Enhanced processing capability: By constructing a dual-dimensional feature system in the time and frequency domains and a low-frequency compensation mechanism, this invention can effectively handle sparse data of low-frequency merchants. Machine learning methods often require a large amount of labeled data, and their performance will significantly decrease when the data is sparse. 5. No need for a large amount of labeled data: This invention predefines typical patterns and does not require a large amount of labeled merchant data with well-defined business models. Machine learning methods usually require a large amount of labeled data for model training, resulting in high data acquisition costs. 6. Higher computational efficiency: The feature extraction and similarity calculation processes of this invention are relatively simple and have low computational complexity, enabling rapid evaluation of a large number of merchants. Complex machine learning models (such as deep learning models) often require a large amount of computational resources and time, which is not conducive to the evaluation of large-scale merchants.

[0109] The processes shown in S101-S106 above construct a multi-dimensional evaluation system in the time and frequency domains to quantify the matching degree between target merchants and reference merchants from multiple dimensions of features, thereby achieving accurate identification of business models.

[0110] like Figure 2 The diagram shown is a flowchart of another merchant business model recognition method based on Fourier transform provided in this application embodiment, including the following steps.

[0111] S201: Determine multiple peaks in the time-domain signal.

[0112] Among them, the peak refers to the maximum amplitude within a wavelength range.

[0113] S202: Select target peaks that meet the first condition from all peaks, and determine the time period corresponding to the target peaks as the peak time period.

[0114] The first condition is that the peak height is greater than the target threshold and the corresponding time period meets the preset time.

[0115] In some examples, the target threshold can be determined based on individual data points in the time-domain signal, for example, by using 1.2 times the 75th percentile of the mean of each data point as the target threshold.

[0116] In some examples, the preset time can be set to 3 hours.

[0117] S203: Calculate the arithmetic mean of the peak heights of each peak to obtain the average peak height.

[0118] In this process, the peak height of each peak in the time-domain signal is determined in advance, and then the arithmetic mean of the peak heights of each peak is calculated to obtain the average peak height.

[0119] S204: Determine the peak intensity ratio based on the ratio of the peak height between the second peak and the first peak.

[0120] Among them, the second peak is the peak with the second highest peak height, and the first peak is the peak with the highest peak height. Generally speaking, the peak height ratio is essentially the ratio of the peak height of the second peak to the peak height of the first peak.

[0121] It should be emphasized that if the time-domain signal has only one peak, the intensity ratio of that peak can be directly set to 0.

[0122] S205: Determine the time interval between adjacent peaks in each peak, and determine the minimum and maximum time intervals from each time interval.

[0123] S206: The peak interval ratio is determined based on the ratio of the minimum time interval to the maximum time interval.

[0124] Among them, the peak spacing ratio can be used to reflect the uniformity of peak distribution in a time-domain signal.

[0125] S207: Based on the peak time period, average peak height, peak intensity ratio, and peak interval ratio, the corresponding time-domain features are determined.

[0126] Among them, the peak time period, average peak height, peak intensity ratio, and peak interval ratio are determined as the corresponding time-domain features, which can effectively reflect the time-domain characteristics of the target merchant's transaction flow data.

[0127] In some examples, the corresponding time-domain features can be determined based on one or more of the following: peak duration, average peak height, peak intensity ratio, and peak interval ratio.

[0128] The processes shown in S201-S207 above can effectively obtain the time-domain characteristics of the target merchant's transaction flow data from the time-domain signal.

[0129] likeFigure 3 The diagram shown is a flowchart of another merchant business model recognition method based on Fourier transform provided in this application embodiment, including the following steps.

[0130] S301: Use the Fourier transform method to convert the time-domain signal into a frequency-domain signal.

[0131] The frequency domain signal includes multiple complex numbers and their corresponding sequence numbers; the complex numbers include frequency, amplitude, and phase.

[0132] In some examples, the Fourier transform method includes, but is not limited to, the Discrete Fourier Transform (DFT) and the Fast Fourier Transform (FFT).

[0133] In some examples, the Discrete Fourier Transform can convert the time-domain signal x[n] into the frequency-domain signal X[k], the specific representation of which can be found in Equation (1).

[0134] (1)

[0135] In formula (1), k represents the frequency index and k = 0, 1, ..., N−1, N is the number of sampling points, and j represents the imaginary unit and j 2 =-1, Represents the rotation factor (a complex exponential function that describes the periodicity of frequency k).

[0136] In some examples, the frequency domain signal X[k] can be viewed as a sequence of complex numbers, where each complex number contains a frequency. ,amplitude Phase Generally speaking, frequency The periodicity and amplitude of the corresponding time-domain signal The strength of the corresponding cycle (which can be understood as the fluctuation range of the number of transactions), phase The starting position of the corresponding cycle (which can be understood as the offset of the peak period).

[0137] In some examples, the results of the Fast Fourier Transform (FFT) are consistent with those of the Discrete Fourier Transform (DFT). Specifically, assuming a data acquisition time interval... =1 hour, then sampling frequency =1 / Frequency resolution = / N, the actual frequency corresponding to frequency index k It can be denoted as formula (2).

[0138] (2)

[0139] In a possible implementation, if N is set to 720, then =k / 720, when k is set to 30 =30 / 720=1 / 24.

[0140] It should be noted that the Fast Fourier Transform (FFT) result of the frequency domain signal x[n] has conjugate symmetry (i.e., X[k]=X∗[N−k]). Only the first N / 2 frequency points (k=1,2,…,N / 2) need to be retained to fully describe the periodicity of the frequency domain signal. For this purpose, the frequency axis can be calculated through the core function of the FFT, and the positive frequency part in the frequency axis can be selected and recorded as the frequency. The corresponding execution statement can be: freq=fft freq(N,d=1)⇒positive_freq=freq[freq>0].

[0141] In some examples, amplitude It is the magnitude of X[k], reflecting the strength of the contribution of that frequency to the time-domain signal; specifically, the amplitude. The expression form can be found in formula (3).

[0142] (3)

[0143] In formula (3), represent The real part, represent The imaginary part.

[0144] In some examples, phase It is the frequency domain The argument reflects the starting position of the frequency in the time domain (which can be understood as the phase shift of the period). Specifically, the phase... The expression form can be found in formula (4).

[0145] (4)

[0146] In formula (4), the arctan2(y,x) function returns the angle of the point (x,y) in polar coordinates, and the range of the angle is (-π,π).

[0147] S302: Extract the target amplitude and target phase corresponding to the target frequency from the frequency domain signal.

[0148] Among them, the target frequency represents the core cycle of a merchant's operation.

[0149] In some examples, the core cycle of a merchant's operation is typically a daily cycle (e.g., 24 hours), corresponding to a frequency of 1 / 24 hours.

[0150] In a possible implementation, the frequency of the daily cycle Corresponding index See formula (5) for details.

[0151] (5)

[0152] In a possible implementation, the target amplitude is equal to the product of the amplitude and the frequency index corresponding to the daily period, specifically it can be regarded as = .

[0153] S303: The daily periodic intensity percentage is determined based on the ratio between the target amplitude and the total amplitude.

[0154] The total amplitude is the sum of the amplitudes of all the signals in the frequency domain.

[0155] In some examples, the daily cycle intensity ratio can be expressed as shown in formula (6).

[0156] (6)

[0157] S304: Based on the target phase, it is determined to be the daily cycle phase.

[0158] S305: Based on complex numbers, the proportion of daily periodic intensity, and daily periodic phase, the corresponding frequency domain characteristics are determined.

[0159] Among them, one or more of the following can be used to determine the corresponding frequency domain features: complex number, daily cycle intensity ratio, and daily cycle phase.

[0160] The processes shown in S301-S305 above can achieve the acquisition of the corresponding frequency domain signal based on the combination of time domain signal and Fourier transform.

[0161] like Figure 4 The diagram shown is a flowchart of another merchant business model recognition method based on Fourier transform provided in this application embodiment, including the following steps.

[0162] S401: Analyze the transaction data of the target merchant to obtain the average number of transactions at different time points.

[0163] The average number of transactions is the average number of transactions at the same time point on different dates.

[0164] In some examples, different time points can be set to 24 hours, which means determining the average number of transactions over 24 hours over multiple days.

[0165] S402: The average number of transactions at different time points is determined as a global feature.

[0166] Among them, the average number of transactions at different time points is determined as a global feature, which can eliminate the daily fluctuations in the transaction flow data of the target merchant.

[0167] S403: Based on time domain features, frequency domain features, and global features, the multi-dimensional features of the target merchant are determined.

[0168] Among them, the multi-dimensional features identified based on time domain features, frequency domain features, and global features can provide a more reliable reference for business model identification.

[0169] The processes shown in S401-S403 above can determine the multi-dimensional features of the target merchant based on time domain features, frequency domain features, and global features, thereby improving the reliability of multi-dimensional features.

[0170] like Figure 5 The diagram shown is a flowchart of another merchant business model recognition method based on Fourier transform provided in this application embodiment, including the following steps.

[0171] S501: Compare the multi-dimensional features of the target merchant with the preset multi-dimensional features of the reference merchant to obtain the multi-dimensional feature comparison results.

[0172] Among them, key features can be selected from the multi-dimensional features of the target merchant and compared with the corresponding features in the preset multi-dimensional features to obtain multi-dimensional feature comparison results.

[0173] In some examples, the daily cycle intensity ratio, peak period, and daily cycle phase can be used as key features and compared with the corresponding features in the preset multi-dimensional features to obtain the multi-dimensional feature comparison results: daily cycle intensity ratio difference, peak period overlap, and phase consistency.

[0174] Optionally, the multi-dimensional feature comparison results should include at least spectral similarity. The process for determining spectral similarity can be found in [link to relevant documentation]. Figure 6 The steps shown are accompanied by corresponding explanations.

[0175] S502: Based on the multi-dimensional feature comparison results and combined with the preset weights of the multi-dimensional features, determine the corresponding similarity evaluation results.

[0176] Among them, based on the multi-dimensional feature comparison results and combined with the preset weights of the multi-dimensional features, the corresponding similarity evaluation result can be determined as shown in formula (7).

[0177] (7)

[0178] In formula (7), This represents the similarity assessment result. Represents the difference in daily cycle intensity ratio, Represents the overlap of peak periods. Represents phase consistency. These represent the preset weights corresponding to the multi-dimensional features.

[0179] In some examples, when using multi-dimensional feature comparison results to determine the corresponding similarity assessment results, it is necessary to normalize the multi-dimensional feature comparison results to improve the reliability of the similarity assessment results.

[0180] The processes shown in S501-S502 above can determine the corresponding similarity evaluation results based on the multi-dimensional feature comparison results and the preset weights of the multi-dimensional features, thereby improving the reliability of the similarity evaluation results.

[0181] like Figure 6 The diagram shown is a flowchart of another merchant business model recognition method based on Fourier transform provided in this application embodiment, including the following steps.

[0182] S601: Based on the frequency domain characteristics of the target merchant, determine the corresponding first vector, and based on the frequency domain characteristics in the preset multi-dimensional characteristics of the reference merchant, determine the corresponding second vector.

[0183] In some examples, the first vector can be determined by extracting the complex numbers of three core frequencies from the frequency domain features of the target merchant. Similarly, the second vector can be determined by extracting the complex numbers of three core frequencies from the frequency domain features of the reference merchant.

[0184] In a possible implementation, the first vector can be considered as The second vector can be regarded as ,in Represents amplitude, Represents phase.

[0185] S602: Calculate the dot product of the first vector and the second vector.

[0186] The calculation process of the dot product of the first vector and the second vector can be found in formula (8).

[0187] (8)

[0188] In formula (8), The conjugate transpose of the complex number representing the reference merchant. The cosine value represents the phase difference.

[0189] S603: The spectral similarity is determined based on the ratio of the real part of the dot product to the product of the target product.

[0190] The target product includes the product of the magnitude of the first vector and the magnitude of the second vector.

[0191] In some examples, the expression for spectral similarity can be found in formula (9).

[0192] (9)

[0193] In formula (9), Represents the magnitude of the first vector. , Represents the magnitude of the second vector. , This indicates that the spectrum of the target merchant and the reference merchant are consistent. This indicates that the target merchant and the reference merchant are not related.

[0194] The processes shown in S601-S603 above can use the first vector and the second vector to determine the spectral similarity between the target merchant and the reference merchant.

[0195] like Figure 7 The diagram shown is a schematic representation of the architecture of a merchant business model recognition device based on Fourier transform provided in an embodiment of this application, including the following units.

[0196] The transaction data processing unit 100 is used to preprocess the collected transaction flow data of the target merchant to obtain the time domain signal of the target merchant; the time domain signal includes multiple data points and corresponding time series; the data points represent the number of transactions that occur within the time series.

[0197] The time-domain feature determination unit 200 is used to determine the corresponding time-domain features based on the time-domain signal.

[0198] Optionally, the time-domain feature determination unit 200 is specifically used for: determining multiple peaks in the time-domain signal; selecting target peaks that meet a first condition from each peak, and determining the time period corresponding to the target peak as the peak time period; the first condition is: the peak height is greater than the target threshold, and the corresponding time period meets the preset time; calculating the arithmetic mean of the peak heights of each peak to obtain the average peak height; determining the peak intensity ratio based on the ratio of the peak heights of the second peak and the first peak; the second peak is the peak with the second highest peak height among all peaks, and the first peak is the peak with the first highest peak height among all peaks; determining the time interval between adjacent peaks in each peak, and determining the minimum time interval and the maximum time interval from each time interval; determining the peak interval ratio based on the ratio of the minimum time interval and the maximum time interval; and determining the corresponding time-domain feature based on the peak time period, the average peak height, the peak intensity ratio, and the peak interval ratio.

[0199] The frequency domain feature determination unit 300 is used to obtain the corresponding frequency domain features based on the time domain signal and combined with the Fourier transform.

[0200] Optionally, the frequency domain feature determination unit 300 is specifically used to: convert the time domain signal into a frequency domain signal using the Fourier transform method; the frequency domain signal includes multiple complex numbers and their corresponding ordinal positions; the complex numbers include frequency, amplitude, and phase; extract the target amplitude and target phase corresponding to the target frequency from the frequency domain signal; the target frequency represents the core cycle of the merchant's operation; determine the daily cycle intensity ratio based on the ratio between the target amplitude and the total amplitude; the total amplitude is the sum of all amplitudes in the frequency domain signal; determine the daily cycle phase based on the target phase; and determine the corresponding frequency domain feature based on the complex number, the daily cycle intensity ratio, and the daily cycle phase.

[0201] The multi-dimensional feature combination unit 400 is used to determine the multi-dimensional features of the target merchant based on time-domain features and frequency-domain features.

[0202] Optionally, the multi-dimensional feature combination unit 400 is specifically used to: analyze the transaction flow data of the target merchant to obtain the average number of transactions at different time points; the average number of transactions is the average number of transactions at the same time point on different dates; based on the average number of transactions at different time points, it is determined as a global feature; based on the time domain feature, frequency domain feature and global feature, it is determined as a multi-dimensional feature of the target merchant.

[0203] The similarity assessment unit 500 is used to obtain the corresponding similarity assessment results based on the multi-dimensional characteristics of the target merchant and the preset multi-dimensional characteristics of the reference merchants; the reference merchants include sample merchants whose business models have credit risks.

[0204] Optionally, the similarity evaluation unit 500 is specifically used to: compare the multi-dimensional features of the target merchant with the preset multi-dimensional features of the reference merchant to obtain the multi-dimensional feature comparison results; and determine the corresponding similarity evaluation results based on the multi-dimensional feature comparison results and the preset weights of the multi-dimensional features.

[0205] Optionally, the multi-dimensional feature comparison results include at least spectral similarity. The process by which the similarity evaluation unit 500 determines spectral similarity is as follows: based on the frequency domain features of the target merchant, a corresponding first vector is determined, and based on the frequency domain features in the preset multi-dimensional features of the reference merchant, a corresponding second vector is determined; the dot product of the first vector and the second vector is calculated; the ratio of the real part of the dot product to the target product is determined as the spectral similarity; the target product includes the product of the magnitude of the first vector and the magnitude of the second vector.

[0206] The business model determination unit 600 is used to determine the business model of the target merchant based on the similarity assessment results.

[0207] Optionally, the business model determination unit 600 is specifically used to: analyze the transaction flow data of the target merchant to obtain the transaction frequency of the target merchant; and determine the business model of the target merchant based on the similarity assessment results and the transaction frequency.

[0208] The units shown above, by constructing a multi-dimensional evaluation system in the time and frequency domains, quantify the matching degree between target merchants and reference merchants from multiple dimensions of features, thereby achieving accurate identification of business models.

[0209] This application also provides a computer-readable storage medium including a stored program, wherein the program executes the merchant business pattern recognition method based on Fourier transform provided in this application.

[0210] This application also provides an electronic device, including a processor, a memory, and a bus. The processor and the memory are connected via the bus. The memory is used to store a program, and the processor is used to run the program. When the program runs, it executes the merchant business pattern recognition method based on Fourier transform provided in this application.

[0211] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0212] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A Fourier transform-based merchant operation pattern recognition method, characterized by, The method comprises the following steps: Preprocessing the target merchant transaction flow data obtained by collection to obtain a time domain signal of the target merchant; the time domain signal comprises a plurality of data points and corresponding time sequences; The data points represent the number of transactions occurring within the time sequence; Based on the time domain signal, the corresponding time domain features are determined; Based on the time domain signal, the corresponding frequency domain features are obtained by combining the Fourier transform; Based on the time domain features and the frequency domain features, the multi-dimensional features of the target merchant are determined; Based on the multi-dimensional features of the target merchant and the pre-set multi-dimensional features of the reference merchant, the corresponding similarity evaluation results are obtained; The reference merchant includes sample merchants with credit risk in the business model; Based on the similarity evaluation results, the business model of the target merchant is determined.

2. The method of claim 1, wherein, Based on the time domain signal, the corresponding time domain features are determined, which comprises: Determine the plurality of wave peaks in the time domain signal; From each of the wave peaks, the target wave peak that meets the first condition is selected, and the time period corresponding to the target wave peak is determined as the wave peak time period; the first condition is that the wave peak height is greater than the target threshold, and the corresponding time period meets the pre-set time; Calculate the arithmetic mean of the wave peak height of each wave peak to obtain the wave peak average height; Based on the wave peak height ratio between the second wave peak and the first wave peak, the wave peak intensity ratio is determined; the second wave peak is the wave peak with the second highest wave peak height among the plurality of wave peaks, and the first wave peak is the wave peak with the first highest wave peak height among the plurality of wave peaks; Determine the time interval between adjacent wave peaks in each of the wave peaks, and determine the minimum time interval and the maximum time interval from each of the time intervals; Based on the ratio of the minimum time interval and the maximum time interval, the wave peak interval ratio is determined; Based on the wave peak time period, the wave peak average height, the wave peak intensity ratio and the wave peak interval ratio, the corresponding time domain features are determined.

3. The method of claim 1, wherein, Based on the time domain signal, the corresponding frequency domain features are obtained by combining the Fourier transform method, which comprises: Using the Fourier transform method, the time domain signal is converted into a frequency domain signal; the frequency domain signal comprises a plurality of complex numbers and corresponding sequence positions; the complex number comprises frequency, amplitude and phase; From the frequency domain signal, the target amplitude and the target phase corresponding to the target frequency are extracted; the target frequency represents the core period of the business; Based on the ratio between the target amplitude and the total amplitude, the daily cycle intensity proportion is determined; the total amplitude is the sum of each amplitude in the frequency domain signal; Based on the target phase, the daily cycle phase is determined; Based on the complex number, the daily cycle intensity proportion and the daily cycle phase, the corresponding frequency domain features are determined.

4. The method of claim 1, wherein, Based on the time domain features and the frequency domain features, the multi-dimensional features of the target merchant are determined, which comprises: Analyzing the target merchant transaction flow data to obtain the average number of transactions at different time nodes; the average number of transactions is the mean value of the number of transactions at the same time node on different dates; Based on the average number of transactions at different time nodes, the global features are determined; Determine the multi-dimensional features of the target merchant based on the time domain features, the frequency domain features, and the global features.

5. The method of claim 1, wherein, Based on the multi-dimensional features of the target merchant and the preset multi-dimensional features of the reference merchant, obtain the corresponding similarity evaluation result, including: Compare the multi-dimensional features of the target merchant with the preset multi-dimensional features of the reference merchant to obtain the multi-dimensional feature comparison result; Based on the multi-dimensional feature comparison result and the preset weight of the multi-dimensional features, determine the corresponding similarity evaluation result.

6. The method of claim 5, wherein, The multi-dimensional feature comparison result at least includes a frequency spectrum similarity, and the determination process of the frequency spectrum similarity is: Based on the frequency domain features of the target merchant, determine a corresponding first vector, and based on the frequency domain features in the preset multi-dimensional features of the reference merchant, determine a corresponding second vector; Calculate the dot product of the first vector and the second vector; Based on the ratio of the real part of the dot product and the target product, determine the frequency spectrum similarity; the target product includes the product of the length of the first vector and the length of the second vector.

7. The method of claim 1, wherein, Based on the similarity evaluation result, determine the operating mode of the target merchant, including: Analyze the target merchant transaction flow data to obtain the transaction frequency of the target merchant; Based on the similarity evaluation result and the transaction frequency, determine the operating mode of the target merchant.

8. A Fourier transform-based merchant operation pattern recognition apparatus, characterized by comprising: Including: A transaction data processing unit for pre-processing the target merchant transaction flow data collected to obtain a time domain signal of the target merchant; the time domain signal includes a plurality of data points and corresponding time sequences; the data points represent the number of transactions occurring in the time sequence; A time domain feature determination unit for determining the corresponding time domain features based on the time domain signal; A frequency domain feature determination unit for obtaining the corresponding frequency domain features based on the time domain signal and the Fourier transform; A multi-dimensional feature combination unit for determining the multi-dimensional features of the target merchant based on the time domain features and the frequency domain features; A similarity evaluation unit for obtaining the corresponding similarity evaluation result based on the multi-dimensional features of the target merchant and the preset multi-dimensional features of the reference merchant; The reference merchant includes a sample merchant with credit risk in the operating mode; An operating mode determination unit for determining the operating mode of the target merchant based on the similarity evaluation result.

9. A storage medium, characterized by The storage medium includes a stored program, wherein the program is run by the processor to execute the Fourier transform-based merchant operating mode identification method of any one of claims 1-7.

10. An electronic device, comprising: Including: A processor, a memory, and a bus; The processor and the memory are connected through the bus; The memory is used to store the program, and the processor is used to run the program, wherein the program is run by the processor to execute the Fourier transform-based merchant operating mode identification method of any one of claims 1-7.

Images