A method and system for monitoring sales behavior based on data visualization

Abstract

The present application relates to the technical field of computer monitoring, and more particularly to a sales behavior monitoring method and system based on data visualization. The method comprises: obtaining historical sales data and constructing behavior sequences; introducing a dual risk weight system for sales behavior: determining a first risk weight representing its inherent risk level; calculating a second risk weight representing dynamic abnormality; determining the alignment cost between sales behaviors by combining the difference between the first and second risk weights and the difference in quantitative index characteristics; calculating the distance between behavior sequences using the dynamic time warping algorithm, conducting clustering analysis based on the generated distance matrix, and identifying abnormal behavior sequences based on the clustering results. The present application significantly improves the accuracy of identifying complex and hidden abnormal sales patterns by constructing a comprehensive risk measurement system.

Description

Technical Field

[0001] This invention relates to the field of computer monitoring technology, and in particular to a sales behavior monitoring method and system based on data visualization. Background Technology

[0002] In modern enterprise management, the compliance and effectiveness of sales activities are directly related to the healthy development of the enterprise. To prevent risks such as sales fraud, irregular operations, and falsification of performance data, and to promptly identify inefficient or abnormal sales patterns, enterprises need to monitor and analyze massive amounts of sales data. This data is distributed across systems such as customer relationship management and enterprise resource planning, covering information from the entire process from customer contact, product quotation, contract signing to after-sales service. It is not only large in volume but also complex in structure.

[0003] Currently, mainstream sales behavior monitoring methods mainly rely on static rule engines or threshold judgments, which are difficult to adapt to the ever-changing market environment and complex business scenarios, and are prone to generating a large number of false alarms and missed alarms.

[0004] While related technologies have introduced sequence analysis methods to transform sales cases into time-ordered behavioral sequences and used algorithms such as dynamic time warping to compare process similarity, significant shortcomings remain. First, the distance metric used for sequence similarity calculation is simply defined, typically considering only quantitative differences in sales behaviors and neglecting the inherent business risks of different sales behaviors. Second, risk assessment is limited in scope; even when focusing on inherent risks in sales behaviors, it fails to consider the dynamic risks or anomalies of specific behaviors in conjunction with data distribution. However, whether a sales behavior is abnormal depends not only on its type but also on the position of its specific quantitative indicators within similar behaviors. In summary, existing technologies lack a sequence similarity measurement method that can comprehensively quantify both the inherent risks and dynamic anomalies of sales behaviors, thus limiting its accuracy in sales behavior monitoring applications. Summary of the Invention

[0005] To address the technical problems in sales behavior monitoring mentioned above, such as the inability to effectively combine the inherent business risks of sales behavior with the anomalies of dynamic data and the difficulty in identifying abnormal sales patterns, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a sales behavior monitoring method based on data visualization, the method comprising the steps of:

[0007] Historical sales data containing multiple sales cases are acquired. Each sales case includes various sales behaviors and types. For each predefined sales behavior type, a first risk weight is calculated to characterize its inherent risk level. A behavior sequence consisting of multiple sales behaviors arranged chronologically is constructed for each sales case. For any sales behavior in the behavior sequence, a second risk weight is calculated based on the statistical distribution of at least one quantitative indicator it contains and the quantitative indicators of similar sales behaviors. The second risk weight is positively correlated with both the numerical deviation and the numerical rarity of the quantitative indicator. Based on the characteristic differences of the quantitative indicators of any two sales behaviors in any two behavior sequences, the differences in the second risk weights of each sales behavior, and the differences in the first risk weights of the corresponding sales behavior types, the alignment cost of any two elements in the two behavior sequences is determined. The alignment cost of all elements in the behavior sequence is acquired one by one. A dynamic time warping algorithm is used to calculate the distance between any two behavior sequences to generate a distance matrix. Based on the results of cluster analysis of the distance matrix, abnormal behavior sequences are identified, thereby achieving monitoring of sales behaviors.

[0008] This invention innovatively constructs a dual-risk assessment system. The first risk weight focuses on the inherent business attributes of sales behavior, accurately characterizing the risk level inherent in different behavior types. The second risk weight targets dynamic data characteristics, capturing abnormal fluctuations during behavior execution through the deviation and rarity of quantitative indicators. This system is integrated into the core of the Dynamic Time Warping (DTW) algorithm, aligning cost calculations. By coupling the dual risk weights with the differences in behavioral quantitative indicators, a similarity measurement standard more closely aligned with actual business needs is formed. This design overcomes the limitations of existing technologies that rely solely on a single numerical dimension for sequence comparison, preserving the objectivity of data characteristics while incorporating subjective judgment of business risks. Ultimately, this assessment logic improves the accuracy and sensitivity in identifying complex and hidden abnormal sales patterns, significantly reducing the false positive and false negative rates caused by the single dimension in traditional methods, providing more reliable technical support for sales risk prevention and control.

[0009] Preferably, obtaining the numerical deviation includes: calculating the difference between the quantitative indicator and the median of the quantitative indicator for the same type of sales behavior, and normalizing the difference using the interquartile range of the same type of sales behavior.

[0010] This invention utilizes the median and interquartile range to effectively resist the interference of outliers on the statistical benchmark. This ensures that the calculation benchmark for the second risk weight is stable and reliable, avoiding the problem of overall assessment standard inaccuracy caused by a few extreme outliers, thereby improving the accuracy of dynamic risk assessment.

[0011] Preferably, the rarity of the numerical value is a quantified value obtained by performing an inverse operation on the ratio value; wherein, the calculation of the ratio value includes: counting the frequency of occurrence of the quantified indicator value within a preset range centered on the quantified indicator value of the current sales behavior in the same type of sales behavior; and calculating the ratio of the occurrence frequency to the total number of the same type of sales behavior to obtain the ratio value.

[0012] This invention objectively measures the rarity of a value by calculating its sparsity within its local data density. This method intuitively reflects the rarity of data points in a behavioral sequence, adding a crucial data distribution perspective to dynamic risk assessment and enhancing the ability to discover uncommon values ​​that may foreshadow new risk patterns.

[0013] Preferably, the alignment cost of any two elements in the two behavioral sequences includes: multiplying the feature difference of the quantitative indicators of the two sales behaviors by the sum of the difference of the first risk weight and the difference of the second risk weight to obtain the alignment cost of the two elements.

[0014] This invention employs a multiplicative structure, allowing the difference between two risk weights to act as an amplification factor for feature differences. This means that even if two behavioral sequences are not significantly different in numerical value, their alignment cost will be drastically increased if their risk levels differ significantly. This mechanism greatly enhances the algorithm's sensitivity to risk factors, effectively penalizing matches between high-risk and low-risk behaviors, and making the identification results more consistent with risk control logic.

[0015] Preferably, the characteristic difference of the quantitative indicators of the two sales behaviors is the Euclidean distance between the quantitative indicators of the two sales behaviors.

[0016] Preferably, the calculation of the first risk weight used to characterize its inherent risk level includes: dynamically adjusting and determining the first risk weight based on a preset basic risk value for the sales behavior type and in combination with the violation frequency of the sales behavior type in historical sales data.

[0017] This invention introduces an adaptive learning mechanism for determining the first risk weight. It makes inherent risk no longer completely static, but capable of self-correction and evolution based on historical data. This feedback loop allows the system to learn from past experience, automatically increasing the risk level of behaviors proven to be more likely to trigger violations, making the risk assessment model more timely and intelligent.

[0018] Preferably, the dynamic adjustment includes: determining a dynamic adjustment factor that is proportional to the violation frequency, and using the dynamic adjustment factor to correct the base risk value to determine the first risk weight.

[0019] Preferably, the clustering analysis of the distance matrix includes: using a density-based noise-based spatial clustering algorithm to cluster the distance matrix.

[0020] Preferably, the step of acquiring historical sales data containing multiple sales cases includes: acquiring historical sales data, which includes text data, time data, and monetary data; and performing at least one of the following operations on the historical sales data: vectorizing the text data, uniformly formatting the time data, and normalizing the monetary data.

[0021] In a second aspect, the present invention provides a sales behavior monitoring system based on data visualization, the sales behavior monitoring system based on data visualization including a memory and a processor, the memory storing computer program instructions, and when the computer program instructions are executed by the processor, implementing the sales behavior monitoring method based on data visualization of the first aspect of the present invention.

[0022] By adopting the above technical solution, a sales behavior monitoring method based on data visualization according to the first aspect of the present invention is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor, thereby creating a terminal device based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows: By constructing a multi-dimensional, adaptive comprehensive risk assessment system and deeply integrating it into the similarity measurement process of behavioral sequences, the invention fundamentally improves the intelligence, accuracy, and depth of sales behavior monitoring. This invention achieves a leap from static rule-based judgment to dynamic cognitive assessment by combining a first risk weight that can be dynamically adjusted based on historical violation frequency and a second risk weight calculated based on robust statistical methods. This gives risk assessment both the depth of business logic and the flexibility of data-driven approaches. More importantly, this dual risk system is integrated into a dynamically time-warped alignment cost function through multiplication, creating a risk amplification effect and greatly enhancing the algorithm's sensitivity to high-risk behavioral combinations. This enables the invention to effectively identify complex abnormal patterns composed of a series of seemingly normal but risk-accumulating events. Finally, combined with a density-based clustering algorithm, it achieves automated and accurate discovery of abnormal sequences, significantly reducing the false positive and false negative rates of traditional methods, and providing enterprises with a more reliable and intelligent risk management tool. Attached Figure Description

[0024] Figure 1 A flowchart illustrating a sales behavior monitoring method based on data visualization, provided as an embodiment of the present invention;

[0025] Figure 2This is a structural block diagram of a sales behavior monitoring system based on data visualization, provided as an embodiment of the present invention. Detailed Implementation

[0026] The first aspect of this invention provides a sales behavior monitoring method based on data visualization, such as... Figure 1 As shown, the method includes steps S100-S400:

[0027] Step S100: Obtain historical sales data containing multiple sales cases. Each sales case contains multiple sales behaviors and multiple sales behavior types. For each preset sales behavior type, calculate the first risk weight used to characterize its inherent risk level.

[0028] It should be noted that to monitor sales activities, it is necessary to obtain historical sales data from the sales system. In order to improve the quality of sales data, it is usually necessary to preprocess the obtained sales data.

[0029] Specifically, all sales activities are extracted from the database of the enterprise CRM system or intelligent monitoring platform, such as customer registration, quote modification, discount application, and policy amendment. The data corresponding to these sales activities is then preprocessed, including: converting text data into numerical vectors using a bag-of-words model; converting time data into uniform timestamps; smoothing the data using a moving average filter; and mapping monetary data to a minimum-maximum normalization method. The interval is used to eliminate the influence of dimensions. The above processing methods are all existing technologies and will not be elaborated here.

[0030] It should be noted that, to quantify the risks of sales activities, a fixed semantic risk weight must be assigned to each type of sales activity based on its business nature and potential risk level. This weight directly reflects the inherent risks and provides a benchmark for risk assessment and anomaly identification. This semantic risk weight is determined jointly by a base risk value and a dynamic risk coefficient. The base risk value is statically determined based on business attributes, compliance requirements, etc. For example, policy amendments have a high base value due to their association with financial security; customer registration has a small impact and a low base value.

[0031] Based on the above logic, the semantic risk weights satisfy the following relation:

[0032] ;

[0033] in, It is the first Semantic risk weights for sales behavior types It is the first The basic risk value preset for each type of sales behavior. It is a dynamic adjustment factor used to control the amplifying effect of violation frequency on the underlying risk. It is the first The frequency of violations for each type of sales behavior in historical sales data.

[0034] In this formula, through Introducing historical violation frequency This allows semantic risk weights to break free from the static limitations of relying solely on business attributes. If the frequency of violations for a certain type of sales behavior increases, the semantic risk weights will... It directly amplifies and promptly reflects actual risk fluctuations. Base risk value. This ensures that the underlying logic of risk quantification does not deviate from the essence of the business. Through this formula, the semantic risk weight of a behavior with low basic risk but historically often used for illegal operations, such as policy modification, will be dynamically increased, enabling the model to have the ability to adaptively learn from potential risks.

[0035] It should be noted that when retrieving the frequency of violations for different sales behavior types from historical sales data, if there is limited historical sales data, all historical sales data can be selected; if several years of sales data exist, the sales data from the past year can be selected. Regarding... The value can be determined by combining industry and historical data, and implementers can also set it according to their needs.

[0036] For example, in the financial and insurance industry, risks are necessitated because violations can easily lead to legal disputes and reputational damage. If sufficient historical data is available, such as sales data from the past year, [the following settings could be implemented]. Because the industry's sales data is rigorously verified and highly reliable, it has a high pass rate. The value amplifies the impact of violation frequency on weighting, strictly restricting high-risk operations. If historical data is limited, with only sales data from the past six months available, then [the following setting is applied]. Because of limited data, the statistical frequency of violations may be significantly biased. In this case, it is necessary to reduce the data. Minimize interference. In the luxury goods industry, discounts and special terms directly impact profits, requiring precise control while maintaining reasonable business flexibility. If sufficient historical data is available, such as sales data from the past year, set... The amplification effect on violations is moderate, protecting profits while allowing for reasonable exceptions. If historical data is limited, with only sales data from the past six months available, then setting... This reduces the impact weight of violation frequency and avoids excessively restricting business expansion due to a small amount of abnormal data.

[0037] It should be noted that the preset base risk value for each sales activity type is based on the nature of the sales activity type and the definition of potential risks. Implementation personnel can also set this value according to their needs.

[0038] For example, sales activities such as customer registration typically involve entering customer identity and contact information. Since this generally doesn't involve direct profits and is a basic business process, the likelihood of violations is low; therefore, its basic risk value can be set at 0.1. Sales activities such as price modification are usually influenced by actual promotional strategies, and the operational process is relatively standard, with controllable risks; therefore, the basic risk value can be set at 0.2. Discount applications, especially special discount applications, require a strict approval process. Abuse can directly lead to lost profits, resulting in a relatively high risk level; therefore, the basic risk value is set at 0.3. Policy amendments often involve modifying and adjusting terms, posing certain compliance risks. Unclear reasons for amendments or lack of customer confirmation can lead to disputes, requiring strengthened supervision; therefore, its basic risk value is set at 0.4.

[0039] Furthermore, regarding the types of sales activities, this invention defines... , , , , Each of these corresponds to a sales activity: customer registration, price modification, discount application, policy amendment, and whether the transaction is completed or not. The implementation personnel can adjust the settings according to the scenario.

[0040] Thus, the semantic risk weights of the sales behavior types used to characterize their inherent risk levels have been obtained. For clarity, the semantic risk weights will be referred to as the first risk weights below.

[0041] Step S200: Construct a sequence of behaviors consisting of multiple sales behaviors in chronological order for each sales case; for any sales behavior in the sequence, calculate the second risk weight of the sales behavior based on the statistical distribution of at least one quantitative indicator it contains and the quantitative indicators of similar sales behaviors.

[0042] It's important to note that the second risk weight complements and deepens the first risk weight. The first risk weight addresses the question of which types of sales behavior are more risky, while the second risk weight aims to address the anomaly of the numerical values ​​in this particular sales behavior. For example, two discount applications might have the same first risk weight, but applying for a 5% discount in one instance and a 95% discount in the other carries drastically different risks at the data level. The second risk weight, through its dimensions of numerical deviation and numerical rarity, precisely captures the dynamic risk arising from the statistical anomalies of specific quantitative indicators.

[0043] Specifically, it includes two operations: constructing the behavioral sequence and calculating the second risk weight. These will be explained separately below.

[0044] Regarding the construction of behavioral sequences, it's important to note that this approach breaks away from the isolated analytical perspective of individual sales behaviors. By using the complete sales cycle of a single customer as a guide, outlining the dynamic evolution path of sales, the temporal relationships between sales behaviors, stage characteristics, and the cumulative effects of quantitative indicators can be clearly presented, helping to accurately identify potential risks.

[0045] Specifically, taking an individual customer as a unit, a sales case is defined as the complete path from the initial contact record in customer registration to the final result of a sale or failure. Each sales case is transformed into a behavioral sequence, which is arranged in ascending order by timestamps. Each event point in the behavioral sequence is a feature vector containing normalized quantitative indicators of the sales behavior.

[0046] In addition, building a sequence of behaviors requires ensuring that the order of events conforms to business logic.

[0047] Let's illustrate this with two sales cases: Sales Case 1 includes customer registration → quote modification → discount application → transaction; Sales Case 2 includes customer registration → quote modification → discount application → policy amendment → transaction. The behavioral sequence corresponding to Sales Case 1 is as follows. , , , , These correspond to customer registration, price modification, discount application, and transaction completion in Sales Case 1. Sales Case 2 corresponds to the following behavioral sequence. , , , , , These correspond to customer registration, quote modification, discount application, policy amendment, and transaction in Sales Case 2, respectively.

[0048] Each sales action within a sequence of actions typically includes more than one quantifiable metric. For example, a discount application usually includes quantifiable metrics such as discount rate, number of discounts, and discount amount. Similarly, a price modification usually includes quantifiable metrics such as number of modifications and price modification amount.

[0049] This completes the acquisition of all behavioral sequences corresponding to all sales cases.

[0050] When calculating the second risk weight, it should be noted that for any sales behavior in the behavioral sequence, its second risk weight is determined by the numerical deviation and rarity of the quantitative indicators it contains.

[0051] Regarding the acquisition of numerical deviation, it should be noted that, in order to measure the degree to which a value deviates from its normal level, this embodiment does not use the mean and standard deviation, which are susceptible to extreme values, but instead uses the more robust median and interquartile range.

[0052] Specifically, the difference between the quantitative indicator and the median of the quantitative indicators of similar sales behaviors is calculated, and the difference is normalized using the interquartile range of similar sales behaviors.

[0053] Taking any behavioral sequence as the current behavioral sequence, the numerical deviation of the quantitative indicator of any sales behavioral sequence in the current behavioral sequence satisfies the following relationship:

[0054] ;

[0055] in, It is the first in the current action sequence The first sales transaction The degree of deviation of the values ​​of a quantitative indicator. It is the first in the current action sequence The first sales transaction A quantitative indicator, It is the first of all action sequences. The first sales transaction The median of a quantitative indicator, It is the first of all action sequences. The first sales transaction The interquartile range of a quantitative indicator is obtained using existing technology, so I won't go into details here. This is a tiny value used to prevent the denominator from being 0. It can be set to 0.01, or as needed. It is the absolute value symbol.

[0056] In this formula, To a certain extent, it reflects the first of all behavioral sequences. The first sales activity, such as the discount application activity, is the... Quantitative indicators such as the degree of dispersion corresponding to the discount rate. Reflecting the current action sequence, the first The first sales activity, such as the discount application activity, is the... Quantitative indicators, such as the relative magnitude of the discount rate deviating from the normal level, are examples. The larger the value, the more severe the deviation.

[0057] Regarding the calculation of numerical rarity, it should be noted that even if a value has a low deviation, if its occurrence is extremely rare, it may still indicate a risk. This invention quantifies numerical rarity by calculating a ratio and performing an inverse operation on it.

[0058] Specifically, first, calculate the proportion. Statistically count the frequency of occurrences of the quantitative indicator values ​​within a preset range centered on the quantitative indicator value of the current sales activity within the same type of sales activity; for example, the first... The value of each quantitative indicator The frequency of occurrence within a given value is calculated, and then this frequency is taken as a percentage of the total number of similar sales transactions. The smaller this percentage is, the fewer values ​​are similar to the current quantifiable indicator. Then, this percentage is used inversely to obtain the final rarity quantifiable value. Thus, the smaller the percentage, the larger the rarity quantifiable value.

[0059] Specifically, inverse operation, in existing technology, refers to an operation that can undo the effect of another operation. Simply put, if an operation will... It became Then its inverse operation can... Turn back Implementers can choose to use common inverse operations, such as taking the reciprocal, according to their needs. This invention performs negative exponentiation on the proportional value, which is also a type of inverse operation. The numerical rarity of this invention satisfies the following relationship:

[0060] ;

[0061] in, It is the first in the current action sequence The first sales transaction The rarity of a quantitative indicator's value. It is the first in the current action sequence The first sales transaction The aforementioned ratio value of the first quantitative indicator, that is, in relation to the first... Of all sales activities of the same type, the first... The first sales transaction The frequency of occurrence of a quantitative indicator value within a preset range centered on the current quantitative indicator value is then calculated, and the proportion of this frequency to the total number of similar sales activities is then calculated.

[0062] This numerical rarity provides a smooth and bounded risk mapping, making the entire algorithm more stable in engineering implementation.

[0063] For example, on the same day, the discount rate for sales staff in discount application behavior was... That is, the first The values ​​of each quantitative indicator are Its calculation range is Therefore, the preset range is Assuming there are 1000 discount applications in the historical discount application process, it was found that... Within this range, such a high discount has never occurred before in history. Therefore, only the current record falls within this range, occurring once. With a ratio of 0.001, its rarity metric is 0.999.

[0064] The preceding text explained the two factors for determining the calculation of the second risk weight; the following text explains how to calculate the second risk weight based on these two factors.

[0065] It should be noted that the second risk weight is the second risk weight of the quantitative indicator of sales behavior in the behavioral sequence, and the above formula can be used in any behavioral sequence. Furthermore, in this invention, the second risk weight is positively correlated with both the numerical deviation and the numerical rarity of the quantitative indicator.

[0066] Based on the above logic, the second risk weight of the current behavior sequence satisfies the following relationship:

[0067] ;

[0068] in, It is the first in the current action sequence The first sales transaction The second risk weight of each quantitative indicator It is the first in the current action sequence The first sales transaction The degree of deviation of the values ​​of a quantitative indicator. It is the first in the current action sequence The first sales transaction The rarity of a quantitative indicator's value. , These are the weights for numerical deviation and numerical rarity, respectively.

[0069] In this formula, Reflecting the current action sequence, the first The first in the sales transaction The relative magnitude of each quantitative indicator deviating from the normal level. The larger the value, the more severe the deviation. Therefore, the [number]th [character / item] in the current action sequence... The first in the sales transaction The greater the second risk weight corresponding to a quantitative indicator, the higher the risk weight, and vice versa. The smaller the value, the less deviation there is, and the more likely the current action sequence is to be in the correct order. The first in the sales transaction The smaller the second risk weight corresponding to each quantitative indicator, the lower the risk weight. The larger the value, the higher the value within the action sequence. The first in the sales transaction The rarer a quantitative indicator is, the greater the corresponding second risk weight is likely to be. When The smaller the value, the higher the value within the behavior sequence. The first in the sales transaction The more conventional a quantitative indicator is, the smaller the corresponding second risk weight may be.

[0070] It should be added that, , These are the weights of numerical deviation and numerical rarity, which are adjusted manually or through data training to prioritize numerical deviation and numerical rarity in risk decision-making, making risk assessment more closely aligned with the risk sensitivities of actual business scenarios. , The implementation personnel need to set the risk priority based on the business scenario. When the magnitude deviation is the core risk, When rare operations are the core risk, Dynamic adjustments can make risk assessments more accurately match actual business needs.

[0071] For example, an insurance company sets up a risk assessment mechanism for policies of high-value clients with annual premiums exceeding 1 million yuan, focusing on preventing two types of risks. One is price risk, such as discount rates or premium adjustments deviating from normal ranges, potentially leading to revenue losses for the company. The other is operational risk, such as unusual patterns in policy amendments or quotation modifications, such as multiple amendments to beneficiaries within a short period, or modification reasons coded as special approvals with extremely rare occurrences, indicating potential irregularities. In this case, the weighting is set accordingly. , This approach not only reflects a greater focus on operational risks but also amplifies risk signals through the coefficient amplification effect, making it logically sound and in line with actual needs.

[0072] It is important to note that a sales activity may have one or more quantitative indicators. When the sales activity has only one quantitative indicator, the second risk weight of that indicator is the second risk weight of the entire sales activity. However, when a sales activity has multiple quantitative indicators, the average or maximum value of the second risk weights of all the indicators can be used as the second risk weight of the entire sales activity. Implementation personnel can choose according to the specific circumstances.

[0073] Therefore, based on the above calculation method, the second risk weight of all sales activities is calculated.

[0074] Step S300: Based on the characteristic differences of the quantitative indicators of any two sales behaviors in any two behavior sequences, the differences of the second risk weights of each sales behavior, and the differences of the first risk weights of the corresponding sales behavior types, determine the alignment cost of any two elements in the two behavior sequences.

[0075] It should be noted that the Dynamic Time Warping (DTW) algorithm is a classic algorithm that uses dynamic programming to find the optimal matching path between two time series of different lengths, thereby calculating their similarity. The core reason for using this algorithm in this invention is its ability to effectively handle inconsistencies in time series length and differences in behavioral rhythm within sales behavior sequences. For example, different customers may have different sales cycles, and key behaviors may occur at different times. By flexibly aligning similar behavioral nodes in the sequence, it accurately measures the overall similarity of behavioral sequences from different sales cases, providing a more quantitative basis for identifying abnormal sales patterns or risky behaviors that aligns with actual business time series characteristics.

[0076] Traditional DTW algorithms measure sequence similarity solely through numerical differences, such as Euclidean distance, neglecting the risk differences in business logic across different sales behaviors. This invention incorporates a first risk weight and a second risk weight into the alignment cost function, dynamically adjusting the sensitivity of high-risk events in sequence matching, thus making the optimized DTW algorithm more focused on potential violation patterns.

[0077] Specifically, based on behavioral sequences and behavioral sequence For example, this section explains how to calculate the alignment cost between any two action sequences. Action sequences In , , , The first risk weights are respectively , , , Behavioral sequence In , , , The second risk weights corresponding to the several sales behaviors are as follows: , , , Similarly, behavioral sequences In , , , , The first risk weights are respectively , , , , Behavioral sequence In , , , , The second risk weights corresponding to the various sales behaviors are as follows: , , , , .

[0078] It should be noted that the design of the aligned cost function aims to modulate and amplify the underlying numerical differences through dual risk differences. This modulation or weighted penalty idea can be embodied using a multiplicative structure.

[0079] Based on the above logic, the behavioral sequence Sales behavior and behavior sequence The alignment costs of sales activities in the process satisfy the following relationship:

[0080] ;

[0081] in, It is a sequence of behaviors The Middle Sales behaviors and behavior sequences The first in Alignment costs for similar sales activities. It is a sequence of behaviors The first in Quantitative indicators and behavioral sequences of sales-related behaviors The first in The Euclidean distance of quantitative indicators of sales-related behaviors. It is a sequence of behaviors The first in The first risk weight for this type of sales activity It is a sequence of behaviors The first in The first risk weight for this type of sales activity It is a sequence of behaviors The first in The second risk weight for similar sales behaviors It is a sequence of behaviors The first in The second risk weight for similar sales behaviors It is the absolute value symbol. It is the standard normalized function. It is a preset microvalue used to prevent It can be set to 0, or 0.01, or as needed.

[0082] This cost function causes the DTW algorithm to tend to align behaviors with similar risk levels during alignment, while avoiding aligning behaviors with vastly different risk levels. This allows the final DTW distance to more accurately reflect the similarity in risk patterns and risk pace between two sales cases.

[0083] Thus, by using the above method, the alignment cost of all sales behaviors in any two behavior sequences has been obtained.

[0084] Step S400: Using the dynamic time warping algorithm, the distance value between any two behavior sequences is calculated based on the alignment cost of sales behavior to generate a distance matrix; based on the results of cluster analysis of the distance matrix, abnormal behavior sequences are identified, thereby realizing the monitoring of sales behavior.

[0085] It should be noted that when using the Dynamic Time Warping (DTW) algorithm, a cost matrix needs to be generated first, and then the optimal warping path needs to be found. The final cumulative cost of this path is the distance between the two action sequences.

[0086] Specifically, the alignment cost calculated earlier is used to fill the cost matrix of the DTW algorithm, and then the optimal normalization path is determined. The final cumulative cost of the path is the distance between the two corresponding behavior sequences. All sales behavior sequences are paired up, and this process is repeated until a... The distance matrix, where This represents the total number of sales cases.

[0087] It should be noted that the distance matrix obtained through the preceding steps already contains risk-based similarity information among all sales cases. The final task is to automatically extract anomalous patterns from this matrix. Cluster analysis is an ideal tool for accomplishing this task. This invention considers using the density-based noise applied spatial clustering (DBSCAN) algorithm, which does not require pre-specifying the number of clusters and can naturally identify outliers that typically directly correspond to the most unique and anomalous sales cases in business contexts.

[0088] Specifically, the generated distance matrix is ​​used as input to the DBSCAN algorithm. Two hyperparameters, neighborhood radius and minimum sample size, are set. The algorithm automatically classifies all sales cases into three categories: Core points: points within dense areas, representing typical and common sales behavior patterns. Boundary points: points at the edges of dense areas. Noise points / outliers: isolated points that do not belong to any dense area. These points are the primary focus for abnormal behavior sequences. Regarding the settings for neighborhood radius and minimum sample size, this invention sets the neighborhood radius to 5 and the minimum sample size to 10; implementers can adjust these settings according to their needs.

[0089] Finally, based on the clustering analysis results, the identified noise points and specific clusters composed of high-risk behaviors are marked as anomalous behavior sequences. The clustering results can then be further visualized, for example, by highlighting outliers and anomalous clusters on a two-dimensional or three-dimensional scatter plot, thereby achieving efficient and accurate monitoring of sales behavior.

[0090] The second aspect of this embodiment provides a sales behavior monitoring system based on data visualization, such as... Figure 2 As shown, the sales behavior monitoring system based on data visualization includes a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement the first aspect of the present invention: a sales behavior monitoring method based on data visualization.

[0091] The sales behavior monitoring system based on data visualization also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0092] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0093] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A sales behavior monitoring method based on data visualization, characterized in that, include: Acquire historical sales data containing multiple sales cases. Each sales case contains multiple sales behaviors and multiple sales behavior types. For each pre-defined sales behavior type, calculate the first risk weight used to characterize its inherent risk level. For each sales case, a sequence of actions consisting of multiple sales behaviors arranged in chronological order is constructed. For any sales behavior in the sequence, a second risk weight is calculated based on the statistical distribution of at least one quantitative indicator contained therein and the quantitative indicators of similar sales behaviors. The second risk weight is positively correlated with both the numerical deviation and the numerical rarity of the quantitative indicator. Based on the characteristic differences of quantitative indicators of any two sales behaviors in any two behavioral sequences, the differences in the second risk weights of each sales behavior, and the differences in the first risk weights of the corresponding sales behavior types, the alignment cost of any two elements in the two behavioral sequences is determined, satisfying the following relationship: ; It is a sequence of behaviors The Middle Sales behaviors and behavior sequences The first in Alignment costs of similar sales activities It is a sequence of behaviors The first in Quantitative indicators and behavioral sequences of sales-related behaviors The first in The Euclidean distance of quantitative indicators of sales-related behaviors. It is a sequence of behaviors The first in The first risk weight for this type of sales activity It is a sequence of behaviors The first in The first risk weight for this type of sales activity It is a sequence of behaviors The first in The second risk weight for similar sales behaviors It is a sequence of behaviors The first in The second risk weight for similar sales behaviors It is the absolute value symbol. It is a standard normalized function; It is a preset microvalue used to prevent =0; The alignment cost of all elements in the behavior sequence is obtained one by one. The distance between any two behavior sequences is calculated using a dynamic time warping algorithm to generate a distance matrix. Based on the clustering analysis of the distance matrix, abnormal behavior sequences are identified, thereby enabling the monitoring of sales behavior.

2. The sales behavior monitoring method based on data visualization according to claim 1, characterized in that, The acquisition of the numerical deviation includes: Calculate the difference between the quantitative indicator and the median of the quantitative indicators of the same type of sales behavior, and normalize the difference using the interquartile range of the same type of sales behavior.

3. The sales behavior monitoring method based on data visualization according to claim 1, characterized in that, The numerical rarity is a quantized value obtained by performing an inverse operation on the proportional value; The calculation of the ratio value includes: The frequency of occurrence of quantitative indicators within a preset range centered on the value of the quantitative indicator for the current sales activity in the same type of sales activity. The occurrence frequency is calculated as a percentage of the total number of similar sales transactions to obtain the percentage value.

4. The sales behavior monitoring method based on data visualization according to claim 1, characterized in that, The calculation of the first risk weight used to characterize its inherent risk level includes: Based on the preset basic risk value for the sales behavior type, and combined with the violation frequency of the sales behavior type in historical sales data, the first risk weight is dynamically adjusted and determined.

5. The sales behavior monitoring method based on data visualization according to claim 4, characterized in that, The dynamic adjustment includes: A dynamic adjustment factor proportional to the frequency of violations is determined, and the base risk value is corrected using the dynamic adjustment factor to determine the first risk weight.

6. The sales behavior monitoring method based on data visualization according to claim 1, characterized in that, The clustering analysis of the distance matrix includes: The distance matrix is ​​clustered using a density-based noise spatial clustering algorithm.

7. The sales behavior monitoring method based on data visualization according to claim 1, characterized in that, The acquisition of historical sales data containing multiple sales cases includes: Obtain historical sales data, which includes text data, time data, and monetary data; The operation involves at least one of the following: vectorizing the text data in the historical sales data, uniformly formatting the time data, and normalizing the monetary data.

8. A sales behavior monitoring system based on data visualization, characterized in that, The data visualization-based sales behavior monitoring system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a data visualization-based sales behavior monitoring method according to any one of claims 1-7.

Images