After-sales evaluation method and system for enterprise marketing consultation service

Abstract

The invention relates to the technical field of after-sales satisfaction evaluation, in particular to an after-sales evaluation method and system for enterprise marketing consultation service, which identifies time lag association between each behavior and score through a Granger causal relationship test method, can establish dynamic mapping of score influence paths in a service process, and improves the evaluation efficiency. A key driving link of score feedback is accurately identified, and multi-cycle behaviors such as plan execution, customer return visit and suggestion revision are matched and rearranged by constructing a unified time coordinate structure, so that the alignment precision between service behavior data and score data is improved, the misjudgment risk caused by score time drift is reduced, and the service quality is improved. By introducing a nonlinear state conversion and periodic aggregation mechanism, a nonlinear change trend existing in a scoring sequence is captured, and error modeling is performed on a scoring evolution process in combination with a long and short-term memory network, so that trend deviation caused by short-term mutation and periodic abnormality can be compensated, and the scoring evolution process is more accurate. And the prediction precision and the scoring behavior explanation capability of dynamic modeling are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of after-sales satisfaction evaluation, and in particular to a after-sales evaluation method and system for enterprise marketing consulting services. Background Art

[0002] The technical field of after-sales satisfaction evaluation mainly focuses on collecting customer feedback information after product or service delivery, and is usually applied in commercial service scenarios. It uses structured means to analyze service experience quality, quantify satisfaction indicators, and guide service optimization decisions accordingly.

[0003] A post-sales evaluation method for enterprise marketing consulting services aims to establish a standardized evaluation process for evaluating service performance through quantitative indicators and feedback data after the delivery of consulting services, and provide decision-making basis for both service providers and customers. It can achieve a comprehensive evaluation of consulting results in multiple dimensions such as actual business improvement, market response changes, and customer acceptance, thereby improving service transparency and customer stickiness.

[0004] Existing technologies focus on collecting customer feedback once delivery is completed. They lack effective capture of the dynamic changes in customer responses, are unable to reveal the time lag effects caused by multiple influencing behaviors in the service process, and fail to consider the nonlinear trends and cyclical repetitions in actual operations, resulting in limited scoring interpretation capabilities. When service behaviors fluctuate in rhythm and customer feedback has delayed characteristics, traditional methods cannot identify the causal sources behind the score fluctuations, reducing the pertinence of service process optimization suggestions. The evaluation results will be biased, affecting the service provider's perception of actual performance and adjustment decisions. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a post-sales evaluation method and system for enterprise marketing consulting services.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a post-sales evaluation method for enterprise marketing consulting services, comprising the following steps: S1: Based on the multiple service action information archived in the enterprise consulting service task record table, the time intervals between project establishment response, approval feedback, and draft submission are compared to screen the combination items that meet the temporal consistency. The Granger causality test is used to calculate the synchronous offset rate between the late-comer variables and the satisfaction score curve in the combination, and a significant marker matrix is ​​established to generate a causal path set. S2: Based on the causal path set, extract the periodic data of the three indicators of plan execution, customer return visit, and suggestion revision, match them with the scoring sequence and rearrange them in bitwise order to construct a unified time coordinate structure to obtain a fusion matrix group; S3: Based on the fusion matrix group, the scoring step size is set and the suggestion execution, response feedback, and project record sequence are input. The nonlinear transformation and periodic aggregation of the state value are completed in chronological order. The error difference of the actual scoring sequence is calculated using a long short-term memory network, the state transfer path is updated, and a dynamic output sequence of satisfaction evolution is generated. S4: Based on the dynamic output sequence of the satisfaction evolution, extract the interview frequency, question and answer response, and delay records under the offset period, screen the difference values ​​and frequencies, classify them into dimensions, and generate an influencing factor group; S5: Based on the influencing factor group, the scoring range and distribution classification of the stable parameter combinations under the four dimensions of suggestion execution, cooperation, integrity, and adoption frequency are determined to generate an impact intensity distribution table corresponding to the satisfaction dimension.

[0007] As a further solution of the present invention, the specific steps of generating the causal path set are: Based on the multiple service action information archived in the enterprise consulting service task record table, the project response time, approval feedback time, and first draft submission time are located in the fields. All time items are arranged in natural time order to form an independent time vector sequence, generating a time sequence action list. Based on the time sequence action list, calculate the difference between any two time items, arrange all the results in ascending order, and then screen whether each difference falls within the set service period to generate a valid interval set; Based on the effective interval set, the Granger causality test is used to map the position of the action corresponding to each late time item on the satisfaction score timeline, and the offset rate calculation is performed on the difference between the score value and the action trigger position. The action points with consistent offset direction are marked and extracted to generate a causal path set.

[0008] As a further solution of the present invention, the Granger causality test is performed according to the formula: in: Indicates in The customer satisfaction rating of the enterprise in each cycle, Indicates in The historical satisfaction rating of each cycle, Indicates the scoring sequence itself The regression influence coefficient of the order lag term on the current rating value, Indicates in The follow-up actions of enterprises implemented in the cycle Indicates the first The basic influence coefficient of the order lag action on the current rating value, Indicates the Normalized coefficient of execution intensity of the order-lag action, Indicates the The sensitivity of the enterprise customer status to the action stimulus score within a period, Indicates the The intensity of external environmental disturbances in each cycle, represents the score sensitivity adjustment coefficient, represents the external interference adjustment coefficient, represents the maximum number of lag periods considered in the satisfaction rating series, represents the maximum number of hysteresis cycles considered in the subsequent action sequence, Indicates the The random disturbance term in the period score regression model.

[0009] As a further solution of the present invention, the specific steps of generating the fusion matrix group are: Based on the causal path set, perform periodic positioning operations on the three fields of plan execution frequency, customer return visit days, and suggestion revision rounds. Count the number of values ​​that appear in each period and bind them to the period number to generate a behavior period detail group. Based on the behavior cycle detail group, extract the cycle number of each time node in the scoring sequence, and perform a one-to-one matching mapping between the behavior indicator value and the corresponding cycle of the scoring value to generate a scoring mapping result set; Based on the score mapping result set, the score values ​​and behavior values ​​under all cycle numbers are combined and filled into the three-dimensional coordinate matrix structure in sequence, and the results of each cycle are filled into the corresponding cell position to generate a fusion matrix group.

[0010] As a further solution of the present invention, the specific steps of generating the dynamic output sequence of satisfaction evolution are: Based on the fusion matrix group, a scoring step value setting operation is performed and the recommended execution items, response feedback items, and project record items are divided into equal-length blocks in chronological order. A scoring group with the same order is generated for each scoring segment to generate a scoring cycle sequence. Based on the scoring cycle sequence, the score difference operation of adjacent cycles in each scoring group is calculated, and the interval classification and sign direction judgment are performed on the score change value of each segment. Then, the scoring results of each cycle are accumulated and merged according to the time period to generate a scoring state sequence; Based on the rating state sequence, a long short-term memory network is used to perform a square operation on the difference between the periodic rating value and the rating value of the same period in the actual rating sequence, and all difference terms are merged into the original rating transmission path. Value replacement is performed and the path results are updated to generate a dynamic output sequence of satisfaction evolution.

[0011] As a further solution of the present invention, the long short-term memory network is according to the formula: in: Indicates that in the cycle The enterprise customer satisfaction score value obtained by long short-term memory network prediction, Indicates that in the cycle The actual customer satisfaction ratings of the enterprises collected within the Represents a period The importance weight of the overall marketing service evaluation path of the enterprise, Represents a period The standard deviation of the customer rating series within the enterprise, Represents a period The relative rate of change in the frequency of interaction with internal enterprise customers is used to measure the intensity of behavioral fluctuations. The adjustment coefficient that represents the influence of the score standard deviation, The adjustment coefficient that represents the degree of influence of the intensity of behavioral fluctuations, Indicates the total number of scoring cycles in the after-sales service evaluation. It represents the total amount of error in the enterprise customer satisfaction evaluation after introducing scoring weights, scoring uncertainty and behavioral fluctuation factors.

[0012] As a further solution of the present invention, the specific steps of generating the impact factor group are: Based on the dynamic output sequence of satisfaction evolution, the absolute value of the score difference in each cycle is extracted, and the score offset points greater than the set interval are screened, and the offset segment cycle numbers are collected and processed to generate an offset cycle set; Based on the offset period set, the number of interview records, the number of question and answer responses, and the number of extension records are matched with the offset period number, and the number of occurrences of various indicators in the period are classified and statistically processed to generate a period parameter distribution table; Based on the periodic parameter distribution table, the repetition frequency of each group of parameters in the period in all periods is judged, and the parameter pairs that change in the same direction are merged and classified. The classification results are mapped to the corresponding dimension labels to generate an impact factor group.

[0013] As a further solution of the present invention, the specific steps of generating the influence intensity distribution table corresponding to the satisfaction dimension are: Based on the influencing factor group, a screening operation is performed on the correspondence between the four indicators of recommended execution, cooperation, completeness, and adoption frequency and the parameter content, and the extreme values ​​of each parameter in the scoring sequence are extracted. Each scoring value is classified and marked by dimension to generate a dimension scoring interval table; Based on the dimension score interval table, statistical processing is performed on the number of occurrences in each score value interval, the number of occurrences is converted into an integer frequency value, and it is determined whether the score values ​​in the same dimension are clustered in a certain range, the highest frequency interval and corresponding parameter combination are retained, and a dimension frequency concentration group is generated; Based on the dimension frequency concentration group, the score difference calculation operation of the highest frequency score interval under each dimension is performed, the segments with differences greater than the set step size are extracted and marked with intensity level numbers, and then all segment numbers are mapped to the label sequence to generate the influence intensity distribution table corresponding to the satisfaction dimension.

[0014] As a further solution of the present invention, the scoring value is discrete integer data, and its value range is set to positive integers between 1 and 10, with a step size of 1, that is, each scoring value in the scoring sequence is an integer within the interval and constitutes the basic data source of the scoring sequence.

[0015] A post-sale evaluation system for enterprise marketing consulting services, wherein the post-sale evaluation system for enterprise marketing consulting services is used to execute the above-mentioned post-sale evaluation method for enterprise marketing consulting services, and the system comprises: Causal Construction Module: Based on the three records of project approval response, approval feedback, and draft submission in the enterprise consulting service task record table, it calculates the time interval, screens the time-consistent items, uses the Granger causality test to calculate the synchronization offset rate between the subsequent action and the scoring curve, constructs the significance matrix, extracts the highly significant combinations, and generates a causal path set. Sequence alignment module: Based on the causal path set, extract the periodic data corresponding to plan execution, customer return visits, and suggestion revisions, match and rearrange them with the scoring sequence item by item, construct a unified time number, establish a consistent time axis structure, and obtain a unified time coordinate fusion matrix group; Trend modeling module: Based on the unified time coordinate fusion matrix group, the scoring step size is set, the recommended execution, response feedback, and project records are input, nonlinear state conversion and cycle aggregation are performed, and the long short-term memory network is used to perform error difference learning on the scoring sequence to generate a dynamic output sequence of satisfaction evolution; Factor extraction module: Based on the dynamic output sequence of satisfaction evolution, extract the interview frequency, question and answer response, and delay records within each score deviation period, screen the difference values ​​and high-frequency combinations, classify them into multi-dimensional influence sources, and construct an influence factor group; Distribution classification module: Based on the influencing factor group, the parameter values ​​under the four scoring dimensions of suggestion execution, cooperation, completeness, and adoption frequency are called to determine the frequency differences of each scoring interval, extract the mapping relationship between the strong difference segment number and the dimension label sequence, and generate the impact intensity distribution table corresponding to the satisfaction dimension.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: In this invention, the Granger causality test method is used to identify the time-lagged association between each behavior and the score, which can establish a dynamic mapping of the score impact path in the service process and accurately identify the key driving links of the score feedback; In this invention, by constructing a unified time coordinate structure, multi-cycle behaviors such as plan execution, customer return visits, and suggestion revisions are matched and rearranged, thereby improving the alignment accuracy between service behavior data and scoring data and reducing the risk of misjudgment caused by scoring time drift; In the present invention, by introducing nonlinear state transformation and periodic aggregation mechanisms, the nonlinear change trend in the scoring sequence is captured, and the long-short-term memory network is combined to perform error modeling on the scoring evolution process, which can compensate for the trend deviation caused by short-term mutations and periodic anomalies, and improve the prediction accuracy of dynamic modeling and the ability to explain scoring behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0020] Example 1 See also Figure 1 The present invention provides a technical solution: a post-sales evaluation method for enterprise marketing consulting services, comprising the following steps: S1: Based on the multiple service action information archived in the enterprise consulting service task record table, the time intervals between project establishment response, approval feedback, and draft submission are compared to screen the combination items that meet the temporal consistency. The Granger causality test is used to calculate the synchronous offset rate between the late-comer variables and the satisfaction score curve in the combination, and a significant marker matrix is ​​established to generate a causal path set. S2: Based on the causal path set, the periodic data of the three indicators of plan execution, customer return visit, and suggestion revision are extracted, and the data are rearranged by bit matching with the scoring sequence to construct a unified time coordinate structure and obtain a fusion matrix group; S3: Based on the fusion matrix group, the scoring step size is set and the suggestion execution, response feedback, and project record sequence are input. The nonlinear transformation and periodic aggregation of state values ​​are completed in chronological order. The error difference of the actual scoring sequence is calculated using a long short-term memory network, the state transfer path is updated, and a dynamic output sequence of satisfaction evolution is generated. S4: Based on the dynamic output sequence of satisfaction evolution, the interview frequency, Q&A response, and delay records under the offset period are extracted, the difference values ​​and frequencies are screened and classified into dimensions, and an influencing factor group is generated; S5: Based on the influencing factor group, the scoring range and distribution classification of the stable parameter combinations under the four dimensions of recommendation execution, cooperation, completeness, and adoption frequency are carried out to generate the impact intensity distribution table corresponding to the satisfaction dimension.

[0021] The specific steps to generate the causal path set are: Based on the multiple service action information archived in the enterprise consulting service task record table, the project response time, approval feedback time, and first draft submission time are located in the fields. All time items are arranged in natural time order to form an independent time vector sequence, generating a time sequence action list. Based on the time sequence action list, the difference between any two time items is calculated and all the results are sorted in ascending order. Each difference is then screened to see if it falls within the set service period to generate a valid interval set. Based on the effective interval set, the Granger causality test is used to map the position of the action corresponding to each post-time item on the satisfaction score time axis. The offset rate calculation is performed on the difference between the score value and the action trigger position. The action points with the same offset direction are marked and extracted to generate a causal path set. Based on the multiple service action information archived in the enterprise consulting service task record table, a field location method was used to extract fields from the record items named "Project Response Time," "Approval Feedback Time," and "Draft Submission Time." The standard SQL search statement SELECT was called, and the field matching function LOCATE was used to locate the columns where the fields were located one by one. The three time fields were type-checked according to the date and timestamp field format and uniformly converted to the YYYY-MM-DDHH:MM:SS format. The TIMESTAMPDIFF function was used to generate a numerical timestamp vector in seconds. The three fields were sorted in ascending order by timestamp size using the ORDER BY command to generate a time-series action list. Based on the time sequence action list, a nested FOR loop structure is used to calculate the difference between the timestamp values ​​of any two time items. The inner loop index variable is used to point to the starting position of the outer loop. The timestamp difference result is calculated by subtraction and stored in a vector structure. The standard sorting method QUICKSORT is used to sort the difference vector from small to large. An interval judgment operation is performed on each difference value. The upper and lower thresholds of the service cycle are set to 3600 seconds and 259200 seconds respectively. The IF conditional statement is called to filter all difference items within the interval and mark the filtering status as TRUE to generate a valid interval set. Based on the effective interval set, the Granger causality test is used to perform an offset rate analysis on the position mapping of the action corresponding to each post-time item on the satisfaction score time axis. The grangercausalitytests function in the statsmodels.tsa.stattools module in Python is used. The input parameters are a two-dimensional array consisting of the score time series and the post-action event time series. The maximum lag order maxlag is set to 5, and the test dimension verbose is set to False. The F test and p-value extraction operations are performed on each combination in turn. The combinations with F statistic values ​​greater than the critical value and p-values ​​less than 0.05 are retained. The difference between the score sequence time point and the event time point in these combinations is calculated, and the average difference is taken and its positive or negative direction is determined as the basis for the offset direction. The action points with the same offset direction are stored in a LIST structure and the mark value is set to "+1" to generate a causal path set.

[0022] The Granger causality test is based on the formula: in: Indicates in The enterprise customer satisfaction score of each cycle, Indicates in The historical satisfaction rating of each cycle, Indicates the scoring sequence itself The regression influence coefficient of the order lag term on the current rating value, Indicates in The follow-up actions of enterprises implemented in the cycle Indicates the first The basic influence coefficient of the order lag action on the current rating value, Indicates the Normalized coefficient of execution intensity of the order-lag action, Indicates the The sensitivity of the enterprise customer status to the action stimulus score within a period, Indicates the The intensity of external environmental disturbances in each cycle, represents the score sensitivity adjustment coefficient, represents the external interference adjustment coefficient, represents the maximum number of lag periods considered in the satisfaction rating series, represents the maximum number of hysteresis cycles considered in the subsequent action sequence, Indicates the The random disturbance term in the cycle score regression model; Implementation process: First, extract time-series customer satisfaction rating data from the company's after-sales service records and the corresponding subsequent action sequence , respectively set the maximum lag order of the scoring sequence and the maximum lag order of the action sequence Then, based on the historical rating trend, a rating lag structure is constructed to calculate each lag rating item. The influence coefficient , and incorporate it into the prediction model. At the same time, for each action event Perform execution intensity normalization to generate normalized weights , used to characterize the strength of the action implementation. Then, based on the rate of change of the score slope, the customer's sensitive response ability in each action cycle is analyzed and quantified as the customer state sensitivity coefficient , and combined with external information sources such as holidays, market opinions or industry news, quantitatively generate the external environment disturbance intensity of the corresponding period , then, the adjustment coefficient is introduced and , the optimal combination is determined by performing grid search on the training set to ensure that the model maintains optimal stability in response to sensitivity and disturbance effects. Finally, the action sequence, score sequence, autonomous and external regulation factors are integrated into the regression structure to calculate the score value of each cycle The prediction results are compared with the original score value to determine the Granger causality of each action in the time series, and then the subsequent actions with significant score impact are screened out to construct the final causal path set.

[0023] The specific steps to generate the fusion matrix group are: Based on the causal path set, periodic positioning operations are performed on the three fields: plan execution frequency, customer return visit days, and suggestion revision rounds. The number of values ​​appearing in each cycle is counted and bound to the cycle number to generate a detailed group of behavior cycles. Based on the behavior cycle detail group, the cycle number of each time node in the scoring sequence is extracted, and a one-to-one matching mapping is performed between the behavior indicator value and the corresponding cycle of the scoring value to generate a scoring mapping result set; Based on the score mapping result set, the score values ​​and behavior values ​​under all cycle numbers are combined and filled into the three-dimensional coordinate matrix structure in order. The results of each cycle are filled into the corresponding cell position to generate a fusion matrix group; Based on the causal path set, we used a periodization method to periodically locate records with the fields named "Plan Execution Frequency," "Customer Return Visit Days," and "Recommendation Revision Rounds." We used the date_range function in Pandas to set the start and end times of the period, with the period interval parameter set to 7 days. We used the cut function to map the timestamps of each field record to the corresponding period segment. We used the groupby method and the count function to count the number of occurrences of the record values ​​within each period segment. We then used the enumerate method to sequentially bind each period number to generate detailed groups of behavior periods. Based on the behavior cycle detail group, the cycle number extraction method is used to call the merge operation on the timestamp of each record in the scoring sequence to perform a left join match. The docking field is the cycle number field and the number field corresponding to the scoring time. After the cycle number is transferred, a one-to-one mapping operation is performed based on the behavior indicator value and the scoring value as the matching field. The zip function is used to construct a paired structure of the indicator value and the scoring value, and the structure is stored in a typical structure object to generate a scoring mapping result set. Based on the score mapping result set, a combined filling operation is performed on the period number, score value, and behavior value contained in each mapping record to construct a three-dimensional coordinate matrix structure. The zeros function in NumPy is called to initialize the three-dimensional matrix array dimension to n×m×k, corresponding to the period number, score value interval, and behavior value counting interval, respectively. The score value filling dimension is set to the middle axis, and the behavior value is filled in the third axis. The index assignment operation is used to fill in the combined value according to the period number to generate a fusion matrix group.

[0024] The specific steps to generate the dynamic output sequence of satisfaction evolution are: Based on the fusion matrix group, the scoring step value is set and the recommended execution items, response feedback items, and project record items are divided into equal-length blocks in chronological order. The same-order scoring groups are generated for each scoring segment to generate a scoring cycle sequence. Based on the scoring cycle sequence, the score difference operation of adjacent cycles in each scoring group is calculated, and the interval classification and symbol direction judgment are performed on the score change value of each segment. Then, the scoring results of each cycle are accumulated and merged according to the time period to generate the scoring status sequence; Based on the rating state sequence, a long short-term memory network is used to perform a square operation on the difference between the periodic rating value and the rating value of the same period in the actual rating sequence. All difference terms are merged into the original rating transfer path, and value replacement is performed and the path results are updated to generate a dynamic output sequence of satisfaction evolution. Based on the fusion matrix group, a fixed segmentation method is used to set the step length of the scoring step value to 5. The cut function in Pandas is used to divide the scoring sequence into equally spaced blocks according to the step length. The sort_values ​​function is called on the recommended execution items, response feedback items, and project record items to sort them in ascending order by the time field. The rolling window method is used to match the sorting results with the step length. The corresponding scoring value is extracted from each time block, and the equal-order scoring group is constructed and stored in a sequence to generate the scoring cycle sequence. Based on the scoring cycle sequence, the NumPy array operation method is used to calculate the difference of the score values ​​at two adjacent cycle positions in each scoring group. The np.diff function is used to generate a first-order difference array and use it as the score change. The sign function is called to determine the positive and negative directions of each difference to identify the trend type. The cut function is used to divide the score change values ​​into five categories according to the set interval boundaries of -3, -1, 0, 1, and 3. After extracting the score change mark for each category, the groupby and sum functions are used to accumulate the score values ​​according to the cycle segment number. The accumulated results are spliced ​​in chronological order to form a single sequence structure to generate a scoring status sequence. Based on the rating state sequence, a long short-term memory network is used to perform a square operation on the difference between the period rating value and the rating value of the same period in the actual rating sequence. The Keras framework is used to construct the network structure, the input layer is set to 64 dimensions, the activation function is tanh, the time step is 5, the number of hidden layer nodes is set to 128, the loss function is specified as mean_squared_error, the optimizer is adam, and the fit method is called to input the period rating value as X, the actual rating value as Y, the training rounds as 50, and the batch size as 16. The difference between the predicted value and the actual rating value of each period node is extracted, and the square calculation is performed item by item. The update_path function is called to replace the square difference with the corresponding position value in the original rating path to generate a dynamic output sequence of satisfaction evolution.

[0025] Long short-term memory network follows the formula: in: Indicates that in the cycle The enterprise customer satisfaction score value obtained by long short-term memory network prediction, Indicates that in the cycle The actual customer satisfaction ratings of the enterprises collected within the Represents a period The importance weight of the overall marketing service evaluation path of the enterprise, Represents a period The standard deviation of the customer rating series within the enterprise, Represents a period The relative rate of change in the frequency of interaction with internal enterprise customers is used to measure the intensity of behavioral fluctuations. The adjustment coefficient that represents the influence of the score standard deviation, The adjustment coefficient that represents the degree of influence of the intensity of behavioral fluctuations, Indicates the total number of scoring cycles in the after-sales service evaluation. It represents the total error in the enterprise customer satisfaction evaluation after introducing the scoring weight, scoring uncertainty and behavior fluctuation factors; Implementation process: First, for each corporate customer in the cycle Generate predicted rating values ​​based on the service feedback data within and compared with the actual satisfaction ratings collected The square of the difference is calculated to quantify the scoring deviation of a single cycle. In order to reflect the criticality of the scoring of each cycle in the overall marketing service delivery path of the enterprise, the importance weight of the corresponding node of each cycle is calculated. , which is determined by the interaction density of the node in the score transfer structure and the hierarchical position of the propagation path. In order to enhance the model's ability to respond to score volatility, the score standard deviation parameter is introduced , through the cycle The scoring samples are extracted by setting sliding windows before and after, and the behavioral fluctuation intensity parameter is introduced. By calculating the same enterprise customer in the cycle The interaction frequency change rate of the previous cycle is obtained, and then the adjustment coefficient and By setting multiple candidate value combinations on the training set and using cross-validation to evaluate the minimum prediction error to determine its value, the square of the score difference, the score importance weight and the two disturbance factors are combined in each cycle to form the adjusted score error through weighted calculation. , the error is embedded into the customer rating path and replaces the original rating node value, generating a dynamic evaluation result for guiding service optimization and satisfaction evolution modeling.

[0026] The specific steps to generate the impact factor group are: Based on the dynamic output sequence of satisfaction evolution, the absolute value of the score difference in each cycle is extracted, and the score offset points greater than the set interval are screened. The offset segment cycle numbers are collected and processed to generate an offset cycle set. Based on the offset period set, the number of interview records, number of Q&A responses, and number of extension records are matched with the offset period number. The number of occurrences of various indicators within the period is classified and statistically processed to generate a period parameter distribution table. Based on the period parameter distribution table, the repetition frequency of each group of parameters in the period in all periods is judged, and the parameter pairs that change in the same direction are merged and classified. The classification results are mapped to the corresponding dimension labels to generate the impact factor group; Based on the dynamic output sequence of satisfaction evolution, the abs function in NumPy is used to perform an absolute value operation on the score value of each period. The offset amplitude value of each score item in the difference sequence is extracted. The offset judgment threshold is set to 3. The Boolean index method is used to filter out all score nodes with an offset amplitude value greater than 3. The np.where function is used to extract the index value of the corresponding score node in the sequence as the period number set. After deduplication using the set function, the index value is converted into a list format using the tolist method to generate the offset period set. Based on the offset period set, a correspondence is established between the three types of field data ("number of interview records," "number of Q&A responses," and "number of deferred records") and the period numbers in the offset period set using the timestamp field. The merge command is used to match data within the time segment, and the time segment boundaries are defined as the start and end times of each period. The between statement is used to determine the inclusion relationship, marking each record that meets the conditions as a valid data point. The three types of indicators are grouped and counted by period number. The groupby and size methods are called to obtain the statistical values ​​within the period, and a period parameter distribution table is generated. Based on the periodic parameter distribution table, the value_counts function is called for the three parameters in each group of periods to calculate the number of repetitions in all periods. The frequency judgment threshold is set to 3, and all parameter combination items with a frequency greater than or equal to 3 are retained. For each pair of items in the remaining parameter combinations, np.sign is called to determine their change direction in adjacent periods. Boolean conditions are used to determine items with consistent directions. Parameter pairs with consistent directions are merged using a composite key. All merged items are stored in a dictionary structure and mapped accordingly according to the dimension label structure fields to generate an impact factor group.

[0027] The specific steps to generate the influence intensity distribution table corresponding to the satisfaction dimension are: Based on the influencing factor group, a screening operation is performed on the correspondence between the four indicators of recommended execution, cooperation, completeness, and adoption frequency and the parameter content. The extreme values ​​of each parameter in the scoring sequence are extracted, and each score value is classified and marked by dimension to generate a dimension score range table; Based on the dimension score interval table, perform statistical processing on the number of occurrences in each score value interval, convert the number of occurrences into integer frequency values, and determine whether the score values ​​in the same dimension are clustered in a certain range. Retain the interval with the highest frequency and the corresponding parameter combination to generate the dimension frequency concentration group; Based on the dimension frequency concentration group, the score difference operation of the highest frequency score interval under each dimension is calculated. The segments with a difference greater than the set step size are extracted and marked with intensity level numbers. Then all segment numbers are mapped to the label sequence to generate the influence intensity distribution table corresponding to the satisfaction dimension; Based on the influencing factor group, the merge function in Pandas is used to align the primary keys of the four indicators ("recommendation execution degree", "cooperation degree", "completeness", and "adoption frequency") with the factor group field. After extracting all parameter values ​​corresponding to each indicator, the amin and amax functions in NumPy are called to obtain the minimum and maximum values ​​of each indicator parameter in the scoring sequence, respectively. The scoring interval length is then set to 2, and the arange function is used to construct equally spaced scoring intervals from the minimum to the maximum value. The cut function is called on each scoring value, and the bins parameter is passed in to classify and mark them according to the dimension field to generate a dimension scoring interval table. Based on the dimension score interval table, the groupby method is called to perform grouped statistics on the number of samples within each score value interval. The count function is used to calculate the number of occurrences in each interval and convert it to an integer value. The idxmax function is used to locate the score interval with the largest frequency value within each dimension. The query function is called to extract the score segment with the highest frequency and its associated parameter combination. The result is encapsulated into a key-value structure and the original number of each score segment is retained. The corresponding relationship between the dimension and the score segment is then generated to generate a dimension frequency concentration group. Based on the dimension frequency concentration group, a combined difference processing is performed on the highest frequency interval number value in each scoring dimension. The NumPy diff function is called to extract the difference items between all numbers. The scoring span step threshold is set to 2. The Boolean filtering method is used to retain all number pairs with differences greater than the threshold. The eligible numbers are assigned a number sequence value by calling the enumerate method. The range function is used to mark them as level numbers, and the numbers are accumulated from 1 to the corresponding scoring segment. The zip function is used to create a mapping list for all level numbers and dimension labels to generate an impact intensity distribution table corresponding to the satisfaction dimension.

[0028] The scoring value is discrete integer data, and its value range is set to positive integers between 1 and 10, with a step size of 1. That is, each scoring value in the scoring sequence is an integer within the interval and constitutes the basic data source of the scoring sequence.

[0029] A post-sale evaluation system for enterprise marketing consulting services is provided. The post-sale evaluation system for enterprise marketing consulting services is used to implement the above-mentioned post-sale evaluation method for enterprise marketing consulting services. The system includes: Causal Construction Module: Based on the three records of project approval response, approval feedback, and draft submission in the enterprise consulting service task record table, it calculates the time interval, screens the time-consistent items, uses the Granger causality test to calculate the synchronization offset rate between the subsequent action and the scoring curve, constructs the significance matrix, extracts the highly significant combinations, and generates a causal path set. Sequence alignment module: Based on the causal path set, it extracts the periodic data corresponding to plan execution, customer return visits, and suggestion revisions, matches and rearranges them item by item with the scoring sequence, constructs a unified time number, establishes a consistent time axis structure, and obtains a unified time coordinate fusion matrix group; Trend Modeling Module: Based on a unified time coordinate fusion matrix group, it sets the scoring step size, inputs suggested execution, response feedback, and project records, performs nonlinear state transitions and periodic aggregation, and uses a long-short-term memory network to perform error difference learning on the scoring sequence to generate a dynamic output sequence of satisfaction evolution. Factor extraction module: Based on the dynamic output sequence of satisfaction evolution, it extracts the interview frequency, Q&A response, and delay records within each rating offset period, screens the difference values ​​and high-frequency combinations, classifies them into multidimensional influencing sources, and constructs an influencing factor group; Distribution classification module: Based on the influencing factor group, the parameter values ​​under the four scoring dimensions of suggestion execution, cooperation, completeness, and adoption frequency are called to determine the frequency differences of each scoring interval, extract the mapping relationship between the strong difference segment number and the dimension label sequence, and generate the impact intensity distribution table corresponding to the satisfaction dimension.

[0030] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A post-sales evaluation method for enterprise marketing consulting services, characterized in that: The following steps are involved: S1: Based on the multiple service action information archived in the enterprise consulting service task record table, the time intervals between project establishment response, approval feedback, and draft submission are compared to screen the combination items that meet the temporal consistency. The Granger causality test is used to calculate the synchronous offset rate between the late-comer variables and the satisfaction score curve in the combination, and a significant marker matrix is ​​established to generate a causal path set. S2: Based on the causal path set, extract the periodic data of the three indicators of plan execution, customer return visit, and suggestion revision, match them with the scoring sequence and rearrange them in bitwise order to construct a unified time coordinate structure to obtain a fusion matrix group; S3: Based on the fusion matrix group, the scoring step size is set and the suggestion execution, response feedback, and project record sequence are input. The nonlinear transformation and periodic aggregation of the state value are completed in chronological order. The error difference of the actual scoring sequence is calculated using a long short-term memory network, the state transfer path is updated, and a dynamic output sequence of satisfaction evolution is generated. S4: Based on the dynamic output sequence of the satisfaction evolution, extract the interview frequency, question and answer response, and delay records under the offset period, screen the difference values ​​and frequencies, classify them into dimensions, and generate an influencing factor group; S5: Based on the influencing factor group, the scoring range and distribution classification of the stable parameter combinations under the four dimensions of suggestion execution, cooperation, integrity, and adoption frequency are determined to generate an impact intensity distribution table corresponding to the satisfaction dimension.

2. The post-sales evaluation method for enterprise marketing consulting services according to claim 1, characterized in that: The specific steps of generating the causal path set are: Based on the multiple service action information archived in the enterprise consulting service task record table, the project response time, approval feedback time, and first draft submission time are located in the fields. All time items are arranged in natural time order to form an independent time vector sequence, generating a time sequence action list. Based on the time sequence action list, calculate the difference between any two time items, arrange all the results in ascending order, and then screen whether each difference falls within the set service period to generate a valid interval set; Based on the effective interval set, the Granger causality test is used to map the position of the action corresponding to each late time item on the satisfaction score timeline, and the offset rate calculation is performed on the difference between the score value and the action trigger position. The action points with consistent offset direction are marked and extracted to generate a causal path set.

3. The post-sales evaluation method for enterprise marketing consulting services according to claim 2, characterized in that: The Granger causality test follows the formula: in: Indicates in The enterprise customer satisfaction score of each cycle, Indicates in The historical satisfaction rating of each cycle, Indicates the scoring sequence itself The regression influence coefficient of the order lag term on the current rating value, Indicates in The follow-up actions of enterprises implemented in the cycle Indicates the first The basic influence coefficient of the order lag action on the current rating value, Indicates the Normalized coefficient of execution intensity of the order-lag action, Indicates the The sensitivity of the enterprise customer status to the action stimulus score within a period, Indicates the The intensity of external environmental disturbances in each cycle, represents the score sensitivity adjustment coefficient, represents the external interference adjustment coefficient, represents the maximum number of lag periods considered in the satisfaction rating series, represents the maximum number of hysteresis cycles considered in the subsequent action sequence, Indicates the The random disturbance term in the period score regression model.

4. The post-sales evaluation method for enterprise marketing consulting services according to claim 1, characterized in that: The specific steps of generating the fusion matrix group are: Based on the causal path set, perform periodic positioning operations on the three fields of plan execution frequency, customer return visit days, and suggestion revision rounds. Count the number of values ​​that appear in each period and bind them to the period number to generate a behavior period detail group. Based on the behavior cycle detail group, extract the cycle number of each time node in the scoring sequence, and perform a one-to-one matching mapping between the behavior indicator value and the corresponding cycle of the scoring value to generate a scoring mapping result set; Based on the score mapping result set, the score values ​​and behavior values ​​under all cycle numbers are combined and filled into the three-dimensional coordinate matrix structure in sequence, and the results of each cycle are filled into the corresponding cell position to generate a fusion matrix group.

5. The post-sales evaluation method for enterprise marketing consulting services according to claim 1, characterized in that: The specific steps for generating the dynamic output sequence of satisfaction evolution are as follows: Based on the fusion matrix group, a scoring step value setting operation is performed and the recommended execution items, response feedback items, and project record items are divided into equal-length blocks in chronological order. A scoring group with the same order is generated for each scoring segment to generate a scoring cycle sequence. Based on the scoring cycle sequence, the score difference operation of adjacent cycles in each scoring group is calculated, and the interval classification and sign direction judgment are performed on the score change value of each segment. Then, the scoring results of each cycle are accumulated and merged according to the time period to generate a scoring state sequence; Based on the rating state sequence, a long short-term memory network is used to perform a square operation on the difference between the periodic rating value and the rating value of the same period in the actual rating sequence, and all difference terms are merged into the original rating transmission path. Value replacement is performed and the path results are updated to generate a dynamic output sequence of satisfaction evolution.

6. The post-sales evaluation method for enterprise marketing consulting services according to claim 5, characterized in that: The long short-term memory network is based on the formula: in: Indicates that in the cycle The enterprise customer satisfaction score value obtained by long short-term memory network prediction, Indicates that in the cycle The actual customer satisfaction ratings of the enterprises collected within the Represents a period The importance weight of the overall marketing service evaluation path of the enterprise, Represents a period The standard deviation of the customer rating series within the enterprise, Represents a period The relative rate of change in the frequency of interaction with internal enterprise customers is used to measure the intensity of behavioral fluctuations. The adjustment coefficient that represents the influence of the score standard deviation, The adjustment coefficient that represents the degree of influence of the intensity of behavioral fluctuations, Indicates the total number of scoring cycles in the after-sales service evaluation. It represents the total amount of error in the enterprise customer satisfaction evaluation after introducing scoring weights, scoring uncertainty and behavioral fluctuation factors.

7. The post-sales evaluation method for enterprise marketing consulting services according to claim 1, characterized in that: The specific steps of generating the impact factor group are: Based on the dynamic output sequence of satisfaction evolution, the absolute value of the score difference in each cycle is extracted, and the score offset points greater than the set interval are screened, and the offset segment cycle numbers are collected and processed to generate an offset cycle set; Based on the offset period set, the number of interview records, the number of question and answer responses, and the number of extension records are matched with the offset period number, and the number of occurrences of various indicators in the period are classified and statistically processed to generate a period parameter distribution table; Based on the periodic parameter distribution table, the repetition frequency of each group of parameters in the period in all periods is judged, and the parameter pairs that change in the same direction are merged and classified. The classification results are mapped to the corresponding dimension labels to generate an impact factor group.

8. The post-sales evaluation method for enterprise marketing consulting services according to claim 1, characterized in that: The specific steps for generating the influence intensity distribution table corresponding to the satisfaction dimension are as follows: Based on the influencing factor group, a screening operation is performed on the correspondence between the four indicators of recommended execution, cooperation, completeness, and adoption frequency and the parameter content, and the extreme values ​​of each parameter in the scoring sequence are extracted. Each scoring value is classified and marked by dimension to generate a dimension scoring interval table; Based on the dimension score interval table, statistical processing is performed on the number of occurrences in each score value interval, the number of occurrences is converted into an integer frequency value, and it is determined whether the score values ​​in the same dimension are clustered in a certain range, the highest frequency interval and corresponding parameter combination are retained, and a dimension frequency concentration group is generated; Based on the dimension frequency concentration group, the score difference calculation operation of the highest frequency score interval under each dimension is performed, the segments with differences greater than the set step size are extracted and marked with intensity level numbers, and then all segment numbers are mapped to the label sequence to generate the influence intensity distribution table corresponding to the satisfaction dimension.

9. The post-sales evaluation method for enterprise marketing consulting services according to claim 8, characterized in that: The scoring value is discrete integer data, and its value range is set to positive integers between 1 and 10, with a step size of 1, that is, each scoring value in the scoring sequence is an integer within the interval and constitutes the basic data source of the scoring sequence.

10. A post-sales evaluation system for enterprise marketing consulting services, characterized in that: The post-sales evaluation method for enterprise marketing consulting services according to any one of claims 1 to 9, wherein the system comprises: Causal Construction Module: Based on the three records of project approval response, approval feedback, and draft submission in the enterprise consulting service task record table, it calculates the time interval, screens the time-consistent items, uses the Granger causality test to calculate the synchronization offset rate between the subsequent action and the scoring curve, constructs the significance matrix, extracts the highly significant combinations, and generates a causal path set. Sequence alignment module: Based on the causal path set, extract the periodic data corresponding to plan execution, customer return visits, and suggestion revisions, match and rearrange them with the scoring sequence item by item, construct a unified time number, establish a consistent time axis structure, and obtain a unified time coordinate fusion matrix group; Trend modeling module: Based on the unified time coordinate fusion matrix group, the scoring step size is set, the recommended execution, response feedback, and project records are input, nonlinear state conversion and cycle aggregation are performed, and the long short-term memory network is used to perform error difference learning on the scoring sequence to generate a dynamic output sequence of satisfaction evolution; Factor extraction module: Based on the dynamic output sequence of satisfaction evolution, extract the interview frequency, question and answer response, and delay records within each score deviation period, screen the difference values ​​and high-frequency combinations, classify them into multi-dimensional influence sources, and construct an influence factor group; Distribution classification module: Based on the influencing factor group, the parameter values ​​under the four scoring dimensions of suggestion execution, cooperation, completeness, and adoption frequency are called to determine the frequency differences of each scoring interval, extract the mapping relationship between the strong difference segment number and the dimension label sequence, and generate the impact intensity distribution table corresponding to the satisfaction dimension.

Images