Interactive voice response path analysis method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses an interactive voice response path analysis method and device, electronic equipment and a storage medium. The method comprises the following steps: collecting a log file of the interactive voice response system, and processing the log file to obtain path characteristics of an interactive voice response path corresponding to the log file; inputting the path features into a pre-constructed space-time diagram neural network model, and predicting the loss probability of the interactive voice response path and the influence degrees of a plurality of nodes in the interactive voice response path on the loss probability according to an output result of the space-time diagram neural network model; and according to the loss probability and the plurality of influence degrees, performing funnel analysis on the interactive voice response path so as to position an abnormal node from the plurality of nodes, and analyzing an abnormal reason of the abnormal node. According to the technical scheme provided by the embodiment of the invention, the interactive voice response path can be efficiently analyzed.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of path analysis technology, and in particular to an interactive voice response path analysis method, device, electronic device, and storage medium. Background Art

[0002] There are many complex interaction nodes in the Interactive Voice Response (IVR) system. Interactive Voice Response path analysis is crucial to improving user experience.

[0003] However, the efficiency of interactive voice response path analysis is currently low and needs to be urgently addressed. Summary of the Invention

[0004] The embodiments of the present invention provide an interactive voice response path analysis method, device, electronic device and storage medium to achieve efficient analysis of the interactive voice response path.

[0005] According to one aspect of the present invention, an interactive voice response path analysis method is provided, which may include: collecting log files of an interactive voice response system, and processing the log files to obtain path characteristics of the interactive voice response path corresponding to the log files; inputting the path characteristics into a pre-constructed spatiotemporal graph neural network model, and predicting the churn probability of the interactive voice response path and the degree of influence of multiple nodes in the interactive voice response path on the churn probability based on the output results of the spatiotemporal graph neural network model; performing a funnel analysis on the interactive voice response path based on the churn probability and the multiple degrees of influence to locate abnormal nodes from multiple nodes, and analyzing the abnormal causes of the abnormal nodes.

[0006] According to another aspect of the present invention, an interactive voice response path analysis device is provided, which may include: a path feature acquisition module, used to collect log files of an interactive voice response system, and process the log files to obtain path features of the interactive voice response path corresponding to the log files; an impact degree prediction module, used to input the path features into a pre-constructed spatiotemporal graph neural network model, and predict the churn probability of the interactive voice response path and the degrees of influence of multiple nodes in the interactive voice response path on the churn probability based on the output results of the spatiotemporal graph neural network model; an abnormal cause analysis module, used to perform a funnel analysis on the interactive voice response path based on the churn probability and multiple impact degrees, so as to locate abnormal nodes from multiple nodes and analyze the abnormal causes of the abnormal nodes.

[0007] According to another aspect of the present invention, an electronic device is provided, which may include: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor implements the interactive voice response path analysis method provided by any embodiment of the present invention when executing the computer program.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, on which computer instructions are stored. The computer instructions are used to enable a processor to implement the interactive voice response path analysis method provided by any embodiment of the present invention when executed.

[0009] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the interactive voice response path analysis method provided by any embodiment of the present invention.

[0010] The technical solution of the embodiment of the present invention collects log files of an interactive voice response system and processes the log files to obtain path features of the interactive voice response path corresponding to the log files. The path features are then input into a pre-built spatiotemporal graph neural network model. Based on the output of the spatiotemporal graph neural network model, the churn probability of the interactive voice response path and the degree of influence of multiple nodes in the interactive voice response path on the churn probability are predicted, thereby achieving efficient prediction of churn probability and degree of influence. Finally, based on the churn probability and multiple degrees of influence, a funnel analysis is performed on the interactive voice response path to locate abnormal nodes from multiple nodes and analyze the abnormal causes of the abnormal nodes, thereby achieving efficient location of abnormal nodes and analysis of abnormal causes. Compared to the related schemes that rely on manual sampling for interactive voice response path analysis, which are less efficient, the above technical scheme, by inputting path features into the spatiotemporal graph neural network model to predict churn probability and degree of influence, and then performing funnel analysis on the interactive voice response path based on the churn probability and degree of influence, can automatically perform interactive voice response path analysis without relying on inefficient manual sampling, thereby achieving efficient analysis of the interactive voice response path.

[0011] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 is a flow chart of an interactive voice response path analysis method provided by an embodiment of the present invention;

[0014] Figure 2 is a schematic diagram of a path feature obtained in an interactive voice response path analysis method provided by an embodiment of the present invention;

[0015] Figure 3 is a flow chart of another interactive voice response path analysis method provided by an embodiment of the present invention;

[0016] Figure 4 is a schematic diagram of training a spatiotemporal graph neural network model in another interactive voice response path analysis method provided by an embodiment of the present invention;

[0017] Figure 5 2 is a schematic diagram of the prediction and output process of a spatiotemporal graph neural network model in another interactive voice response path analysis method provided by an embodiment of the present invention;

[0018] Figure 6 is a schematic diagram of an optional example of another interactive voice response path analysis method provided according to an embodiment of the present invention;

[0019] Figure 7 This is a structural block diagram of an interactive voice response path analysis device provided by an embodiment of the present invention;

[0020] Figure 8 The figure is a schematic diagram of the structure of an electronic device for implementing the interactive voice response path analysis method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. The situations of "target", "original", etc. are similar and will not be repeated here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0023] Before introducing the embodiments of the present invention, the application scenarios of interactive voice response path analysis, the implementation process of the current solutions adopted for interactive voice response path analysis, and the reasons for the low efficiency of interactive voice response path analysis are exemplified to better understand why the solution proposed in the embodiments of the present invention can achieve efficient analysis of interactive voice response paths.

[0024] There are numerous complex interaction nodes in the IVR system. Every choice and input by the user determines which node the interactive voice response will subsequently enter. With the rapid development and widespread application of intelligent IVR, interactive voice response path analysis is particularly important. In interactive voice response path analysis, identifying unreasonable abnormal nodes is crucial to improving user experience.

[0025] Current interactive voice response path analysis relies on inefficient manual spot checks. For example, the solution for interactive voice response path analysis relies on inefficient manual spot checks based on the historical behavior of the IVR system. However, the current interactive voice response path analysis solution lacks analysis of the impact of time and space dimensions, making it difficult to promptly detect abnormal nodes with unreasonable configurations. The analysis efficiency is low, and it is also difficult to promptly identify abnormal causes of abnormal situations such as abnormal traffic and sudden changes in conversion rates at abnormal nodes.

[0026] To address this, compared to the less efficient solutions that rely on manual sampling for interactive voice response path analysis, the embodiments of the present invention input path features into a spatiotemporal graph neural network model to predict the probability and impact of churn, and then perform a funnel analysis on the interactive voice response path based on the churn probability and impact. This eliminates the need for inefficient manual sampling and allows for automated interactive voice response path analysis, thereby achieving efficient analysis of the interactive voice response path. This will be elaborated on below.

[0027] Figure 1 This is a flow chart of a method for analyzing an interactive voice response path provided in an embodiment of the present invention. This embodiment is applicable to interactive voice response path analysis. The method can be performed by an interactive voice response path analysis device provided in an embodiment of the present invention. This device can be implemented using software and / or hardware and can be integrated into an electronic device, such as various user terminals or servers.

[0028] See also Figure 1 The method of the embodiment of the present invention specifically includes the following steps:

[0029] S110: Collect log files of the interactive voice response system, process the log files, and obtain path features of the interactive voice response paths corresponding to the log files.

[0030] Among them, the interactive voice response system can be understood as a system that guides users to complete service requests through voice menus.

[0031] The log file can be understood as a file corresponding to the log of the interactive voice response system.

[0032] In an embodiment of the present invention, log files can be collected. For example, the interaction data of users in the interactive voice response system can be collected in real time as a log file. For another example, the log capture function of the interactive voice response system can be used to capture at least one of the unique session identity identification number (Identity document, ID) of each user recorded in the original data, the node access sequence, the dwell time, the input type (such as key selection and / or voice command, etc.), and the session result (transfer to manual or hanging up, etc.) as a log file.

[0033] An interactive voice response path can be understood as the behavioral path of a user who conducts voice interaction through an interactive voice response system; an interactive voice response path can be, for example, a behavioral path from the user entering an IVR to completing a self-service (such as including nodes such as inquiring about bills and / or handling business); the number of interactive voice response paths can be at least one.

[0034] Path features can be understood as features of the interactive voice response path; path features can include at least one of the following: process flow features of the interactive voice response system, user behavior features, dwell time trend features, and path change features.

[0035] In an embodiment of the present invention, the log file can be processed to obtain path features. For example, the log file can be cleaned and structured features in the cleaned data can be extracted as path features. For another example, key information data can be extracted from each conversation in the log file, and the path features can be determined based on the extracted key information data. The key information data can include at least one of a unique conversation ID, a node, a dwell time, an input type, and a conversation result. For another example, the log file can be analyzed to analyze the user's selection path in the IVR menu (such as button selection and / or voice command, etc.). Based on the analysis results, the high-frequency paths and abnormal jump-out nodes in the interactive voice response path can be identified, and then the path features can be determined based on the high-frequency paths and abnormal jump-out nodes.

[0036] For example, see Figure 2 , the data collector in the data collection unit can be used to collect logs including user operation behaviors and system node jumps to obtain log files; the log files can be processed by the data preprocessor through standardization preprocessing to obtain path features in the form of standardized sequences.

[0037] S120. Input the path features into a pre-built spatiotemporal graph neural network model, and based on the output results of the spatiotemporal graph neural network model, predict the churn probability of the interactive voice response path and the degree of influence of multiple nodes in the interactive voice response path on the churn probability.

[0038] Among them, the Spatio-Temporal Graph Neural Network (STGNN) model can be understood as a deep learning model for processing complex data with both spatial and temporal dependencies. It can be used to predict the probability of churn and the degree of influence of multiple nodes on the churn probability. The spatio-temporal graph neural network model can, for example, jointly model spatial dependency (graph structure) and temporal dependency (time series changes) to predict behavior and / or churn probability and impact degree and other classification results, thereby achieving in-depth mining and precise optimization of user behavior and / or classification results.

[0039] The output result can be understood as the output result of the spatiotemporal graph neural network model.

[0040] A node may be understood as a node on an interactive voice response path; a node may be, for example, a menu option in an interactive voice response menu.

[0041] It is understandable that the termination of a session or the transfer of a user to a manual agent can be regarded as churn, that is, the termination of a session or the transfer of a user to a manual agent in the interactive voice response path can be regarded as the existence of churn in the interactive voice response path. Specifically, the termination of a session or the transfer of a user to a manual agent at a node can be regarded as the existence of churn at the node. On this basis, the churn probability can be understood as the probability of churn (and / or conversion) in the interactive voice response path. For example, if the log file has a total of 10 sessions corresponding to the interactive voice response path, and there is churn in 5 sessions, then the churn probability is 50%. Further, on this basis, the impact degree can be understood as the impact degree of the node on the churn probability. For example, if the log file has a total of 10 sessions corresponding to the interactive voice response path, and there is churn in 5 sessions, and 4 of the 5 churned sessions are lost at node A, then the impact degree of node A on the churn probability is 80%.

[0042] In an embodiment of the present invention, the path features can be input into the spatiotemporal graph neural network model, and based on the output results, the churn probability and the degree of influence of multiple nodes on the churn probability can be predicted.

[0043] In an embodiment of the present invention, the corresponding user behavior of the interactive voice response path can also be predicted based on the output results of the spatiotemporal graph neural network model, so as to subsequently perform funnel analysis on the interactive voice response path based on user behavior, churn probability and multiple levels of influence.

[0044] S130: Perform funnel analysis on the interactive voice response path based on the churn probability and multiple impact levels to locate abnormal nodes from multiple nodes and analyze abnormal causes of the abnormal nodes.

[0045] The abnormal node may be understood as a node with an abnormal process, and the abnormal node may be, for example, a node with a loss situation.

[0046] The cause of the abnormality can be understood as the reason why the abnormal node occurs abnormally.

[0047] In an embodiment of the present invention, a funnel analysis may be performed on the interactive voice response path based on the churn probability and multiple impact levels to locate abnormal nodes from multiple nodes and analyze the causes of the abnormalities.

[0048] It can be understood that the funnel analysis mentioned in the embodiment of the present invention is a data analysis method that combines the interactive voice response path and the conversion funnel. Funnel analysis can track the user's behavioral trajectory in the voice response process and quantify the conversion or loss at each step of the behavior. Its core goal is to identify the key path of the user from the starting point to the end point of the voice response, analyze the conversion rate of each node in the path, and thus locate the abnormal nodes of the bottleneck. The use of technical optimization tools such as funnel analysis can help improve user experience, reduce costs and increase efficiency, and enhance the overall interactive voice response path analysis effect.

[0049] In an embodiment of the present invention, after analyzing the abnormal cause of the abnormal node, an optimization node can be determined from the abnormal node according to the abnormal cause, so that the interactive voice response system can be dynamically optimized and adjusted according to the optimization node and the abnormal cause of the optimization node through methods such as artificial intelligence (AI). For example, according to the abnormal cause, the optimization node that causes the user to be transferred to a manual agent can be determined from the abnormal node, so that according to the optimization node and the abnormal cause of the optimization node, at least one of the menu design or voice guidance in the interactive voice response system can be optimized, thereby reducing the manual transfer rate, shortening the voice conversation duration, reducing invalid operations, improving self-service efficiency, reducing operating costs, and improving customer satisfaction (Customer Satisfaction Score, CSAT) and first contact resolution rate (First Contact Resolution, FCR).

[0050] In an embodiment of the present invention, interactive voice response path analysis can be performed automatically to automatically detect the data of nodes in the interactive voice response system, and abnormal nodes can be automatically located and the causes of the abnormalities can be automatically analyzed. After automatically analyzing the causes of the abnormalities of the abnormal nodes, alarm prompts can be issued for the abnormal nodes and the causes of the abnormalities, so as to timely correct the unreasonable logic in the interactive voice response system and improve the user experience.

[0051] In an embodiment of the present invention, path heat maps, dynamic funnel maps, and interactive real-time detection dashboards generated during the funnel analysis of the interactive voice response path can also be obtained to display indicator data such as node visits, conversion rates, and churn rates, so that users can clearly understand the current situation of the interactive voice response system.

[0052] In an embodiment of the present invention, the current transition probability of the interactive voice response path (the probability of the user transitioning to the path) can also be determined in real time. If the transition probability drops by more than 2 standard deviations, the interactive voice response path will be marked as a high-risk path.

[0053] The technical solution of the embodiment of the present invention collects log files of an interactive voice response system and processes the log files to obtain path features of the interactive voice response path corresponding to the log files. The path features are then input into a pre-built spatiotemporal graph neural network model. Based on the output of the spatiotemporal graph neural network model, the churn probability of the interactive voice response path and the degree of influence of multiple nodes in the interactive voice response path on the churn probability are predicted, thereby achieving efficient prediction of churn probability and degree of influence. Finally, based on the churn probability and multiple degrees of influence, a funnel analysis is performed on the interactive voice response path to locate abnormal nodes from multiple nodes and analyze the abnormal causes of the abnormal nodes, thereby achieving efficient location of abnormal nodes and analysis of abnormal causes. Compared to the related schemes that rely on manual sampling for interactive voice response path analysis, which are less efficient, the above technical scheme, by inputting path features into the spatiotemporal graph neural network model to predict churn probability and degree of influence, and then performing funnel analysis on the interactive voice response path based on the churn probability and degree of influence, can automatically perform interactive voice response path analysis without relying on inefficient manual sampling, thereby achieving efficient analysis of the interactive voice response path.

[0054] An optional technical solution processes a log file to obtain path characteristics of an interactive voice response path corresponding to the log file, including: segmenting the log file according to session identifiers to obtain candidate voice response paths; eliminating candidate voice response paths that are interrupted due to timeout from each candidate voice response path to obtain an interactive voice response path; if there are duplicate nodes in the interactive voice response path, eliminating the duplicate nodes and updating the interactive voice response path; converting the interactive voice response path into a time series to obtain path characteristics of the interactive voice response path corresponding to the log file.

[0055] The session identifier may be understood as an identifier of a voice session conducted through the interactive voice response system.

[0056] The candidate voice response paths can be understood as paths obtained by segmenting the log file according to the session identifier.

[0057] In an embodiment of the present invention, the log file may be segmented according to the session identifier to obtain candidate voice response paths. For example, each session in the log file may be segmented according to the session identifier to obtain candidate voice response paths.

[0058] In an embodiment of the present invention, candidate voice response paths that are interrupted due to timeout can be eliminated from each candidate voice response path to obtain an interactive voice response path. For example, candidate voice response paths that are interrupted due to timeout can be eliminated from each candidate voice response path, and the interactive voice response path can be determined based on each candidate voice response path obtained after elimination. For another example, candidate voice response paths that are interrupted due to timeout can be eliminated from each candidate voice response path, and each candidate voice response path obtained after elimination can be used as an interactive voice response path; and so on.

[0059] In the embodiment of the present invention, if there are duplicate nodes in the interactive voice response path, the duplicate nodes are removed and the interactive voice response path is updated.

[0060] In an embodiment of the present invention, the interactive voice response path can be converted into a time series to obtain a path feature (for example, the path feature can be reflected in the form of a time series such as [node A@t1→node B@t2→hang up]).

[0061] Exemplarily, a data cleaning module for processing missing values ​​and denoising in a data preprocessor can be used to eliminate candidate voice response paths that are interrupted due to timeouts in each candidate voice response path to obtain an interactive voice response path; if there are duplicate nodes in the interactive voice response path, the data cleaning module is used to eliminate the duplicate nodes and update the interactive voice response path; a serialization module for converting the interactive voice response path represented by the user path into a time series in the data preprocessor is used to convert the interactive voice response path into a time series, and a feature engineering module is used to extract at least one feature of the residence time, time interval and node attributes (such as whether it is a computing node and / or input node, etc.) in the time series, and the feature is standardized to obtain a path feature represented in the form of a standardized sequence.

[0062] In an embodiment of the present invention, the log file can be segmented according to the session identifier to obtain candidate voice response paths, and then the candidate voice response paths that are interrupted due to timeout are eliminated from each candidate voice response path to obtain an interactive voice response path. If there are duplicate nodes in the interactive voice response path, the duplicate nodes are eliminated and the interactive voice response path is updated. Finally, the interactive voice response path is converted into a time series to obtain path features, which can improve the reliability of the path features.

[0063] Another optional technical solution is to perform a funnel analysis on the interactive voice response path based on the churn probability and multiple influence levels to locate abnormal nodes from multiple nodes and analyze the abnormal reasons for the abnormal nodes, including: quantifying the contribution of each node to the churn probability through multiple influence levels, and locating the abnormal nodes that cause the churn of the interactive voice response path from multiple nodes based on each contribution level, and analyzing the abnormal reasons for the abnormal nodes.

[0064] The contribution degree can be understood as the contribution of the node to the churn probability, or it can be understood as the quantification of the influence of the node on the churn probability.

[0065] In the embodiment of the present invention, the contribution of each node to the churn probability may be quantified using multiple influence levels. For example, the contribution of each node may be quantified based on the influence level of the node.

[0066] In the embodiment of the present invention, abnormal nodes can be located from multiple nodes based on their contribution levels. For example, nodes with contribution levels greater than a preset contribution threshold can be identified as abnormal nodes, and the cause of the abnormality can be analyzed.

[0067] In an embodiment of the present invention, the contribution of each node to the churn probability is quantified through multiple influence degrees, and based on each contribution, abnormal nodes are located from multiple nodes, and the cause of the abnormality is analyzed, which can improve the accuracy of the located abnormal nodes.

[0068] Based on the above scheme, there is another optional technical scheme. After locating the abnormal node that causes the loss of the interactive voice response path from multiple nodes and analyzing the abnormal cause of the abnormal node, the interactive voice response path analysis method also includes: according to the abnormal node and the abnormal cause, mining the key nodes from the interactive voice response path, and obtaining the key path constructed based on each key node.

[0069] Among them, key nodes can be understood as nodes related to abnormal situations such as loss, and can also be understood as nodes related to abnormal nodes and / or abnormal causes.

[0070] It is understandable that abnormal situations such as loss may not only occur due to the abnormality of the abnormal node, but may occur due to the joint action of other nodes and the abnormal node. Other nodes that work together with the abnormal node to cause the abnormal situation can be regarded as key nodes.

[0071] The critical path can be understood as the path constructed by each key node.

[0072] In an embodiment of the present invention, key nodes can be mined from the interactive voice response path based on abnormal nodes and abnormal causes. For example, the marginal contribution of the characteristics of each node to the abnormal node can be quantified based on the abnormal nodes and abnormal causes to explain the model behavior, and key nodes related to churn can be mined from the interactive voice response path based on the obtained quantification results.

[0073] In the embodiment of the present invention, it is also possible to determine the timing factors corresponding to the key nodes that cause the loss, such as abnormal waiting, and adjust the interactive voice response system according to the timing factors.

[0074] In the embodiment of the present invention, a critical path constructed based on each key node is obtained.

[0075] In an embodiment of the present invention, after the critical path is constructed, the interactive voice response system can be adjusted according to the critical path.

[0076] In an embodiment of the present invention, key nodes are mined from the interactive voice response path based on abnormal nodes and abnormal causes, and a key path constructed based on each key node is obtained. This can obtain a more comprehensive key path related to abnormal situations such as loss, and can provide an accurate reference for adjusting the interactive voice response system.

[0077] Figure 3 It is a flowchart of another interactive voice response path analysis method provided in an embodiment of the present invention. This embodiment is optimized based on the above-mentioned technical solutions. In this embodiment, optionally, the spatiotemporal graph neural network model includes a graph structure builder, a time feature extractor, a spatial feature extractor, a spatiotemporal feature fusion device and a prediction and output device. The spatiotemporal graph neural network model processes the path features in the following manner: through the graph structure builder, a graph structure is constructed according to the path features; through the time feature extractor, feature extraction is performed on the graph structure to obtain time features; through the spatial feature extractor, feature extraction is performed on the graph structure to obtain spatial features; through the spatiotemporal feature fusion device, the time features and spatial features are fused to obtain fused features; through the prediction and output device, the probability of churn and multiple degrees of influence are predicted and output according to the fused features.

[0078] The explanations of terms that are the same as or corresponding to the above embodiments are not repeated here.

[0079] See also Figure 3 The method of this embodiment may specifically include the following steps:

[0080] S210: Collect log files of the interactive voice response system and process the log files to obtain path features of the interactive voice response paths corresponding to the log files.

[0081] S220. Input the path features into a pre-built spatiotemporal graph neural network model so that the spatiotemporal graph neural network model processes the path features through steps S2201-S2205, wherein the spatiotemporal graph neural network model includes a graph structure builder, a temporal feature extractor, a spatial feature extractor, a spatiotemporal feature fusion device, and a predictor and output device.

[0082] Among them, the graph structure builder can be understood as a component of the spatiotemporal graph neural network model that can construct a graph structure based on path features.

[0083] The temporal feature extractor can be understood as a component of a spatiotemporal graph neural network model that can extract features from a graph structure to obtain temporal features; the temporal feature extractor can, for example, use time series modeling technology to extract temporal features.

[0084] The spatial feature extractor can be understood as a component of the spatiotemporal graph neural network model that can extract features from the graph structure to obtain spatial features.

[0085] The spatiotemporal feature fuser can be understood as a component of the spatiotemporal graph neural network model that can fuse temporal features and spatial features.

[0086] The predictor and outputter can be understood as a component of a spatiotemporal graph neural network model that can predict and output the probability of churn and multiple degrees of impact based on fused features.

[0087] In an embodiment of the present invention, a training data set and a test data set can be generated in advance. For example, the data set can be divided into a training data set and a test data set according to a certain ratio; an initial neural network model is selected, for example, the initial number of convolution layers and the number of convolution kernels of the convolutional neural network are selected, and the convolutional neural network after the number of convolution layers and the number of convolution kernels are selected is used as the initial neural network model; the initial neural network model is trained through the training set to cyclically perform network architecture adjustment and parameter tuning of the initial neural network model. The adjusted parameters can be, for example, batch training size and learning rate. Selecting the optimal batch training size and learning rate is a key factor in training a model with higher prediction accuracy; the trained initial neural network model is tested and verified through the test data set. If the verification test passes, the trained initial neural network model is used as the spatiotemporal graph neural network model.

[0088] For example, see Figure 4The training dataset can be subjected to standardized data preprocessing, and the obtained features in the form of standardized sequences can be processed through a graph structure builder, a temporal feature extractor, a spatial feature extractor, and a spatiotemporal feature fusion. The fused features obtained by the spatiotemporal feature fusion are used as parameters and data for deep learning trainers for learning and training, thereby generating an optimal spatiotemporal graph neural network model. The predictors and outputters in the spatiotemporal graph neural network model can be obtained through a deep learning trainer. For example, the predictors and outputters can be part of the structure of the deep learning trainer. After training is completed, the predictors and outputters in the deep learning trainer are retained, and other parts of the deep learning trainer may or may not be retained.

[0089] It can be understood that by performing interactive voice response path analysis based on the spatiotemporal graph neural network model including a graph structure builder, a temporal feature extractor, a spatial feature extractor, a spatiotemporal feature fusion unit, and a predictor and output unit, the interactive voice response path analysis solution of the embodiment of the present invention can be widely used in fields such as traffic forecasting, weather forecasting, and social network analysis.

[0090] Optionally, the spatial feature extractor is implemented by a graph convolutional neural network; and / or, the temporal feature extractor is implemented by a temporal graph network.

[0091] Among them, Graph Neural Network (GNN) can be understood as a deep learning model for processing graph data.

[0092] A temporal graph network can be understood as a network used to process dynamic graph data with time information.

[0093] In an embodiment of the present invention, the spatial feature extractor is implemented by a graph convolutional neural network, and / or the temporal feature extractor is implemented by a temporal graph network, which can enable the spatial feature extractor to efficiently capture static structural features, and / or enable the temporal feature extractor to explicitly model temporal dependencies and support dynamic prediction and alignment.

[0094] S2201. Build a graph structure based on path features using a graph structure builder.

[0095] The graph structure can be understood as a feature in the form of a graph structure constructed by a graph structure builder based on path features.

[0096] In an embodiment of the present invention, a graph structure may be constructed according to path features through a graph structure builder.

[0097] S2202. Perform feature extraction on the graph structure through a time feature extractor to obtain time features.

[0098] Among them, the temporal feature can be understood as the characteristic information of the graph structure at the temporal level.

[0099] In an embodiment of the present invention, the temporal feature extractor can capture the dynamic evolution law of the time series data in the graph structure (such as the change in the user's stay time in the interactive voice response path), and encode the pattern of node and path changes over time into a learnable feature representation to obtain temporal features. Among them, the spatial feature extractor can, for example, use a temporal convolutional network (TCN) time series modeling to extract temporal features; the gated recurrent unit (GRU) used by the spatial feature extractor can be reset by resetting the gate r t and update gate z t Control the timing information flow, that is, through the formula r t =σ(W r [h t-1 , x t ]), z t =σ(W z [h t-1 , x t ]), and Control the timing information flow, the x t It can be understood as the input feature at the current moment (such as the node residence time), which can be understood as element-by-element multiplication; the temporal convolution network used by the spatial feature extractor uses dilated causal convolution (Dilated Causal Convolution) To capture long-term dependencies, d can be understood as an expansion factor, which can grow exponentially with the number of layers (e.g., d = 1, 2, 4). k It can be understood as the convolution kernel weight.

[0100] In an embodiment of the present invention, a time feature extractor can be used to extract features of a graph structure to obtain time features. For example, a time feature extractor can be used to extract features of at least one aspect of a graph structure, such as user behavior, system feedback, and flow, in a time dimension to obtain time features.

[0101] S2203. Perform feature extraction on the graph structure through a spatial feature extractor to obtain spatial features.

[0102] Among them, spatial features can be understood as the characteristic information of the graph structure at the spatial level.

[0103] In an embodiment of the present invention, the spatial feature extractor can aggregate multi-level neighbor information (node ​​features at each layer are updated by aggregating neighbor information) and dynamically assign interaction weights to encode complex dependencies in the graph structure into low-dimensional feature vectors, capturing the spatial neighborhood relationships of nodes in the graph structure (such as interactive voice response menu jump logic) to obtain spatial features. The spatial feature extractor can, for example, use the Graph Convolutional Network (GCN) formula To extract spatial features, It can be understood as an adjacency matrix with self-loops (A is the original adjacency matrix). It can be understood as a degree matrix, The H (l) It can be understood as the node feature matrix of the lth layer, the W (l) It can be understood as a trainable weight matrix, and σ can be understood as an activation function (such as ReLU).

[0104] In an embodiment of the present invention, a spatial feature extractor can be used to extract features of a graph structure to obtain spatial features. For example, a spatial feature extractor can be used to extract features of at least one aspect of a graph structure, such as user behavior, system feedback, and flow, in a spatial dimension to obtain spatial features.

[0105] S2204: Fusing the time feature and the space feature through the spatiotemporal feature fuser to obtain a fused feature.

[0106] Among them, the fusion feature can be understood as the feature obtained by fusing the time feature and the space feature; the fusion feature can be reflected in the form of a space-time graph.

[0107] In the embodiment of the present invention, the spatiotemporal feature fusion device can dynamically assign the importance weights of the spatiotemporal dimension (such as key nodes or key time points), deeply fuse the spatial features with the temporal features, and thus capture the spatiotemporal coupling effect in the complex interactive voice response system to obtain the fusion feature. Among them, the spatiotemporal feature fusion device can, for example, in terms of spatial attention, use the formula Calculate the dynamic association weight between nodes. t It can be understood as the node feature matrix at time step t. Q , W K It can be understood as a trainable parameter matrix; the spatiotemporal feature fusion can, for example, be used in terms of temporal attention through the formula Calculate the association weights of different time steps, where E can be understood as a time embedding matrix (such as sinusoidal position encoding); the spatiotemporal feature fusion can be, for example, in the gated fusion aspect, through the formula g = σ (W g [h spatial ||H temporal ]) and Hfused =g⊙H spatial +(1-g)⊙H temporal , achieving information flow control through learnable gating weights.

[0108] In an embodiment of the present invention, a spatiotemporal feature fuser may be used to fuse time features and space features to obtain fused features. For example, the spatiotemporal feature fuser may be used to perform weighted fusion on time features and space features so as to jointly support spatiotemporal coupling analysis through time features and space features to obtain fused features.

[0109] S2205. Through the prediction and output device, based on the fusion features, the churn probability and multiple impact levels are predicted and output.

[0110] In the embodiment of the present invention, the prediction and output device can predict and output the churn probability and multiple impact levels based on the fusion features.

[0111] For example, see Figure 5 , the data in the real dataset can be processed into path features of standardized sequences through a graph structure builder, a time feature extractor, a spatial feature extractor, a spatiotemporal feature fusion device, and a prediction and output device, and the churn probability and multiple impact levels can be predicted and output.

[0112] S230. Predict the churn probability of the interactive voice response path and the degree of influence of multiple nodes in the interactive voice response path on the churn probability based on the output results of the spatiotemporal graph neural network model.

[0113] S240: Perform funnel analysis on the interactive voice response path based on the churn probability and multiple impact levels to locate abnormal nodes from multiple nodes and analyze abnormal causes of the abnormal nodes.

[0114] The technical solution of the embodiment of the present invention is that the spatiotemporal graph neural network model includes a graph structure builder, a temporal feature extractor, a spatial feature extractor, a spatiotemporal feature fusion device and a prediction and output device. The spatiotemporal graph neural network model processes path features in the following manner: the graph structure builder constructs a graph structure based on the path features; the temporal feature extractor extracts features from the graph structure to obtain temporal features; the spatial feature extractor extracts features from the graph structure to obtain spatial features; the spatiotemporal feature fusion device fuses temporal features and spatial features to obtain fused features; the prediction and output device predicts and outputs the churn probability and multiple impact levels based on the fused features. The above solution can predict the churn probability and impact level through the spatiotemporal graph neural network model including the graph structure builder, the temporal feature extractor, the spatial feature extractor, the spatiotemporal feature fusion device and the prediction and output device, and can realize the capture of dynamically changing spatiotemporal features from the graph structure data, thereby improving the accuracy of interactive voice response path analysis based on the capture of dynamically changing spatiotemporal features.

[0115] An optional technical solution, a graph structure builder constructs a graph structure in the following manner: obtaining an interactive voice response menu represented in a tree form in an interactive voice response system, and constructing nodes and edges based on the interactive voice response menu; determining the weight of the edge based on the path characteristics, and constructing a graph structure based on the nodes, edges and weights.

[0116] Among them, the interactive voice response menu can be understood as a menu of business objects of the interactive voice response system represented in a tree-like manner; the interactive voice response menu can, for example, include a menu corresponding to an option that guides the user to select a service through voice prompts and / or key (or voice) interaction.

[0117] In an embodiment of the present invention, an interactive voice response menu may be obtained.

[0118] An edge can be understood as a legal jump between entities (nodes) of the interactive voice response menu.

[0119] In an embodiment of the present invention, nodes and edges are constructed based on the interactive voice response menu. For example, the interactive voice response menu can be abstractly constructed as a node that is endowed with static attributes (such as at least one of function type and hierarchy, etc.) and dynamic characteristics (such as at least one of real-time traffic and dwell time, etc.), and edges are constructed based on the logical relationships and / or interactive behaviors (user transfer paths) between entities of the interactive voice response menu.

[0120] The weight can be understood as the weight of the edge, specifically the weight that quantifies the strength of the relationship between the edges.

[0121] In an embodiment of the present invention, the weight of the edge can be determined according to the path characteristics. For example, the transfer frequency and average response time of the edge can be determined according to the path characteristics. The weight can be determined according to the transfer frequency and the average response time (the weight can support dynamic update) to construct a graph structure based on the nodes, edges and weights.

[0122] Exemplarily, the graph structure component may include a static graph construction module and a dynamic edge weight module. The static graph construction module can be used to obtain an interactive voice response menu, and nodes and edges can be constructed based on the interactive voice response menu, and a directed graph can be constructed based on the nodes and edges; the dynamic edge weight module can determine the weight of the edge based on the path characteristics, so as to construct a graph structure based on the directed graph and the weight.

[0123] In the embodiment of the present invention, the weight and historical mean of the right of the current edge can also be compared in real time (the historical mean can be calculated through sliding window statistics, for example), and the comparison result can be displayed.

[0124] In an embodiment of the present invention, an interactive voice response menu can be obtained, and nodes and edges can be constructed based on the interactive voice response menu. The weights of the edges are then determined based on the path characteristics, so that a graph structure can be constructed based on the nodes, edges and weights, which can improve the accuracy of the constructed graph structure.

[0125] It is understandable that the relevant interactive voice response path analysis solutions have the following problems: static path solidification, that is, the menu hierarchy and jump logic corresponding to the node are designed based on manual experience, and cannot be dynamically adjusted according to real-time user behavior, making it difficult to promptly discover and adjust unreasonable menu hierarchies and jump logic; single analysis dimension, that is, reliance on post-statistical reports (such as node churn rate and average session duration, etc.), lack of modeling capabilities for the spatiotemporal correlation of user behavior; insufficient attribution accuracy, that is, only a single node can be located, and the spatiotemporal context cannot be associated; lack of real-time response, that is, there is a significant lag in node problem discovery and optimization; cross-node correlation is ignored, that is, the churn situation of each node is analyzed independently, ignoring the global causal relationship in the path; slow response speed, that is, based on manual analysis, the process from locating the abnormal node to analyzing the cause of the abnormality is long.

[0126] In order to better understand the technical solution of the above embodiment of the present invention and to solve the problems existing in the above related interactive voice response path analysis solution, an optional example is provided here. Figure 6Interactive voice response path analysis can be achieved through a data collection unit, a spatiotemporal graph neural network model unit and a path funnel analysis unit; the data collection unit includes a data collector and a data preprocessor, the spatiotemporal graph neural network model unit includes a spatiotemporal graph neural network model, the spatiotemporal graph neural network model includes a graph structure builder, a time feature extractor, a spatial feature extractor, a spatiotemporal feature fusion unit and a prediction and output unit, and the path funnel analysis unit includes a key path miner and an attribution analyzer. Logs can be collected through a data collector to obtain log files; the data preprocessor can be used to segment the log files according to the session identifier to obtain candidate voice response paths, and the candidate voice response paths can be preprocessed according to the spatiotemporal feature extraction requirements of the spatiotemporal graph neural network model unit through keyword extraction to obtain path features in the form of standardized sequences to meet the input requirements of subsequent spatiotemporal graph neural network model units; the graph structure builder can be used to construct a graph structure based on the path features; the temporal feature extractor can be used to extract features from the graph structure to obtain temporal features; the spatial feature extraction The detector extracts features from the graph structure to obtain spatial features; the spatiotemporal feature fusion device fuses time features and spatial features to obtain fused features; the prediction and output device predicts and outputs the churn probability and multiple impact levels based on the fused features; the key path miner performs funnel analysis on the interactive voice response path based on the churn probability and multiple impact levels to locate abnormal nodes from multiple nodes and analyze the abnormal causes of the abnormal nodes; the attribution analyzer mines key nodes from the interactive voice response path based on the abnormal nodes and the abnormal causes, and obtains the key path constructed based on each key node.

[0127] Compared with related solutions with low analysis efficiency, poor real-time performance and inability to dynamically capture the spatiotemporal correlation of user behavior, the above-mentioned solution of the embodiment of the present invention proposes an interactive voice response path funnel analysis solution based on the spatiotemporal graph neural network model. It uses deep learning related technologies and introduces the spatiotemporal graph neural network model to more fully mine spatiotemporal information. Through path funnel analysis, it upgrades static and isolated statistical indicators to dynamic and global intelligent decision-making tools, which are particularly valuable in refined operations and real-time feedback scenarios. It solves the problems existing in the above-mentioned related interactive voice response path analysis solutions, realizes the funnel dynamic analysis of the interactive voice response path, high-accuracy abnormal node prediction and path optimization, thereby improving user service efficiency and user experience.

[0128] Figure 7This is a block diagram of the structure of an interactive voice response path analysis device provided in an embodiment of the present invention. The device is used to execute the interactive voice response path analysis method provided in any of the above embodiments. The device and the interactive voice response path analysis method of the above embodiments belong to the same inventive concept. For details not fully described in the embodiments of the interactive voice response path analysis device, please refer to the embodiments of the above interactive voice response path analysis method. Figure 7 The device may specifically include: a path feature acquisition module 310, an impact degree prediction module 320 and an abnormality cause analysis module 330.

[0129] Among them, the path feature acquisition module 310 is used to collect the log files of the interactive voice response system and process the log files to obtain the path features of the interactive voice response path corresponding to the log files; the impact degree prediction module 320 is used to input the path features into a pre-built spatiotemporal graph neural network model, and predict the loss probability of the interactive voice response path and the degree of influence of multiple nodes in the interactive voice response path on the loss probability based on the output results of the spatiotemporal graph neural network model; the abnormal cause analysis module 330 is used to perform funnel analysis on the interactive voice response path based on the loss probability and multiple impact degrees, so as to locate abnormal nodes from multiple nodes and analyze the abnormal causes of the abnormal nodes.

[0130] Optionally, the spatiotemporal graph neural network model includes a graph structure builder, a temporal feature extractor, a spatial feature extractor, a spatiotemporal feature fusion device and a prediction and output device. The spatiotemporal graph neural network model processes path features through the following sub-modules: a graph structure construction sub-module, which is used to construct a graph structure according to path features through the graph structure builder; a temporal feature acquisition sub-module, which is used to extract features of the graph structure through the temporal feature extractor to obtain temporal features; a spatial feature acquisition sub-module, which is used to extract features of the graph structure through the spatial feature extractor to obtain spatial features; a fusion feature acquisition sub-module, which is used to fuse temporal features and spatial features through the spatiotemporal feature fusion device to obtain fusion features; an impact degree prediction sub-module, which is used to predict and output the churn probability and multiple impact degrees based on the fusion features through the prediction and output device.

[0131] Optionally, based on the above-mentioned device, the graph structure builder constructs a graph structure through the following units: an edge construction unit, used to obtain an interactive voice response menu represented by a tree in the interactive voice response system, and construct nodes and edges based on the interactive voice response menu; a graph structure construction unit, used to determine the weight of the edge according to the path characteristics, so as to construct a graph structure based on the nodes, edges and weights.

[0132] Optionally, based on the above device, the spatial feature extractor is implemented by a graph convolutional neural network; and / or, the temporal feature extractor is implemented by a temporal graph network.

[0133] Optionally, the path feature obtaining module 310 includes: a candidate voice response path obtaining submodule, which is used to segment the log file according to the session identifier to obtain candidate voice response paths; an interactive voice response path obtaining submodule, which is used to eliminate candidate voice response paths that are interrupted due to timeout from each candidate voice response path to obtain an interactive voice response path; an interactive voice response path updating submodule, which is used to eliminate duplicate nodes if there are duplicate nodes in the interactive voice response path and update the interactive voice response path; a path feature obtaining submodule, which is used to convert the interactive voice response path into a time series to obtain the path feature of the interactive voice response path corresponding to the log file.

[0134] Optionally, the abnormal cause analysis module 330 includes: an abnormal cause analysis sub-module, which is used to quantify the contribution of each node to the loss probability through multiple influence degrees, and locate the abnormal node that causes the loss of the interactive voice response path from multiple nodes based on each contribution degree, and analyze the abnormal cause of the abnormal node.

[0135] Optionally, based on the above-mentioned device, the device may further include: a critical path obtaining module, which is used to locate the abnormal node that causes the loss of the interactive voice response path from multiple nodes, analyze the abnormal cause of the abnormal node, and then mine the key nodes from the interactive voice response path according to the abnormal node and the abnormal cause, and obtain the critical path constructed based on each key node.

[0136] The interactive voice response path analysis device provided by the embodiment of the present invention collects the log files of the interactive voice response system through a path feature acquisition module, processes the log files, and obtains the path features of the interactive voice response path corresponding to the log files. Then, the path features are input into a pre-constructed spatiotemporal graph neural network model through an impact degree prediction module, and the churn probability of the interactive voice response path and the impact degrees of multiple nodes in the interactive voice response path on the churn probability are predicted based on the output results of the spatiotemporal graph neural network model, so as to achieve efficient prediction of churn probability and image degree. Finally, the abnormal cause analysis module performs funnel analysis on the interactive voice response path based on the churn probability and multiple impact degrees, so as to locate abnormal nodes from multiple nodes and analyze the abnormal causes of the abnormal nodes, thereby achieving efficient positioning of abnormal nodes and analysis of abnormal causes. Compared with the related schemes that are less efficient and rely on manual sampling to analyze the interactive voice response path, the above-mentioned device inputs the path features into the spatiotemporal graph neural network model to predict the probability of loss and the degree of impact, and then performs a funnel analysis on the interactive voice response path based on the probability of loss and the degree of impact. It can automatically perform interactive voice response path analysis without relying on less efficient manual sampling, thereby realizing efficient analysis of the interactive voice response path.

[0137] The interactive voice response path analysis device provided in the embodiment of the present invention can execute the interactive voice response path analysis method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0138] It is worth noting that in the embodiment of the above-mentioned interactive voice response path analysis device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0139] Figure 8 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0140] like Figure 8 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0141] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0142] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the interactive voice response path analysis method.

[0143] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product that includes a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication unit 19, or installed from the storage unit 18, or installed from the ROM 12. When the computer program is executed by the processor 11, the above-mentioned functions defined in the method of the embodiment of the present invention are performed.

[0144] In some embodiments, the interactive voice response path analysis method can be implemented as a computer program that is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the interactive voice response path analysis method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the interactive voice response path analysis method in any other appropriate manner (e.g., by means of firmware).

[0145] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0146] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0147] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0148] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0149] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0150] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0151] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0152] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. An interactive voice response path analysis method, characterized in that: include: Collecting log files of an interactive voice response system and processing the log files to obtain path features of interactive voice response paths corresponding to the log files; Inputting the path features into a pre-built spatiotemporal graph neural network model, and predicting the churn probability of the interactive voice response path and the degree of influence of multiple nodes in the interactive voice response path on the churn probability based on the output of the spatiotemporal graph neural network model; According to the churn probability and the multiple impact levels, a funnel analysis is performed on the interactive voice response path to locate abnormal nodes from the multiple nodes, and the abnormal causes of the abnormal nodes are analyzed.

2. The method according to claim 1, characterized in that The spatiotemporal graph neural network model includes a graph structure builder, a temporal feature extractor, a spatial feature extractor, a spatiotemporal feature fusion unit, and a prediction and output unit. The spatiotemporal graph neural network model processes the path features in the following manner: Constructing a graph structure according to the path features by the graph structure builder; Extracting features from the graph structure using the temporal feature extractor to obtain temporal features; Extracting features from the graph structure using the spatial feature extractor to obtain spatial features; The time feature and the space feature are fused by the time-space feature fusion device to obtain a fused feature; The prediction and output device predicts and outputs the churn probability and the multiple impact levels based on the fusion features.

3. The method according to claim 2, characterized in that The graph structure builder constructs the graph structure in the following manner: Obtaining an interactive voice response menu represented in a tree form in the interactive voice response system, and constructing nodes and edges according to the interactive voice response menu; The weight of the edge is determined according to the path feature, so as to construct a graph structure according to the nodes, the edges and the weight.

4. The method according to claim 2, characterized in that The spatial feature extractor is implemented by a graph convolutional neural network; and / or the temporal feature extractor is implemented by a temporal graph network.

5. The method according to claim 1, wherein The processing of the log file to obtain a path feature of an interactive voice response path corresponding to the log file includes: Segmenting the log file according to the session identifier to obtain candidate voice response paths; Eliminating the candidate voice response paths that are interrupted due to timeout from the candidate voice response paths to obtain an interactive voice response path; If there are duplicate nodes in the interactive voice response path, removing the duplicate nodes and updating the interactive voice response path; The interactive voice response path is converted into a time series to obtain a path feature of the interactive voice response path corresponding to the log file.

6. The method according to claim 1, characterized in that The performing of a funnel analysis on the interactive voice response path based on the churn probability and the multiple impact levels to locate abnormal nodes from the multiple nodes and analyzing abnormal causes of the abnormal nodes includes: By using multiple degrees of influence, the contribution of each node to the loss probability is quantified, and based on each contribution, the abnormal node that causes the loss of the interactive voice response path is located from the multiple nodes, and the abnormal cause of the abnormal node is analyzed.

7. The method according to claim 6, characterized in that After locating the abnormal node causing the loss of the interactive voice response path from the plurality of nodes and analyzing the abnormal cause of the abnormal node, the method further includes: According to the abnormal nodes and the abnormal causes, key nodes are mined from the interactive voice response path, and a key path constructed based on each of the key nodes is obtained.

8. An interactive voice response path analysis device, characterized in that: include: A path feature obtaining module is used to collect log files of the interactive voice response system and process the log files to obtain path features of the interactive voice response paths corresponding to the log files; an impact degree prediction module, configured to input the path features into a pre-built spatiotemporal graph neural network model, and predict the churn probability of the interactive voice response path and the degree of influence of multiple nodes in the interactive voice response path on the churn probability based on the output of the spatiotemporal graph neural network model; The abnormality cause analysis module is used to perform a funnel analysis on the interactive voice response path according to the loss probability and the multiple impact levels, so as to locate abnormal nodes from the multiple nodes and analyze the abnormal causes of the abnormal nodes.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the interactive voice response path analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the interactive voice response path analysis method according to any one of claims 1 to 7 when executed.

Images