Logistics call risk prediction method, device and equipment

Abstract

The invention discloses a logistics call risk prediction method, device and equipment, the method is applied to the logistics call field, and the method comprises the steps: obtaining a target multi-source data set of a logistics call in a logistics system; dividing the target multi-source data set into data sets of a plurality of time windows by adopting a time slicing model, and extracting a task feature vector of the data set corresponding to each time window; dTW matching is carried out on task feature vectors of adjacent time windows or cross-time windows, and path morphological features of an optimal alignment path are extracted; inputting the path morphological features corresponding to each time window into a hidden Markov model for hidden state feature extraction to obtain a state transition probability matrix, an observation probability matrix and initial state distribution features; and obtaining call risk prediction results of a plurality of time windows in the future. According to the invention, the communication compliance can be accurately judged.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a logistics call risk prediction method, device and equipment. Background Art

[0002] Existing communication compliance assessment methods primarily rely on the number of calls and basic duration parameters (such as ringing duration and connection time). For example, three missed calls are automatically considered compliant, a single call exceeding a ringing threshold is considered valid, and a call that maintains a set duration after connection is considered valid. However, these methods suffer from the following drawbacks: an inability to distinguish between actual user behavior and communication failures; misclassification of non-subjective rejections, such as network anomalies and device failures, as valid calls; a lack of in-depth analysis of call termination reasons; and insufficient compliance assessment accuracy, leading to service disputes. Summary of the Invention

[0003] The present invention provides a logistics call risk prediction method, device and equipment, which can accurately judge the compliance of communications.

[0004] In one aspect, the present invention provides a method for predicting logistics call risk, the method comprising: Obtain a target multi-source dataset of logistics conversations in a logistics system; A time slicing model is used to divide the target multi-source dataset into datasets of multiple time windows, and a task feature vector of the dataset corresponding to each time window is extracted; Perform DTW matching on the task feature vectors of adjacent time windows or across time windows, calculate the optimal alignment path of the feature vectors in different time windows by accumulating the distance matrix, and extract the path morphological features of the optimal alignment path; The path morphological features corresponding to each time window are input into the hidden Markov model to extract hidden state features, and the state transition probability matrix, observation probability matrix and initial state distribution features are obtained; Multi-dimensional risk prediction is performed on the state transition probability matrix, observation probability matrix and initial state distribution characteristics to obtain call risk prediction results for multiple future time windows.

[0005] In an exemplary embodiment, obtaining a target multi-source dataset of logistics conversations in a logistics system includes: Collect the logistics system's communication logs, user behavior data, logistics regulations, and historical compliance event records through a multi-source data interface to obtain a multi-source data set; Use natural language processing technology to perform semantic analysis on the logistics regulations text, and use named entity recognition algorithm to extract compliance elements from the analysis results as structured labels, and build a compliance rule library; A rule engine and fuzzy matching technology are used to uniformly process the field formats of the multi-source dataset to obtain a target multi-source dataset.

[0006] In an exemplary embodiment, the method further comprises: Parsing the sample multi-source dataset of the sample calls according to the compliance rule library to obtain a sample risk level label for each sample call; Inputting the sample multi-source data set into a machine learning model to extract a sample user behavior pattern; and predicting a sample risk level result of the sample call based on the sample user behavior pattern; Based on the difference between the sample risk level result and the sample risk level label, the machine learning model is trained to obtain a risk level prediction model; The target multi-source dataset is input into the risk level prediction model to perform risk level prediction to obtain a target risk level of the target multi-source dataset.

[0007] In an exemplary embodiment, the method further comprises: Using a clustering algorithm to perform clustering processing on the target multi-source data set to obtain a clustering result; Building a risk assessment strategy based on the compliance rule base; Based on the risk assessment strategy and the clustering result, a target risk level of the target multi-source dataset is determined.

[0008] In an exemplary embodiment, extracting compliance elements from parsing results as structured labels using a named entity recognition algorithm and constructing a compliance rule base include: Use named entity recognition algorithm to extract compliant elements from parsing results as structured labels; Compile the structured tags into a preset network structure based on the Drools or Rete algorithm, build a compliance rule base, and then embed it into the automation engine; The method further comprises: When input data exists in the engine, the input data is evaluated based on the preset network structure to obtain an evaluation result; When the evaluation result indicates that the input data satisfies the rule corresponding to the preset network structure, an execution action corresponding to the rule is executed.

[0009] In an exemplary embodiment, compiling the structured tags into a preset network structure based on the Drools or Rete algorithm to construct a compliance rule base includes: Based on the Rete algorithm, the conditional part of the structured label corresponding rule is constructed into a tree-shaped matching network; When input data exists, the input data is propagated and matched in the network starting from the root of the tree to obtain a compliance judgment result of the input data.

[0010] In an exemplary embodiment, after obtaining the compliance judgment result of the input data, the method further includes: Define the various stages of the compliance process and state transition conditions based on the finite state machine; When it is detected that the communication behavior corresponding to the input data has compliance issues, the finite state machine transfers from the initial state to the audit state and performs an audit operation to obtain an audit result; If there are problems in the audit results, the current state will be transferred to the state of generating a rectification report.

[0011] On the other hand, a logistics call risk prediction device is provided, the device comprising: Multi-source data acquisition module, used to obtain the target multi-source data set of logistics calls in the logistics system; A task feature extraction module is used to divide the target multi-source dataset into a plurality of time window datasets using a time slicing model, and extract a task feature vector of the dataset corresponding to each time window; The path feature extraction module is used to perform DTW matching on the task feature vectors of adjacent time windows or across time windows, calculate the optimal alignment path of the feature vectors in different time windows by accumulating the distance matrix, and extract the path morphological features of the optimal alignment path; The probability feature extraction module is used to input the path morphological features corresponding to each time window into the hidden Markov model to extract hidden state features, and obtain the state transition probability matrix, observation probability matrix and initial state distribution features; The risk prediction module is used to perform multi-dimensional risk prediction on the state transition probability matrix, observation probability matrix and initial state distribution characteristics to obtain call risk prediction results for multiple time windows in the future.

[0012] In an exemplary embodiment, the multi-source data acquisition module includes: A multi-source data acquisition unit is used to collect communication logs, user behavior data, logistics regulations and historical compliance event records of the logistics system through a multi-source data interface to obtain a multi-source data set; A rule base construction unit is used to perform semantic analysis on the logistics regulations text using natural language processing technology, extract compliance elements from the analysis results as structured labels using a named entity recognition algorithm, and construct a compliance rule base; The target data acquisition unit is used to uniformly process the field formats of the multi-source data sets using a rule engine and fuzzy matching technology to obtain a target multi-source data set.

[0013] In an exemplary embodiment, the apparatus further comprises: a parsing module, configured to parse the sample multi-source dataset of the sample calls according to the compliance rule base to obtain a sample risk level label for each sample call; A sample level prediction module is configured to input the sample multi-source dataset into a machine learning model to extract a sample user behavior pattern; and predict a sample risk level result of the sample call based on the sample user behavior pattern; A model training module, configured to train the machine learning model based on the difference between the sample risk level result and the sample risk level label to obtain a risk level prediction model; The target level prediction module is used to input the target multi-source data set into the risk level prediction model to perform risk level prediction and obtain the target risk level of the target multi-source data set.

[0014] In an exemplary embodiment, the apparatus further comprises: A clustering module, configured to perform clustering processing on the target multi-source data set using a clustering algorithm to obtain a clustering result; A strategy building module, configured to build a risk assessment strategy based on the compliance rule base; A target level determination module is configured to determine a target risk level of the target multi-source data set based on the risk assessment strategy and the clustering result.

[0015] In an exemplary embodiment, the rule base construction unit includes: The label determination subunit is used to extract the compliance elements in the parsing results as structured labels using the named entity recognition algorithm; A rule base construction subunit is used to compile the structured tags into a preset network structure based on the Drools or Rete algorithm, build a compliance rule base, and then embed it into the automation engine; The device further comprises: An evaluation result determination module, configured to evaluate input data based on the preset network structure to obtain an evaluation result when input data exists in the engine; The action execution module is used to execute the execution action corresponding to the rule when the evaluation result indicates that the input data meets the rule corresponding to the preset network structure.

[0016] In an exemplary embodiment, the rule base construction sub-unit is also used to construct the conditional part of the structured tag corresponding rule into a tree-shaped matching network based on the Rete algorithm; and when there is input data, the input data is propagated and matched in the network starting from the root of the tree to obtain the compliance judgment result of the input data.

[0017] In an exemplary embodiment, the apparatus further comprises: The state definition module is used to define the various stages of the compliance process and the state transition conditions based on the finite state machine; An audit result determination module, configured to, when monitoring compliance issues in the communication behavior corresponding to the input data, transfer the finite state machine from an initial state to an audit state and perform an audit operation to obtain an audit result; The state transfer module is used to transfer the current state to the state of generating a rectification report if there are problems in the audit results.

[0018] On the other hand, an electronic device is provided, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the logistics call risk prediction method as described above.

[0019] On the other hand, a computer storage medium is provided, which stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the logistics call risk prediction method as described above.

[0020] Another aspect provides a computer program product or computer program, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to implement the logistics call risk prediction method described above.

[0021] The logistics call risk prediction method, device, and equipment provided by the present invention have the following technical effects: The present invention obtains a target multi-source dataset of logistics calls in a logistics system; uses a time slicing model to divide the target multi-source dataset into datasets of multiple time windows, and extracts the task feature vectors of the dataset corresponding to each time window; performs DTW matching on the task feature vectors of adjacent time windows or across time windows, calculates the optimal alignment path of the feature vectors in different time windows through the cumulative distance matrix, and extracts the path morphological features of the optimal alignment path; inputs the path morphological features corresponding to each time window into a hidden Markov model for hidden state feature extraction, and obtains a state transition probability matrix, an observation probability matrix, and initial state distribution features; performs multi-dimensional risk prediction on the state transition probability matrix, observation probability matrix, and initial state distribution features, and obtains call risk prediction results for multiple future time windows. The present invention breaks through the limitations of single duration judgment and integrates multi-dimensional parameters such as communication network status and device response characteristics for the first time; uses an online learning algorithm to continuously optimize the judgment model to adapt to network environment differences in different regions; thereby accurately predicting the compliance of call behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions and advantages of the embodiments of this specification or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the prior art descriptions. 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 paying any creative work.

[0023] Figure 1 This is a schematic diagram of a logistics call risk prediction system provided by an embodiment of this specification; Figure 2 This is a flow chart of a logistics call risk prediction method provided by an embodiment of this specification; Figure 3 This is a flowchart of a method for obtaining a target multi-source data set of logistics conversations in a logistics system provided by an embodiment of this specification; Figure 4 This is a flowchart of a method for determining a target risk level of a target multi-source data set provided by an embodiment of this specification; Figure 5 This is a flow chart of a method for state transfer based on a finite state machine provided in an embodiment of this specification; Figure 6 This is a structural diagram of a logistics call risk prediction device provided by an embodiment of this specification; Figure 7 This is a structural diagram of a server provided in an embodiment of this specification. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of this specification to clearly and completely describe the technical solutions in the embodiments of this specification. 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 any creative efforts shall fall within the scope of protection of the present invention.

[0025] 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 numbers used in this way are 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. 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 server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] See also Figure 1 , Figure 1 This is a schematic diagram of a logistics call risk prediction system provided by an embodiment of this specification. Figure 1 As shown, the logistics call risk prediction system can include at least a server 01 and a client 02.

[0027] Specifically, in the embodiments of this specification, the server 01 may include a standalone server, a distributed server, or a server cluster consisting of multiple servers. It may also be a cloud server that provides basic cloud computing services, such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Server 01 may include a network communication unit, a processor, and memory, among other components. Specifically, server 01 may be used to predict call risk prediction results for multiple future time windows.

[0028] Specifically, in the embodiments of this specification, the client 02 may include a physical device such as a smartphone, desktop computer, tablet computer, laptop computer, digital assistant, smart wearable device, smart speaker, in-vehicle terminal, smart TV, etc. It may also include software running on the physical device, such as a webpage provided by a service provider to a user, or an application provided by the service provider to a user. Specifically, the client 02 may be used to query call risk prediction results for a future time period online.

[0029] The following describes a logistics call risk prediction method of the present invention. Figure 2 It is a flow chart of a logistics call risk prediction method provided in the embodiment of this specification. This specification provides the method operation steps as described in the embodiment or flow chart, but it may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many orders, and does not represent the only order of execution. When the actual system or server product is executed, it can be executed in sequence or in parallel (for example, in a parallel processor or multi-threaded processing environment) according to the method shown in the embodiment or the accompanying drawings. Specifically, Figure 2 As shown, the method may include: S201: Acquire a target multi-source dataset of logistics conversations in a logistics system. In the embodiments of this specification, communication logs, user behavior data, regulatory texts (such as GDPR, ITU-T standards), historical compliance event records, etc. can be collected through a multi-source data interface as a target multi-source data set.

[0030] S203: Divide the target multi-source dataset into datasets of multiple time windows using a time slicing model, and extract a task feature vector of the dataset corresponding to each time window.

[0031] In this embodiment, a time slice model is constructed to dynamically divide communication tasks into time windows. Compliance risks within each time window are quantified using a compliance rule base. For example, metrics such as the data encryption rate and the frequency of user permission changes within a specific time period are analyzed. A continuous time series is divided into time slices of equal or varying lengths (e.g., a 5-minute / hour / day window). Task feature vectors (e.g., data fluctuation amplitude, frequency distribution, and abnormal event density) are extracted within each slice.

[0032] S205: Perform DTW matching on the task feature vectors in adjacent time windows or across time windows, calculate the optimal alignment path of the feature vectors in different time windows by accumulating the distance matrix, and extract the path morphological features of the optimal alignment path.

[0033] In the embodiments of this specification, DTW matching is performed on task feature sequences that are adjacent or span time windows: The optimal alignment path of features in different windows is calculated by accumulating the distance matrix to eliminate asynchronous interference on the time axis (such as feature offset caused by differences in task execution speeds) and path morphological features (such as path slope and local curvature) are extracted as quantitative indicators of temporal dynamic changes.

[0034] S207: Input the path morphological features corresponding to each time window into the hidden Markov model to extract hidden state features, and obtain a state transition probability matrix, an observation probability matrix, and initial state distribution features.

[0035] In this embodiment, potential risks are abstracted as hidden states of an HMM (e.g., {normal state, low-risk state, high-risk state}), while the observed states are feature sequences aligned using DTW. Parameter initialization and training utilize the Baum-Welch algorithm to iteratively optimize three sets of parameters: the state transition probability matrix (reflecting the evolution of risk states), the observation probability matrix (connecting hidden states to DTW features), and the initial state distribution. A timing constraint can also be introduced: high-risk states are only allowed to transition to states of the same or higher level to prevent irrational state jumps.

[0036] S209: Perform multi-dimensional risk prediction on the state transition probability matrix, the observation probability matrix, and the initial state distribution characteristics to obtain call risk prediction results for multiple future time windows.

[0037] In this embodiment, the Viterbi algorithm is used to decode the most likely hidden state sequence and simultaneously output the following for the next N time slices: state transition probability (e.g., the probability of entering a high-risk state in the next three windows) and the cumulative risk intensity (calculated based on the expected value of the state duration). Dynamic feedback optimization establishes a dual closed-loop mechanism: Short-term closed-loop: Real-time update of the HMM emission probability matrix using the latest observation data; Long-term closed-loop: Periodic retraining of the DTW alignment strategy and HMM state space partitioning. Historical logistics calls in the logistics system can be used to predict whether future calls will be risky.

[0038] In the embodiments of this specification, Figure 3 As shown, the target multi-source data set of logistics conversations in the logistics system is obtained, including: S20101: Collect the logistics system's communication logs, user behavior data, logistics regulations, and historical compliance event records through a multi-source data interface to obtain a multi-source dataset. S20103: Using natural language processing technology to perform semantic analysis on the logistics regulations text, and using a named entity recognition algorithm to extract compliance elements from the analysis results as structured labels, and construct a compliance rule library; S20105: Using a rule engine and fuzzy matching technology to uniformly process the field formats of the multi-source dataset to obtain a target multi-source dataset.

[0039] In the examples of this specification, communication logs, user behavior data, regulatory texts (such as GDPR and ITU-T standards), and historical compliance event records are collected through a multi-source data interface. Natural language processing (NLP) technology is used to parse regulatory clauses and extract key compliance requirements (such as data minimization and privacy protection) as structured tags.

[0040] NLP model: Use the BERT or RoBERTa pre-trained model to perform semantic analysis of regulatory texts, combined with named entity recognition (NER) to extract compliance elements (such as "sensitive information type" and "data retention period").

[0041] Data standardization algorithm: Based on the rule engine and fuzzy matching technology (such as Levenshtein distance), it unifies the field formats of different data sources to ensure compatibility for subsequent analysis.

[0042] In the embodiments of this specification, Figure 4 As shown, the method further includes: S401: Parsing a sample multi-source dataset of sample calls according to the compliance rule library to obtain a sample risk level label for each sample call; S403: Inputting the sample multi-source dataset into a machine learning model to extract a sample user behavior pattern; and predicting a sample risk level result of the sample call based on the sample user behavior pattern. S405: Training the machine learning model based on the difference between the sample risk level result and the sample risk level label to obtain a risk level prediction model; S407: Inputting the target multi-source dataset into the risk level prediction model to perform risk level prediction to obtain a target risk level of the target multi-source dataset.

[0043] In an embodiment of this specification, a machine learning algorithm is used to predict future compliance risks, for example, by identifying potential privacy leaks or violations through user behavior patterns. A supervised learning model employs a support vector machine (SVM) or deep neural network (DNN) to classify user behavior compliance, with training data labeled with historical violation cases. The sample multi-source dataset and the target multi-source dataset share the same categories, and the sample risk level labels can include high, medium, and low risk. Multiple risk levels can also be set based on actual needs. The sample multi-source dataset is input into a machine learning model to extract sample user behavior patterns. Based on the sample user behavior patterns, a sample risk level result for the sample call is predicted. Target loss data is determined based on the difference between the sample risk level result and the sample risk level label. The machine learning model parameters are adjusted based on the target loss data until training termination conditions are met. The machine learning model at the end of training is then designated as the risk level prediction model. The risk level prediction model can include a time slicing model, a hidden Markov model, and a risk assessment model. The risk assessment model performs multidimensional risk prediction on the state transition probability matrix, the observation probability matrix, and the initial state distribution characteristics, generating call risk prediction results for multiple future time windows. Multi-dimensional classifiers can be built based on Random Forest or XGBoost, taking inputs such as historical violations and real-time operation logs and outputting a risk level (high / medium / low). Reinforcement Learning (RL): Intelligent agents can be designed to dynamically adjust compliance policies, for example, optimizing data encryption strength or access control rules based on real-time risk feedback.

[0044] In the embodiment of this specification, the method further includes: Using a clustering algorithm to perform clustering processing on the target multi-source data set to obtain a clustering result; Building a risk assessment strategy based on the compliance rule base; Based on the risk assessment strategy and the clustering result, a target risk level of the target multi-source dataset is determined.

[0045] In the embodiments of this specification, a clustering algorithm (such as K-means or DBSCAN) is used to identify abnormal communication patterns, such as high-frequency data exports or access to sensitive resources during non-working hours.

[0046] In the embodiments of this specification, the use of a named entity recognition algorithm to extract compliance elements from the parsing results as structured labels and to construct a compliance rule base includes: Use named entity recognition algorithm to extract compliant elements from parsing results as structured labels; Compile the structured tags into a preset network structure based on the Drools or Rete algorithm, build a compliance rule base, and then embed it into the automation engine; The method further comprises: When input data exists in the engine, the input data is evaluated based on the preset network structure to obtain an evaluation result; When the evaluation result indicates that the input data satisfies the rule corresponding to the preset network structure, an execution action corresponding to the rule is executed.

[0047] In the embodiments of this specification, regulatory requirements are converted into executable logical rules (e.g., "unauthorized cross-border data transmission is prohibited") and embedded in an automated execution engine. Specifically, a dynamic rule base based on the Drools or Rete algorithms is constructed to support real-time matching of communication behaviors with compliance clauses. A workflow engine employs a finite state machine (FSM) or Petri net model to manage compliance processes, such as triggering automated audits or generating rectification reports.

[0048] In the embodiments of this specification, Drools is a Java-based open source rule engine that allows developers to use a declarative rule language to define business rules, separating business logic from application code, thereby improving the maintainability and flexibility of the system.

[0049] How it works: Drools uses an improved version of the Rete algorithm to implement rule matching and execution. It compiles rules into an efficient network structure, and when data enters the engine, it is matched within this network. Drools evaluates the input data according to the rule conditions. When the data meets the rule conditions, the corresponding rule action is executed.

[0050] Role in the Compliance Rules Engine: Drools makes it easy to define and manage various regulatory requirements in the form of rules. For example, a regulatory requirement such as "prohibiting unauthorized cross-border data transmission" can be converted into a Drools rule. When the system detects data transmission, the Drools engine performs real-time matching against the defined rules to determine compliance.

[0051] In the embodiment of this specification, the structured tags are compiled into a preset network structure based on the Drools or Rete algorithm to build a compliance rule base, including: Based on the Rete algorithm, the conditional part of the structured label corresponding rule is constructed into a tree-shaped matching network; When input data exists, the input data is propagated and matched in the network starting from the root of the tree to obtain a compliance judgment result of the input data.

[0052] In the embodiments of this specification, the Rete algorithm was proposed by Charles Forgy in 1979. It is an efficient pattern matching algorithm for implementing a production rule system and is one of the core algorithms of a rule engine.

[0053] How it works: The Rete algorithm constructs the conditional components of a rule into a tree-like matching network. When facts (data) enter the system, they are propagated and matched throughout the network, starting from the root of the tree. By sharing nodes and other methods, the algorithm avoids duplicate matching of facts, significantly improving matching efficiency. Conditions shared by multiple rules are only calculated once, saving computing resources and time.

[0054] Role in the Compliance Rules Engine: When building a dynamic rule base, the Rete algorithm enables rapid, real-time matching of communication behaviors against compliance clauses. It efficiently organizes and manages a large number of compliance rules. When new communication behaviors occur, it can quickly determine whether they comply with defined compliance clauses, providing an efficient matching mechanism for compliance checks.

[0055] In the embodiment of this specification, after obtaining the compliance judgment result of the input data, as follows Figure 5 As shown, the method further includes: S501: Define the various stages of the compliance process and state transition conditions based on the finite state machine; S503: When it is detected that the communication behavior corresponding to the input data has a compliance issue, the finite state machine transitions from the initial state to the audit state, and performs an audit operation to obtain an audit result; S505: If there are problems in the audit results, the current state is transferred to the state of generating a rectification report.

[0056] In the embodiments of this specification, a finite state machine (FSM) is an abstract computational model consisting of a finite number of states, transitions between states, and events that trigger these transitions. At any given moment, the FSM is in a specific state. When an event is received, it transitions to another state according to predefined rules based on the current state and the event type.

[0057] How it works: An FSM describes the behavior of a system by defining a set of states, a set of events, a state transition function, and an initial state. When an event occurs, the state transition function determines the next state based on the current state and the event. For example, in a compliance process, the initial state might be "Unaudited." When the "Audit Start" event is triggered, the state might transition to "Auditing." If the audit passes, the state transitions to "Compliant" upon receiving the "Audit Completed" event. If the audit fails, the state transitions to "Remediation in Progress," and so on.

[0058] Role in the Compliance Workflow Engine: In compliance process management, the FSM can clearly define the various stages and state transitions of the compliance process. For example, if a communication behavior is detected that may have compliance issues, the FSM can transition from the initial state to the "Automatic Audit" state to perform the audit. If issues are found, the FSM will transition to the "Generate Correction Report" state, ensuring that the compliance process proceeds in an orderly manner according to the predetermined logic.

[0059] In the embodiments of this specification, Petri nets are a graphical modeling tool used to describe parallel, concurrent, and asynchronous behavior in a system. Proposed by Carl Adam Petri in 1962, they consist of places, transitions, arcs, and tokens. Places represent system states or resources, transitions represent system events or operations, arcs connect places and transitions, and tokens represent resource quantities or status information within a place.

[0060] How it works: Petri nets operate based on the flow of tokens. When all input places of a transition have sufficient tokens, the transition is triggered. This triggers the corresponding number of tokens removed from the input places and adds new tokens to the output places, thereby changing the system state. For example, in a compliance process, the input places of the "Audit Completed" transition might be the "Auditing" place and the "Audit Passed" conditional place. When both places have tokens, the "Audit Completed" transition is triggered, changing the system state from "Auditing" to a subsequent state such as "Compliant" or "Remediation Underway."

[0061] Role in Compliance Workflow Engines: In compliance process management, Petri net models can more intuitively describe complex compliance processes, especially when concurrent and parallel operations are involved. They can clearly demonstrate the relationships and dependencies between various compliance steps. For example, when multiple compliance check tasks are executed in parallel, Petri nets can accurately describe the synchronization and coordination relationships between these tasks, ensuring the correct execution and management of the compliance process.

[0062] In the embodiments of this specification, compliance status can be displayed through a visual dashboard (such as a risk heat map and real-time alerts), and the model can be continuously optimized through a feedback mechanism. Visualization Model: Generate interactive charts based on D3.js or ECharts, such as Sankey diagrams to display data flow and compliance nodes. Dynamic Optimization Algorithm: Use Bayesian Optimization to adjust model hyperparameters or verify strategy effectiveness through A / B testing.

[0063] This technical solution provides a communication compliance assessment method based on multi-dimensional call feature analysis. It collects communication logs, user behavior data, regulatory texts, and historical compliance event records through multi-source data interfaces. Natural language processing (NLP) technology is used to parse regulatory clauses and extract key compliance requirements as structured labels. A pre-trained BERT or RoBERTa model is used to perform semantic parsing of regulatory texts, combined with named entity recognition (NER) to extract compliance elements. Based on a rule engine and fuzzy matching technology, the field formats of different data sources are unified. A time slice model is constructed to dynamically divide communication tasks into time windows, and compliance risks are quantified within each time period using a compliance rule base. For example, metrics such as the data encryption rate and the frequency of user permission changes within a specific time period are analyzed. This allows for accurate assessment of communication compliance.

[0064] From the technical solutions provided by the above embodiments of this specification, it can be seen that the embodiments of this specification obtain a target multi-source dataset of logistics calls in a logistics system; use a time slicing model to divide the target multi-source dataset into datasets of multiple time windows, and extract the task feature vectors of the dataset corresponding to each time window; perform DTW matching on the task feature vectors of adjacent time windows or across time windows, calculate the optimal alignment path of the feature vectors in different time windows through the cumulative distance matrix, and extract the path morphological features of the optimal alignment path; input the path morphological features corresponding to each time window into the hidden Markov model for hidden state feature extraction to obtain the state transition probability matrix, observation probability matrix and initial state distribution features; perform multi-dimensional risk prediction on the state transition probability matrix, observation probability matrix and initial state distribution features to obtain call risk prediction results for multiple time windows in the future. The present invention breaks through the limitations of single duration judgment and integrates multi-dimensional parameters such as communication network status and device response characteristics for the first time; uses an online learning algorithm to continuously optimize the judgment model to adapt to the differences in network environments in different regions; thereby accurately predicting the compliance of call behavior.

[0065] The embodiment of this specification also provides a logistics call risk prediction device, such as Figure 6 As shown, the device includes: The multi-source data acquisition module 610 is used to acquire a target multi-source data set of logistics calls in the logistics system; A task feature extraction module 620 is configured to divide the target multi-source dataset into a plurality of time window datasets using a time slicing model, and extract a task feature vector of the dataset corresponding to each time window; The path feature extraction module 630 is used to perform DTW matching on the task feature vectors of adjacent time windows or across time windows, calculate the optimal alignment path of the feature vectors in different time windows by accumulating the distance matrix, and extract the path morphological features of the optimal alignment path; Probabilistic feature extraction module 640 is used to input the path morphological features corresponding to each time window into the hidden Markov model to extract hidden state features, and obtain the state transition probability matrix, observation probability matrix and initial state distribution features; The risk prediction module 650 is used to perform multi-dimensional risk prediction on the state transition probability matrix, the observation probability matrix and the initial state distribution characteristics to obtain call risk prediction results for multiple future time windows.

[0066] In an exemplary embodiment, the multi-source data acquisition module includes: A multi-source data acquisition unit is used to collect communication logs, user behavior data, logistics regulations and historical compliance event records of the logistics system through a multi-source data interface to obtain a multi-source data set; A rule base construction unit is used to perform semantic analysis on the logistics regulations text using natural language processing technology, extract compliance elements from the analysis results as structured labels using a named entity recognition algorithm, and construct a compliance rule base; The target data acquisition unit is used to uniformly process the field formats of the multi-source data sets using a rule engine and fuzzy matching technology to obtain a target multi-source data set.

[0067] In an exemplary embodiment, the apparatus further comprises: a parsing module, configured to parse the sample multi-source dataset of the sample calls according to the compliance rule base to obtain a sample risk level label for each sample call; A sample level prediction module is configured to input the sample multi-source dataset into a machine learning model to extract a sample user behavior pattern; and predict a sample risk level result of the sample call based on the sample user behavior pattern; A model training module, configured to train the machine learning model based on the difference between the sample risk level result and the sample risk level label to obtain a risk level prediction model; The target level prediction module is used to input the target multi-source data set into the risk level prediction model to perform risk level prediction and obtain the target risk level of the target multi-source data set.

[0068] In an exemplary embodiment, the apparatus further comprises: A clustering module, configured to perform clustering processing on the target multi-source data set using a clustering algorithm to obtain a clustering result; A strategy building module, configured to build a risk assessment strategy based on the compliance rule base; A target level determination module is configured to determine a target risk level of the target multi-source data set based on the risk assessment strategy and the clustering result.

[0069] In an exemplary embodiment, the rule base construction unit includes: The label determination subunit is used to extract the compliance elements in the parsing results as structured labels using the named entity recognition algorithm; A rule base construction subunit is used to compile the structured tags into a preset network structure based on the Drools or Rete algorithm, build a compliance rule base, and then embed it into the automation engine; The device further comprises: An evaluation result determination module, configured to evaluate input data based on the preset network structure to obtain an evaluation result when input data exists in the engine; The action execution module is used to execute the execution action corresponding to the rule when the evaluation result indicates that the input data meets the rule corresponding to the preset network structure.

[0070] In an exemplary embodiment, the rule base construction sub-unit is also used to construct the conditional part of the structured tag corresponding rule into a tree-shaped matching network based on the Rete algorithm; and when there is input data, the input data is propagated and matched in the network starting from the root of the tree to obtain the compliance judgment result of the input data.

[0071] In an exemplary embodiment, the apparatus further comprises: The state definition module is used to define the various stages of the compliance process and the state transition conditions based on the finite state machine; An audit result determination module, configured to, when monitoring compliance issues in the communication behavior corresponding to the input data, transfer the finite state machine from an initial state to an audit state and perform an audit operation to obtain an audit result; The state transfer module is used to transfer the current state to the state of generating a rectification report if there are problems in the audit results.

[0072] The device and method embodiments in the device embodiments are based on the same inventive concept.

[0073] An embodiment of this specification provides an electronic device, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the logistics call risk prediction method provided in the above method embodiment.

[0074] An embodiment of the present invention also provides a computer storage medium, which can be set in a terminal to store at least one instruction or at least one program related to a logistics call risk prediction method in a method embodiment. The at least one instruction or at least one program is loaded and executed by the processor to implement the logistics call risk prediction method provided by the above method embodiment.

[0075] Embodiments of the present invention further provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to implement the logistics call risk prediction method provided in the above method embodiment.

[0076] Optionally, in the embodiments of this specification, the storage medium may be located in at least one of the multiple network servers in the computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard drive, a magnetic disk, or an optical disk, among other media capable of storing program code.

[0077] The memory described in the embodiments of this specification can be used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required for functions, etc.; the data storage area can store data created according to the use of the device, etc. In addition, the memory may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory may also include a memory controller to provide the processor with access to the memory.

[0078] The logistics call risk prediction method embodiment provided in the embodiments of this specification can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Taking running on a server as an example, Figure 7This is a hardware structure diagram of a server for a logistics call risk prediction method provided by an embodiment of this specification. Figure 7 As shown, the server 700 may vary significantly due to different configurations or performance, and may include one or more central processing units (CPUs) 710 (CPUs 710 may include, but are not limited to, processing devices such as microprocessors (MCUs) or programmable logic devices (FPGAs), memory 730 for storing data, and one or more storage media 720 (e.g., one or more mass storage devices) for storing applications 723 or data 722. The memory 730 and storage media 720 may be either transient or persistent storage. The program stored in the storage medium 720 may include one or more modules, each of which may include a series of instruction operations on the server. Furthermore, the CPU 710 may be configured to communicate with the storage medium 720 to execute the series of instruction operations in the storage medium 720 on the server 700. The server 700 may also include one or more power supplies 760, one or more wired or wireless network interfaces 750, one or more input and output interfaces 740, and / or one or more operating systems 721, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0079] The input / output interface 740 can be used to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the server 700. In one embodiment, the input / output interface 740 may include a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the input / output interface 740 may be a radio frequency (RF) module for wireless communication with the Internet.

[0080] It can be understood by those skilled in the art that Figure 7 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 7 More or fewer components than shown, or with Figure 7 Different configurations shown.

[0081] As can be seen from the embodiments of the logistics call risk prediction method, device, electronic device or storage medium provided by the present invention, the present invention obtains a target multi-source data set for logistics calls in a logistics system; uses a time slicing model to divide the target multi-source data set into data sets of multiple time windows, and extracts the task feature vector of the data set corresponding to each time window; performs DTW matching on the task feature vectors of adjacent time windows or across time windows, calculates the optimal alignment path of the feature vectors in different time windows through the cumulative distance matrix, and extracts the path morphological features of the optimal alignment path; inputs the path morphological features corresponding to each time window into a hidden Markov model for hidden state feature extraction to obtain a state transition probability matrix, an observation probability matrix and an initial state distribution feature; performs multi-dimensional risk prediction on the state transition probability matrix, the observation probability matrix and the initial state distribution feature to obtain call risk prediction results for multiple future time windows. The present invention breaks through the limitations of single duration judgment and integrates multi-dimensional parameters such as communication network status and device response characteristics for the first time; uses an online learning algorithm to continuously optimize the judgment model to adapt to the differences in network environments in different regions; thereby accurately predicting the compliance of call behavior.

[0082] It should be noted that the order in which the embodiments of this specification are presented is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions are of specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0083] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, equipment, and storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant portions, refer to the descriptions of the method embodiments.

[0084] Those skilled in the art will understand that all or part of the steps of implementing the above embodiments may be accomplished by hardware, or by a program instructing the relevant hardware to accomplish the steps. The program may be stored in a computer storage medium, and the above-mentioned storage medium may be a read-only memory, a disk, or an optical disk, etc.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A logistics call risk prediction method, characterized in that: The method comprises: Obtain a target multi-source dataset of logistics conversations in a logistics system; A time slicing model is used to divide the target multi-source dataset into datasets of multiple time windows, and a task feature vector of the dataset corresponding to each time window is extracted; Perform DTW matching on the task feature vectors of adjacent time windows or across time windows, calculate the optimal alignment path of the feature vectors in different time windows by accumulating the distance matrix, and extract the path morphological features of the optimal alignment path; The path morphological features corresponding to each time window are input into the hidden Markov model to extract hidden state features, and the state transition probability matrix, observation probability matrix and initial state distribution features are obtained; Multi-dimensional risk prediction is performed on the state transition probability matrix, observation probability matrix and initial state distribution characteristics to obtain call risk prediction results for multiple future time windows.

2. The method according to claim 1, characterized in that The method of obtaining a target multi-source data set of logistics conversations in a logistics system includes: Collect the logistics system's communication logs, user behavior data, logistics regulations, and historical compliance event records through a multi-source data interface to obtain a multi-source data set; Use natural language processing technology to perform semantic analysis on the logistics regulations text, and use named entity recognition algorithm to extract compliance elements from the analysis results as structured labels, and build a compliance rule library; A rule engine and fuzzy matching technology are used to uniformly process the field formats of the multi-source dataset to obtain a target multi-source dataset.

3. The method according to claim 2, characterized in that The method further comprises: Parsing the sample multi-source dataset of the sample calls according to the compliance rule library to obtain a sample risk level label for each sample call; Inputting the sample multi-source data set into a machine learning model to extract a sample user behavior pattern; and predicting a sample risk level result of the sample call based on the sample user behavior pattern; Based on the difference between the sample risk level result and the sample risk level label, the machine learning model is trained to obtain a risk level prediction model; The target multi-source dataset is input into the risk level prediction model to perform risk level prediction to obtain a target risk level of the target multi-source dataset.

4. The method according to claim 2, characterized in that The method further comprises: Using a clustering algorithm to perform clustering processing on the target multi-source data set to obtain a clustering result; Building a risk assessment strategy based on the compliance rule base; Based on the risk assessment strategy and the clustering result, a target risk level of the target multi-source dataset is determined.

5. The method according to claim 1, characterized in that The use of named entity recognition algorithms to extract compliance elements from parsing results as structured labels and to construct a compliance rule base includes: Use named entity recognition algorithm to extract compliant elements from parsing results as structured labels; Compile the structured tags into a preset network structure based on the Drools or Rete algorithm, build a compliance rule base, and then embed it into the automation engine; The method further comprises: When input data exists in the engine, the input data is evaluated based on the preset network structure to obtain an evaluation result; When the evaluation result indicates that the input data satisfies the rule corresponding to the preset network structure, an execution action corresponding to the rule is executed.

6. The method according to claim 5, characterized in that The structured tags are compiled into a preset network structure based on the Drools or Rete algorithm to build a compliance rule base, including: Based on the Rete algorithm, the conditional part of the structured label corresponding rule is constructed into a tree-shaped matching network; When input data exists, the input data is propagated and matched in the network starting from the root of the tree to obtain a compliance judgment result of the input data.

7. The method according to claim 6, characterized in that After obtaining the compliance judgment result of the input data, the method further includes: Define the various stages of the compliance process and state transition conditions based on the finite state machine; When it is detected that the communication behavior corresponding to the input data has compliance issues, the finite state machine transfers from the initial state to the audit state and performs an audit operation to obtain an audit result; If there are problems in the audit results, the current state will be transferred to the state of generating a rectification report.

8. A logistics call risk prediction device, characterized in that: The device comprises: Multi-source data acquisition module, used to obtain the target multi-source data set of logistics calls in the logistics system; A task feature extraction module is used to divide the target multi-source dataset into a plurality of time window datasets using a time slicing model, and extract a task feature vector of the dataset corresponding to each time window; The path feature extraction module is used to perform DTW matching on the task feature vectors of adjacent time windows or across time windows, calculate the optimal alignment path of the feature vectors in different time windows by accumulating the distance matrix, and extract the path morphological features of the optimal alignment path; The probability feature extraction module is used to input the path morphological features corresponding to each time window into the hidden Markov model to extract hidden state features, and obtain the state transition probability matrix, observation probability matrix and initial state distribution features; The risk prediction module is used to perform multi-dimensional risk prediction on the state transition probability matrix, observation probability matrix and initial state distribution characteristics to obtain call risk prediction results for multiple time windows in the future.

9. An electronic device, characterized in that: The device includes: a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the logistics call risk prediction method as described in any one of claims 1-7.

10. A computer storage medium, characterized in that The computer storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the logistics call risk prediction method as described in any one of claims 1-7.

Images