Mobile phone function test method and device, equipment and storage medium

Abstract

The invention relates to a mobile phone function test method and device, equipment and a storage medium, and the method comprises the following steps: carrying out the feature analysis of a running log in an application program of a target mobile phone, and obtaining an original feature data set; performing multi-dimensional feature extraction on the original feature data set to obtain an application behavior feature sequence; performing dynamic function state analysis on the target mobile phone based on the application behavior feature sequence to obtain a function state transition graph; identifying whether an abnormal mode exists in the functional state transition map or not, and if yes, performing causal relationship inference on the functional state transition map based on the abnormal mode to obtain a fault point list; and performing static analysis on each potential fault in the fault point list to obtain a function optimization suggestion. The technical problems that a traditional mobile phone function test method often depends on manual operation and predefined test cases, time and labor are consumed, and all possible application scenes are difficult to fully cover are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of mobile phone functions, and particularly relates to a method, device, equipment and storage medium for testing mobile phone functions. Background Art

[0002] With the rapid development and popularization of smart phone technology, the complexity of mobile phone functions has also increased. When manufacturers launch new models or update existing devices, they must ensure the stability, compatibility and security of various applications and services. However, traditional mobile phone function testing methods often rely on manual operations and predefined test cases, which are not only time-consuming and laborious, but also difficult to comprehensively cover all possible application scenarios, especially for those dynamically changing and unexpected behavior patterns. In the face of an increasingly complex software environment and diverse user needs, the effectiveness of traditional testing means has been greatly challenged, so a more intelligent and automated solution is needed to improve testing efficiency and accuracy.

[0003] A key problem in the research background is the limited ability of current testing methods to detect abnormal behaviors. In actual use, application programs may exhibit abnormal behaviors due to various factors, such as resource competition, memory leaks or conflicts with other applications. These abnormal situations are difficult to capture through conventional testing because they are usually not caused by a single reason, but the result of the combined action of multiple factors. In addition, even if abnormal patterns can be identified, existing analysis tools lack the ability to deeply explore the root causes of problems, thus affecting the speed and effectiveness of troubleshooting. To improve the ability to handle such complex problems, it is necessary to introduce more advanced data analysis techniques and algorithms to enhance abnormal detection and causal relationship inference in the testing process.

[0004] To solve the above problems, this study proposes a mobile phone function testing method based on feature parsing and recurrent neural network (RNN). This method first extracts raw feature data from the logs of application programs, and obtains the behavioral feature sequence of application programs through multi-dimensional feature extraction to construct a detailed dynamic function state map. Then, the RNN model is used to identify abnormal patterns in the map and further perform causal relationship inference, and finally generate a list containing potential fault points and their optimization suggestions. This method not only improves the automation degree of testing, but also enhances the understanding and processing ability of complex abnormal patterns, provides a more accurate and effective fault location tool for developers, helps to speed up the product iteration cycle and improve the user experience. Summary of the Invention

[0005] The main object of the present invention is to provide a method, device, equipment and storage medium for mobile phone function testing, which solves the technical problems that traditional mobile phone function testing methods often rely on manual operations and predefined test cases, which are not only time-consuming and laborious, but also difficult to comprehensively cover all possible application scenarios.

[0006] To achieve the above object, the present invention provides a method for mobile phone function testing, including the following steps: Perform feature parsing on the running logs in the application of the target mobile phone to obtain an original feature data set; Perform multi-dimensional feature extraction on the original feature data set to obtain an application behavior feature sequence; Based on the application behavior feature sequence, perform dynamic function state analysis on the target mobile phone to obtain a function state transition map; Use a recurrent neural network to identify whether there is an abnormal pattern in the function state transition map. If so, perform causal relationship inference on the function state transition map based on the abnormal pattern to obtain a list of fault points; Perform static analysis on each potential fault in the list of fault points to obtain function optimization suggestions.

[0007] Further, the performing feature parsing on the running logs in the application of the target mobile phone to obtain an original feature data set includes: Perform session segmentation on the application running logs through a log segmentation algorithm to obtain a sequence of log segments; Perform syntax structure parsing on the sequence of log segments based on abstract syntax tree technology to obtain syntax feature vectors; Perform pattern matching on the syntax feature vectors through a regular expression engine to obtain a set of log templates; Perform topic extraction on the set of log templates to obtain a log topic distribution; wherein, the log topic distribution includes function categories and state transition information; Construct an index for the log topic distribution through dynamic trie technology to obtain a hierarchical feature index; wherein, the hierarchical feature index includes feature paths, keyword weights, and association metrics; Perform parallel processing on the hierarchical feature index through a preset MapReduce framework to obtain an original feature data set.

[0008] Further, the performing multi-dimensional feature extraction on the original feature data set to obtain an application behavior feature sequence includes: Perform user interaction event analysis on the original feature data set to obtain a user behavior sequence; Use a preset wavelet transform technology to perform multi-scale decomposition on the user behavior sequence to obtain time feature vectors; Perform multi-dimensional extraction on the time feature vector to obtain behavior pattern features; Classify the behavior pattern features to obtain behavior classification features; wherein, the behavior classification features include click frequency features, sliding mode features, long-press behavior features, browsing depth features, and interaction density features; Through a preset time series algorithm, perform time series pattern mining on the behavior classification features to obtain an application behavior feature sequence.

[0009] Furthermore, perform dynamic functional state analysis on the target mobile phone based on the application behavior feature sequence to obtain a functional state transition map, including: Perform state space mapping on the application behavior feature sequence to obtain a discretized representation of the functional state; Adopt the random walk theory and transfer entropy algorithm to perform transfer probability modeling on the discretized state representation to obtain a state transfer probability map; Through a preset graph convolutional network, perform node feature encoding on the state transfer probability map to obtain an enhanced state representation vector; Based on the enhanced state representation vector, perform dynamic functional state analysis on the target mobile phone to obtain a weighted state transfer graph; Based on the weighted state transfer graph, capture the potential correlation between the discretized representations of the functional states of the target mobile phone to obtain a state correlation vector; Perform spectral clustering analysis on the state correlation vector to obtain a functional state clustering set; Adopt a preset small-world network model to perform topological reconstruction on the functional state clustering set to obtain a state transfer network; Perform dynamic path analysis on the state transfer network to obtain a functional state transition map; wherein, the functional state transition map includes nodes and edges, the nodes represent functional states, and the edges represent the transition relationships between functional states.

[0010] Furthermore, use a recursive neural network to identify whether there is an abnormal pattern in the functional state transition map. If so, perform abnormal causality inference on the target mobile phone based on the abnormal pattern to obtain a list of fault points, including: Adopt a deep neural network algorithm to perform a deep scan on the functional state transition map to obtain a deep scan result; Judge whether there is an abnormality in the deep scan result. If so, use a preset Bayesian network inference algorithm to perform abnormal pattern inference on the deep scan result to obtain an abnormal pattern occurrence probability distribution; Through a preset causal relationship mining algorithm, explore and infer the abnormal causal relationship of the target mobile phone based on the probability distribution of the abnormal pattern occurrence, and obtain a preliminary causal relationship chain; Call the historical abnormal causal chain and the historical abnormal nodes on the historical abnormal causal chain from the database, and perform key path screening on the preliminary causal relationship chain based on the historical abnormal causal chain to obtain the screened causal relationship path; Locate the fault nodes of the screened causal relationship path through the historical abnormal nodes to obtain a list of fault points.

[0011] Further, the exploration and inference of the abnormal causal relationship of the target mobile phone based on the probability distribution of the abnormal pattern occurrence through a preset causal relationship mining algorithm to obtain a preliminary causal relationship chain includes: Extract the time series features of the probability distribution of the abnormal pattern through a preset bidirectional long short-term memory network to obtain an abnormal time series feature vector; wherein, the abnormal time series feature vector includes the abnormal duration, the abnormal occurrence frequency, and the abnormal severity; Perform causal relationship modeling on the abnormal time series feature vector to obtain a causal variable association graph; Use a preset graph attention network to calculate the dynamic weights of the causal variable association graph to obtain a weighted causal graph; Explore the causal path of the target mobile phone based on the weighted causal graph to obtain a set of candidate causal chains; Perform probability dependence analysis on the set of candidate causal chains based on the conditional random field model to obtain a causal chain probability distribution; Use the entropy weight method to rank and infer the importance of the causal chain for the causal chain probability distribution to obtain a preliminary causal relationship chain.

[0012] Further, multiple mobile phone function modules are set on the target mobile phone. The static analysis of each potential fault in the list of fault points to obtain function optimization suggestions includes: Extract keywords for each potential fault in the list of fault points through a feature matching algorithm to obtain a fault type mapping table; Segmentally scan the application program code in the target mobile phone based on the fault type mapping table to obtain a set of fault code blocks; Restore the scenario of the set of fault code blocks through a preset code execution path analysis technique to obtain a list of fault trigger conditions; wherein, the list of fault trigger conditions includes the user operation sequence, the system state condition, and the resource occupancy threshold; Classify and label multiple mobile phone function modules based on the list of fault trigger conditions to obtain a function risk level table; Input the functional risk level table into a preset similarity analysis algorithm for historical problem comparison and static analysis to obtain a problem solution knowledge set; Generate repair suggestions for each of the potential faults based on the problem solution knowledge set to obtain functional optimization suggestions.

[0013] The present invention also provides a mobile phone function testing device, including: A parsing module, configured to parse the features of the running logs in the application program of the target mobile phone to obtain an original feature data set; An extraction module, configured to perform multi-dimensional feature extraction on the original feature data set to obtain an application program behavior feature sequence; A first analysis module, configured to perform dynamic function status analysis on the target mobile phone based on the application program behavior feature sequence to obtain a function status transition map; An inference module, configured to identify whether there is an abnormal pattern in the function status transition map through a recurrent neural network. If so, perform causal relationship inference on the function status transition map based on the abnormal pattern to obtain a list of fault points; A second analysis module, configured to perform static analysis on each of the potential faults in the list of fault points to obtain functional optimization suggestions.

[0014] The present invention also provides a computer device, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0015] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in any one of the above are implemented.

[0016] The mobile phone function test method provided by the present invention includes the following steps: parsing the characteristics of the running logs in the application programs of the target mobile phone to obtain an original feature data set; extracting multi-dimensional features from the original feature data set to obtain an application program behavior feature sequence; based on the application program behavior feature sequence, performing dynamic function state analysis on the target mobile phone to obtain a function state transition map; identifying whether there is an abnormal pattern in the function state transition map through a recurrent neural network, and if so, inferring the causal relationship based on the abnormal pattern for the function state transition map to obtain a list of fault points; performing static analysis on each potential fault in the list of fault points to obtain function optimization suggestions. Through the above technical means, the technical problem that traditional mobile phone function test methods often rely on manual operations and predefined test cases, which is not only time-consuming and laborious, but also difficult to comprehensively cover all possible application scenarios, is solved. It realizes the causal relationship inference of abnormal patterns in the function state transition map based on a recurrent neural network, can effectively trace the root cause of abnormal behaviors, and accurately locate specific fault points. This method not only improves the speed of fault troubleshooting, but also enhances the understanding and processing ability of complex system problems with beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the steps of the mobile phone function test method in an embodiment of the present invention; Figure 2 is a structural block diagram of the mobile phone function test device in an embodiment of the present invention; Figure 3 is a schematic structural block diagram of a computer device in an embodiment of the present invention.

[0018] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0020] As Figure 1 shown, Figure 1 is a schematic diagram of the steps of a mobile phone function test method in an embodiment of the present invention; An embodiment of the present invention provides a mobile phone function test method, including the following steps: Step S1, parsing the characteristics of the running logs in the application programs of the target mobile phone to obtain an original feature data set.

[0021] Specifically, the step of parsing the running logs within the applications of the target mobile phone to obtain the original feature dataset is the foundation of the entire mobile phone function testing method. In this process, first, the running logs from various applications on the target mobile phone need to be collected. These logs record various events and status changes during the execution of the applications, such as startup, shutdown, exception throwing, and other information. To extract useful information from these massive logs, we will use specific algorithms and technical means to parse these logs, namely the so-called "feature parsing". This process involves identifying and extracting key elements in the logs, such as timestamps, operation types, objects or resources involved, and result statuses, and converting them into a structured data format, thus forming an original feature dataset containing rich information. For example, in an actual application scenario, assume that we are testing a social network application that allows users to post text and image content and interact with friends. When a user uploads a photo or sends a message in this application, the application will generate corresponding log entries. By performing feature parsing on these logs, we can capture details such as the time point of each upload or send action, the network type used (Wi-Fi or mobile data), the size of the uploaded file, and whether it was successfully completed. After these information are organized into a structured form, they constitute part of our original feature dataset. This dataset not only contains the specific operation information mentioned above but may also include additional metadata such as application version numbers and device models, providing a solid foundation for subsequent multi-dimensional feature extraction. In this way, we ensure that we can comprehensively and accurately reflect the actual behavior of the application, laying a good foundation for the subsequent functional status analysis. At the same time, such a parsing process also ensures that even complex and variable application behaviors can be effectively recorded and understood, thereby supporting more in-depth dynamic functional status analysis and the discovery of potential problems.

[0022] Step S2: Perform multi-dimensional feature extraction on the original feature dataset to obtain the application behavior feature sequence.

[0023] Specifically, after parsing the features of the running logs within the application of the target mobile phone to obtain the original feature dataset, the next step is to perform multi-dimensional feature extraction on the original feature dataset to obtain the application behavior feature sequence. This process aims to mine representative and explanatory information from the original, structured data, which can more clearly describe the behavior patterns of the application and their dynamic changes. Specifically, after completing the feature parsing, we have obtained a dataset rich in information, which records various operations and state transitions of the application. To further understand this data and discover patterns from it, we need to perform multi-dimensional feature extraction. This involves selecting different perspectives or aspects to analyze the original feature dataset, such as the operation frequency in the time dimension, resource usage such as CPU and memory occupancy rates, and changes in the network connection status. By processing the data in these different dimensions, we can construct a more abstract but also more insightful application behavior feature sequence. This sequence not only summarizes the main activity trajectories of the application but also reveals the underlying running logic and potential problems. For example, when testing a social network application, we can observe the specific behaviors of users uploading pictures or sending messages through multi-dimensional feature extraction. In this process, we consider multiple factors: first, the features in the time dimension, such as which time period of the day the user is most active or the number of operations within a specific time period; second, the features of resource consumption, such as the amount of traffic consumed for each picture upload, the required time, and the impact on device resources; third, the characteristics of the network environment, including whether uploads are always through Wi-Fi and the success rate of uploads under mobile data, etc. By integrating the information from these different dimensions, we can generate a detailed application behavior feature sequence that accurately reflects the interaction mode between the user and the application and the performance of the application itself. This method of multi-dimensional feature extraction enables us to more precisely identify the differences between normal and abnormal behavior patterns in subsequent dynamic functional state analysis and provides a solid foundation for the final functional optimization suggestions. At the same time, this method also enhances the ability to understand complex application scenarios, ensuring that even in the face of highly dynamic user behaviors, key features can be effectively captured, thus supporting comprehensive and in-depth functional testing.

[0024] Step S3, based on the application behavior feature sequence, perform dynamic functional state analysis on the target mobile phone to obtain a functional state transition map.

[0025] Specifically, after performing multi-dimensional feature extraction on the original feature dataset to obtain the application behavior feature sequence, the subsequent step is to conduct a dynamic functional state analysis of the target mobile phone based on the application behavior feature sequence to obtain a functional state transition map. This process aims to depict the complete path of the application's transformation under different operations and states by deeply analyzing the application behavior feature sequence, thereby providing an intuitive and systematic perspective for understanding its dynamic behavior. Specifically, after completing the multi-dimensional feature extraction, we have obtained the behavior feature sequence that can reflect various activity patterns during the operation of the application. These sequences record the interaction details between the user and the application, as well as the application's responses under different conditions. To further reveal the logic and rules behind these behaviors, we need to conduct a dynamic functional state analysis on these sequences. This includes identifying all possible functional states in the application and determining how they transform into each other. For example, for a social network application, we can define states such as "start", "browse the home page", "upload pictures", "send messages", etc., and then by analyzing the behavior feature sequence, find out the specific paths and frequencies of the user's transition from one state to another. Such analysis not only helps us understand the user's usage habits but also enables us to discover whether certain specific operations will lead to anomalies or performance degradation. Based on this, we visualize these states and their transition relationships to form the so-called functional state transition map. This map is like a map that clearly shows the workflow of the application in different scenarios. It can help us identify which state transitions are common and which are rare. More importantly, it can point out those state transitions that may cause problems, such as a specific operation always being accompanied by a high failure rate or a long waiting time. Continuing with the above social network application as an example, assume that we observe that when the user tries to upload large-sized pictures via mobile data, the upload often fails. Then this state transition will be highlighted in the map. In this way, we can not only intuitively see the problem but also provide strong support for subsequent fault point location and optimization suggestions. In addition, the functional state transition map also lays the foundation for the recursive neural network (RNN) to identify abnormal patterns. Once a detailed state transition map is established, the RNN can use this information to learn normal behavior patterns and issue an alarm when it detects patterns that deviate from the norm. This method enables us to timely capture potential problems even in the face of complex dynamic changes, ensuring the stability of the application and the continuity of the user experience. In summary, by conducting a dynamic functional state analysis on the application behavior feature sequence and constructing a functional state transition map, we can discover problems earlier, locate fault points more accurately, and ultimately propose effective solutions during the testing process.

[0026] Step S4, identify whether there is an abnormal pattern in the functional state transition graph through a recurrent neural network. If so, perform causal relationship inference on the functional state transition graph based on the abnormal pattern to obtain a list of fault points.

[0027] Specifically, after analyzing the dynamic functional state of the target mobile phone based on the application behavior feature sequence and obtaining the functional state transition map, the next crucial step is to identify whether there are abnormal patterns in the functional state transition map through a Recurrent Neural Network (RNN). This process utilizes the powerful pattern recognition ability of deep learning, aiming to automatically detect those abnormal state transitions that deviate from the normal operation process, and further infer the causal relationships of these abnormal patterns to finally obtain a list of fault points. Specifically, once the functional state transition map is constructed, it becomes the basis for the training and analysis of the RNN. The RNN is a neural network structure particularly suitable for processing sequential data and can capture long-term dependencies in time series, which is especially important for understanding the application behavior feature sequence and its state transitions. In our social network application case, the RNN will receive the state transition sequence extracted from the functional state transition map as input. After being processed by multiple layers of neurons, it outputs the probability assessment of whether each state transition is abnormal. For example, if a user frequently transitions from the "browse home page" state to the "upload failed" state, especially when attempting to upload large-sized pictures in a mobile data environment, the RNN may mark this state transition as abnormal because it does not conform to the regular usage habits of most users. When the RNN identifies possible abnormal patterns, the next task is to infer the causal relationships of these abnormal patterns. This step involves in-depth analysis of the reasons leading to abnormal state transitions to find specific fault points. In the above example, assuming the RNN does discover the abnormal pattern of "failed to upload large-sized pictures in a mobile data environment", then we can explore the root cause of the problem by tracing back the log records and other context information associated with this pattern. It may be caused by factors such as mobile data bandwidth limitations, server response delays, or improper error handling in the client code. By checking and validating these factors one by one, we can determine the most likely fault point and include it in the list of fault points. To ensure the accuracy of the inference results, we can also combine other auxiliary tools and technologies, such as performance monitoring data, system logs, and internal documents provided by developers, to cross-validate our findings. For example, in a social network application, if we suspect that the problem is caused by insufficient mobile data bandwidth, we can check the network performance of other applications on the same device during the same period or compare the upload success rate in a Wi-Fi environment. In this way, not only is the accuracy of fault location improved, but also solid data support is provided for subsequent function optimization. In summary, by identifying abnormal patterns in the functional state transition map through a recurrent neural network and inferring the causal relationships based on these abnormal patterns, we can quickly and accurately find potential fault points in a complex application environment.This method not only enhances the intelligence level of the testing process, but also provides valuable feedback information to the development team, helping them improve the application performance more effectively and enhance the user experience.

[0028] Step S5: Perform static analysis on each potential fault in the fault point list to obtain function optimization suggestions.

[0029] Specifically, after identifying whether there is an abnormal pattern in the functional state transition graph through a recurrent neural network and inferring the causal relationship of the functional state transition graph based on the abnormal pattern to obtain a list of fault points, the next step is to perform a static analysis on each potential fault in the list of fault points to obtain functional optimization suggestions. This process aims to find the root cause of the problem and propose specific improvement measures by conducting an in-depth technical review and evaluation of the identified fault points, thereby ensuring a significant improvement in the functionality of the application and the user experience. Specifically, when the recurrent neural network helps us identify abnormal patterns in a social network application, such as "large-sized picture upload fails in a mobile data environment", and generates a list containing these potential fault points, we need to conduct a static analysis on each fault point. Static analysis means that there is no need to actually run the application, but to discover problems by checking the code, configuration files, and relevant documents. For the above-mentioned problem of large-sized picture upload failure, we can first review the code segments related to the upload function to look for possible logical errors or improper resource management. For example, check whether the chunked upload mechanism is correctly implemented to meet the requirements under different network conditions; or confirm whether there is an appropriate timeout setting to prevent operation failure due to long waiting. At the same time, we can also analyze the server-side log records and API response situations to understand whether there is a transmission interruption caused by high server load or unreasonable interface design. In addition, considering the unique limitations in the mobile data environment, such as bandwidth fluctuations and unstable connections, we should also review the design of the network status monitoring and automatic retry mechanisms in the client code to ensure that the upload success rate can be guaranteed as much as possible even under poor network conditions. Through this multi-angle static analysis, we can not only determine the specific causes of the faults, but also provide a clear direction for subsequent functional optimization. For example, in our social network application case, through static analysis, we found that the client code indeed lacks an effective implementation of chunked upload, which is particularly likely to cause problems when dealing with large-sized pictures. At the same time, we also noticed that the server side has not been specifically optimized for requests from the mobile data environment, resulting in some upload operations failing due to timeouts. For these problems, we can propose a series of functional optimization suggestions: one is to improve the client code by adding functions for intelligent chunked upload and dynamic adjustment of upload parameters; the other is to optimize the server-side processing logic by increasing support for weak network environments, such as extending the timeout time and providing more detailed error feedback. By implementing these suggestions, not only can the current problems be solved, but the robustness of the application under various network conditions can also be improved, thereby enhancing the overall user experience. In short, by performing a static analysis on each potential fault in the list of fault points, we can deeply understand the essence of the problem and find the best solution without relying on the actual operating environment.This method not only helps to quickly locate and solve problems, but also provides valuable improvement suggestions for the development team, promoting the continuous optimization and development of the product. Ultimately, it enables us to maintain efficient and accurate testing and maintenance capabilities when facing complex application behaviors, ensuring that each release version can meet the needs and expectations of users.

[0030] In a specific embodiment, parsing the features of the running logs within the application of the target mobile phone to obtain an original feature data set, including: Performing session segmentation on the application running logs through a log segmentation algorithm to obtain a sequence of log segments; Performing syntax structure parsing on the sequence of log segments based on the abstract syntax tree technology to obtain syntax feature vectors; Performing pattern matching on the syntax feature vectors through a regular expression engine to obtain a set of log templates; Performing topic extraction on the set of log templates to obtain a log topic distribution; wherein, the log topic distribution includes function categories and state transition information; Constructing an index for the log topic distribution through a dynamic trie technology to obtain a hierarchical feature index; wherein, the hierarchical feature index includes a feature path, keyword weights, and association metrics; Performing parallel processing on the hierarchical feature index through a preset MapReduce framework to obtain an original feature data set.

[0031] Specifically, first, by performing session segmentation on the application running logs, we can break down the continuous log stream into individual session units, namely a sequence of log fragments. Each session represents an interaction process between the user and the application or a series of operation processes within the application. For example, in a social network application, a session may start with the user opening the application to browse the home page, then uploading a picture or sending a message, and finally closing the application. Through the log segmentation algorithm, we can accurately identify these session boundaries, thus providing a clear timeline and context information for subsequent analysis. This helps reduce noise interference and improve the depth of understanding of specific events or behaviors. Next, based on the Abstract Syntax Tree (AST) technology, parsing the syntax structure of the log fragment sequence can further refine the understanding of each log entry. The abstract syntax tree is a data structure used to represent the syntax structure of source code, which shows the statements and expressions in the program in a tree-like manner. In our scenario, although not directly targeting the source code of programming languages, by parsing the log content in a similar way, we can capture the operation types, parameter values, and their relationships contained in the log entries, forming what is called a syntax feature vector. For a social network application, this may mean being able to distinguish different types of log entries such as "start", "upload picture", "send message", etc., and understanding the specific parameters they carry, such as file size, network type, etc. This structured representation makes the log information easier to process and analyze. Subsequently, using a regular expression engine to perform pattern matching on the syntax feature vectors aims to induce a set of general log template sets. Regular expressions are a powerful text search tool that can define complex matching rules to identify log entries with similar structures. At this stage, we use the previously parsed syntax feature vectors as input and, through a series of predefined regular expression rules, find those frequently occurring and consistently formatted log patterns. For a social network application, this step can help us identify common log templates such as "Picture upload successful" and "Picture upload failed: Reason X". These templates are not only a compressed representation of the original logs but also provide a basis for subsequent topic extraction. Immediately following, perform topic extraction on the log template set to reveal the underlying meaning behind the logs and obtain the log topic distribution. The topic here refers to the implicit functional categories in the logs and their corresponding state transition information. Through clustering analysis or other machine learning methods, different topics can be mined from a large number of log templates, such as "user interaction", "resource management", "exception handling", etc. At the same time, the state transition relationships between different topics can also be identified, such as the normal process from "picture upload" to "upload completed" or the abnormal situation from "upload started" to "upload failed". This is crucial for understanding the behavioral logic of the application, especially when we want to evaluate its stability and performance.Then, an index is constructed for the log topic distribution through the dynamic trie technology, and finally a hierarchical feature index is obtained. The dynamic trie is an efficient data structure suitable for fast look-up and update operations. Here, it is used to organize the feature paths, keyword weights, and association metrics derived from the log topic distribution to form a well-structured index system. The hierarchical feature index not only improves the query efficiency but also enhances the ability to understand complex log structures. For example, in a social network application, we can, based on the feature paths of user behavior (such as "browse the home page -> upload pictures -> share links"), combined with keyword weights (such as the importance of "upload failed" relative to "upload successful"), and the association metrics between different operations (such as the correlation between "upload pictures" and "server response time"), more accurately describe the usage habits of users and the problems they encounter. Finally, the hierarchical feature index is processed in parallel through a preset MapReduce framework to obtain the original feature dataset. MapReduce is a distributed computing model particularly suitable for task execution on large-scale datasets. At this stage, we divide the hierarchical feature index into multiple subtasks and allocate them to different nodes in the cluster for parallel processing. Each node is responsible for processing a part of the data, performing operations such as statistical summarization and pattern discovery, and finally combining the results to form the complete original feature dataset. This method not only accelerates the speed of data analysis but also effectively addresses the challenges brought by massive logs. For a social network application, this means that a large number of user log records can be processed in a short time, potential problems can be discovered in a timely manner, and improvement measures can be proposed to ensure the stable operation of the application and a high-quality user experience. In summary, through the above series of feature analyses of the in-application running logs of the target mobile phone, we can not only comprehensively and deeply understand the running status of the application but also provide a solid foundation for subsequent function optimization and fault diagnosis. This method makes full use of the advantages of modern information technology to achieve the effective transformation from raw logs to useful knowledge, which is of great significance for improving the quality and reliability of mobile applications.

[0032] In a specific embodiment, the multi-dimensional feature extraction of the original feature dataset to obtain the application behavior feature sequence includes: Performing user interaction event analysis on the original feature dataset to obtain the user behavior sequence; Using the preset wavelet transform technology to perform multi-scale decomposition on the user behavior sequence to obtain the time feature vector; Performing multi-dimensional extraction on the time feature vector to obtain the behavior pattern feature; Classifying the behavior pattern features to obtain the behavior classification features; wherein, the behavior classification features include click frequency features, sliding mode features, long-press behavior features, browsing depth features, and interaction density features; By means of a preset timing algorithm, perform timing pattern mining on the behavior classification features to obtain an application program behavior feature sequence.

[0033] Specifically, first, we start with the analysis of user interaction events in the original feature dataset. The goal of this step is to extract a series of operation events triggered by users from the original data to form a user behavior sequence. In social network applications, these interaction events can include launching the application, browsing the home page, uploading pictures, sending messages, etc. By parsing the timestamps, operation types, and result statuses of these events, we can construct complete user behavior trajectories, that is, user behavior sequences. For example, suppose a user opens the social network application multiple times in a day, browses several feeds, uploads a photo and shares it with friends. Then this user behavior sequence records a series of operations of this user and their chronological order. Such detailed records provide a solid foundation for subsequent behavior pattern analysis. Next, in order to understand the changing patterns of these user behaviors over time in more depth, we use the preset wavelet transform technology to perform multi-scale decomposition on the user behavior sequence, thereby obtaining time feature vectors. Wavelet transform is a signal processing method, especially suitable for the analysis of non-stationary signals. It can capture local changes in the signal at different scales, which is of great significance for revealing periodicity and suddenness in user behaviors. Taking social network applications as an example, through wavelet transform, we can discover different active time periods of users on weekdays and weekends, or identify behavior peaks brought about by certain specific activities (such as during festivals or major events). The time feature vectors not only contain the absolute time information of when user behaviors occur, but also reflect the changing trends of behavior intensity and frequency, providing rich materials for our further multi-dimensional feature extraction. Subsequently, based on the above time feature vectors, we perform multi-dimensional extraction on them to obtain behavior pattern features. The focus of this stage is to describe user behaviors from different perspectives in order to more comprehensively characterize their characteristics. For example, we can calculate the time intervals between each user behavior, evaluate the duration and repetition rate of behaviors, or count the proportions of different types of behaviors. Continuing with the above social network application example, we may notice that some users frequently refresh the feed page, while others are more involved in commenting and liking. In this way, we can not only distinguish the usage habits of different users, but also capture the behavioral differences of the same user in different situations. The behavior pattern features provide us with a more abstract but also more explanatory perspective to help us understand the logic behind user behaviors. On this basis, we classify the behavior pattern features to obtain behavior classification features, which include click frequency features, sliding mode features, long-press behavior features, browsing depth features, and interaction density features. Each feature represents a way or degree of interaction between the user and the application program.The click frequency feature can tell us the number of times a user clicks on the screen within a unit of time, reflecting their operation speed and sense of urgency; the sliding mode feature helps identify the user's content browsing habits, such as whether they tend to flip through quickly or read carefully; the long - press behavior feature may be related to the use of certain special functions, such as saving pictures or copying links; the browsing depth feature shows the degree to which a user explores the content of the application, whether they only stay on the surface or dig deeper; and the interaction density feature comprehensively considers the frequency and complexity of various operations of a user within a period of time. By defining and measuring these behavioral classification features, we can establish a systematic set of evaluation indicators to quantify and compare the behavioral patterns of different user groups. Finally, in order to integrate these scattered behavioral classification features into a coherent whole, we perform temporal pattern mining on the behavioral classification features through a preset temporal algorithm, thereby obtaining the application behavior feature sequence. The temporal algorithm can help us identify the temporal relationships between behavioral classification features, revealing the development path and evolution trend of user behavior. For example, in a social network application, through temporal analysis, we can find that some users show a high interaction density when first exposed to the application, but gradually form a fixed usage pattern over time, such as regularly posting updates or following specific topics. In addition, temporal pattern mining can also help us predict future behaviors and take proactive measures to optimize the user experience. In this way, we can not only summarize past data but also provide guidance for future improvements, ensuring that the application always meets the needs and expectations of users. In summary, through multi - dimensional feature extraction from the original feature dataset, we can construct a detailed and highly interpretable application behavior feature sequence. This method not only deepens our understanding of user behavior but also provides strong support for subsequent function optimization and fault diagnosis. By combining advanced mathematical tools and technical means, we have achieved an effective transformation from raw data to meaningful knowledge, making mobile phone function testing more intelligent and efficient. The application of this method, especially in a complex social network environment, greatly enhances our ability to discover potential problems and propose targeted solutions, ultimately promoting the continuous development of the application and the continuous improvement of the user experience.

[0034] In a specific embodiment, the dynamic functional state analysis of the target mobile phone based on the application behavior feature sequence to obtain a functional state transition map includes: Perform state - space mapping on the application behavior feature sequence to obtain a discrete representation of the functional state; Use the random - walk theory and transfer - entropy algorithm to model the transition probability of the state discrete representation to obtain a state transition probability map; Through a preset graph convolutional network, perform node feature encoding on the state transition probability map to obtain an enhanced state representation vector; Performing dynamic functional state analysis on the target mobile phone based on the enhanced state representation vector to obtain a weighted state transition graph; Capturing the potential correlation between the discretized representations of the functional states of the target mobile phone based on the weighted state transition graph to obtain a state correlation vector; Performing spectral clustering analysis on the state correlation vector to obtain a functional state clustering set; Using a preset small-world network model to perform topological reconstruction on the functional state clustering set to obtain a state transition network; Performing dynamic path analysis on the state transition network to obtain a functional state transition map; wherein, the functional state transition map includes nodes and edges, the nodes represent functional states, and the edges represent the transition relationships between functional states.

[0035] Specifically, first, we start with the state space mapping of the application behavior feature sequence. This step aims to convert continuous behavior features into discrete functional state representations. In social network applications, user behaviors can be highly diverse, such as browsing, uploading pictures, sending messages, etc. To facilitate analysis, we need to map these complex behaviors onto a finite set of states to form a discrete representation of functional states. For example, we can define "launch the application", "browse the home page", "upload pictures", "send messages", etc. as different functional states. Each state not only describes the user's current operation but also includes the specific environment and conditions in which the operation occurs. Through this state space mapping, we transform the original behavior feature sequence into a series of clear state changes, providing a basis for further analysis. Next, the random walk theory and transfer entropy algorithm are used to model the transition probabilities of the state discrete representation to obtain a state transition probability graph. The random walk theory is a method for describing the evolution of a system's state over time. It assumes that the probability of the system transitioning from one state to another depends only on the current state and is not affected by historical states. Combining with the transfer entropy algorithm, we can calculate the likelihood of transitions between different states. For social network applications, this means that we can evaluate the probability of a user transitioning from one functional state (such as "browse the home page") to another state (such as "upload pictures"). The state transition probability graph not only reveals the possible paths of user behavior but also helps us identify which state transitions are more common or important. Then, through a pre-set graph convolutional network (GCN), the node features of the state transition probability graph are encoded to obtain an enhanced state representation vector. The graph convolutional network is a neural network specifically designed to process graph-structured data, which can capture the complex relationships between nodes and their neighbors. At this stage, we regard each node in the state transition probability graph as an input unit and use GCN to learn the deep feature representation of each state. For example, in social network applications, after encoding by GCN, we can more accurately understand the unique attributes of each functional state and how they affect other states. The enhanced state representation vector not only contains the initial state information but also integrates the context features of surrounding states, making subsequent analysis more accurate and comprehensive. On this basis, based on the enhanced state representation vector, a dynamic functional state analysis of the target mobile phone is performed to obtain a weighted state transition graph. This process involves re-evaluating the transition relationships between states and assigning corresponding weights according to the new feature representation. For social network applications, this means that we can recalculate the likelihood of a user transitioning from one functional state to another considering more factors. For example, if a certain user often directly "uploads pictures" after "browsing the home page", the weight on this path may be relatively high.The weighted state transition graph not only reflects the actual transition frequencies between states but also embodies the strength of the internal connections between them, providing a basis for us to further explore the relationships between states. Next, based on the weighted state transition graph, capture the potential correlations between the discretized representations of the functional states of the target mobile phone to obtain a state correlation vector. The goal of this step is to discover the implicit relationships between states that are not easily detectable but do exist. For example, in a social network application, although there may not seem to be a direct transition path between certain states on the surface, there may actually be an indirect influence. By analyzing the weighted state transition graph, we can identify these potential correlations and quantify them with a state correlation vector. This in-depth mining helps us better understand the internal working mechanism of the application, especially when we want to evaluate its stability and performance. Subsequently, perform spectral clustering analysis on the state correlation vector to obtain a set of functional state clusters. Spectral clustering is a clustering method based on graph theory that can find the similarities between data points through matrix decomposition techniques and group them. Here, we use the state correlation vector as the input and identify groups of states with similar characteristics through spectral clustering analysis. For a social network application, this may mean that we can discover some common interaction patterns or behavior chains, such as "browse the home page -> upload a picture -> share a link". The set of functional state clusters not only simplifies the management of a large number of states but also helps us focus on the most frequently occurring or important state combinations, thereby improving the analysis efficiency. Immediately afterwards, adopt a preset small-world network model to perform topological reconstruction on the set of functional state clusters to obtain a state transition network. The small-world network model is a special network structure that has the characteristics of both a high clustering coefficient and a short average path length, making it very suitable for describing complex systems in the real world. By mapping the set of functional state clusters into such a network model, we can construct a connection pattern that is both tight and efficient. For a social network application, this means that we can create a network structure that reflects the user behavior patterns and the internal logic of the application. The state transition network not only shows the direct and indirect connections between various functional states but also reveals their relative positions and roles in the entire system. Finally, perform dynamic path analysis on the state transition network to obtain a functional state transition map. Dynamic path analysis is a method used to study the reachability between nodes in a network, which can help us trace all possible paths from one state to another. In our scenario, through dynamic path analysis, we can determine which state transitions are the most common, which are rare, and more importantly, we can also identify those state transitions that may cause problems, such as a particular operation always being accompanied by a high failure rate or a long waiting time. The functional state transition map finally presents a clear map where the nodes represent functional states and the edges represent the transition relationships between functional states.This spectrum map is not only an important tool for understanding the behavior of the application, but also provides a solid foundation for subsequent fault diagnosis and optimization suggestions. In summary, through the above series of dynamic functional state analyses of the application behavior feature sequence, we can not only comprehensively and deeply understand the running status of the application, but also provide strong support for subsequent function optimization and fault diagnosis. This method makes full use of the advantages of modern information technology to achieve an effective transformation from behavior features to the functional state transition spectrum map, which is of great significance for improving the quality and reliability of mobile applications. Especially in a complex social network environment, this method greatly enhances our ability to discover potential problems and propose targeted solutions, ultimately promoting the continuous development of the application and the continuous improvement of the user experience.

[0036] In a specific embodiment, the method for identifying whether there is an abnormal pattern in the functional state transition spectrum map through a recurrent neural network. If there is an abnormal pattern, then based on the abnormal pattern, an abnormal causal relationship inference is performed on the target mobile phone to obtain a list of fault points, including: Using a deep neural network algorithm to perform a deep scan on the functional state transition spectrum map to obtain a deep scan result; Determine whether there is an abnormality in the deep scan result. If there is an abnormality, then use a preset Bayesian network inference algorithm to perform an abnormal pattern inference on the deep scan result to obtain an abnormal pattern occurrence probability distribution; Through a preset causal relationship mining algorithm, based on the abnormal pattern occurrence probability distribution, explore and infer the abnormal causal relationship of the target mobile phone to obtain a preliminary causal relationship chain; Call the historical abnormal causal chain and the historical abnormal nodes on the historical abnormal causal chain from the database, and based on the historical abnormal causal chain, perform a critical path screening on the preliminary causal relationship chain to obtain a screened causal relationship path; Locate the fault nodes through the historical abnormal nodes on the screened causal relationship path to obtain a list of fault points.

[0037] Specifically, first, we use the deep neural network (DNN) algorithm to perform a deep scan on the functional state transition map to obtain the deep scan results. Deep neural networks are a powerful machine learning model that is particularly good at processing complex nonlinear data. At this stage, we use the previously constructed functional state transition map as input and use the powerful pattern recognition ability of DNN to perform layer-by-layer analysis and feature extraction. For example, in social network applications, DNN can automatically detect those uncommon state transition paths or state combinations that appear frequently but do not conform to conventional logic. The results of this deep scan not only reveal subtle differences in normal behavior patterns, but also capture abnormal situations that may cause problems. Next, we need to determine whether there are anomalies in the deep scan results. Once DNN identifies potential abnormal patterns, we use the preset Bayesian network inference algorithm to perform abnormal pattern inference on the deep scan results to obtain the probability distribution of abnormal pattern occurrence. Bayesian network is a probabilistic graphical model that can calculate the probability of different events based on known conditions. In our case, this means that we can evaluate the actual probability of occurrence of each abnormal pattern and determine which anomalies are most worthy of attention. For example, if a specific operation is always accompanied by a high failure rate, then the abnormal pattern corresponding to this operation will be given a higher probability of occurrence. In this way, we can prioritize the issues that are most likely to affect the user experience. Then, in order to further understand the reasons behind these abnormal patterns, we use the preset causal relationship mining algorithm to explore and infer the abnormal causal relationship of the target mobile phone based on the probability distribution of the abnormal pattern occurrence, and obtain a preliminary causal relationship chain. The causal relationship mining algorithm aims to find the causal relationship between events and help us explain why certain abnormalities occur. For social network applications, this may mean that we need to explore the reasons why users fail to upload pictures, such as whether it is related to insufficient mobile data bandwidth, server response delays, or client code errors. By analyzing the interaction between these factors, we can construct a preliminary causal relationship chain to reveal the root cause of the abnormality. On this basis, in order to improve the reliability and accuracy of the causal relationship chain, we call the historical abnormal causal chain and the historical abnormal nodes on the historical abnormal causal chain from the database, and screen the key paths of the preliminary causal relationship chain based on the historical abnormal causal chain to obtain the screened causal relationship path. The historical abnormal causal chain contains solutions and experience summaries of similar problems in the past, which provide valuable references for current causal relationship inference. For example, in a social networking application, if we find that the current "image upload failure" problem is similar to the same problem in a certain version in the past, we can learn from the solution at that time and quickly identify the critical path. In this way, we can not only speed up the fault location, but also improve the accuracy of diagnosis.Finally, fault node localization is performed on the filtered causal relationship path through the historical abnormal nodes to obtain a list of fault points. A historical abnormal node refers to the specific location or component that was identified as a fault point in past problems. By comparing the filtered causal relationship path, we can precisely identify the key to the current problem and include it in the list of fault points. For example, in a social network application, assume that after a series of analyses, we find that the root cause of "failed picture upload" is the lack of an effective chunked upload mechanism in the client code. Then we will clearly record this problem as the focus of subsequent optimization. In this way, we can not only provide specific improvement suggestions but also offer a clear direction for the development team to ensure that each release version meets the needs and expectations of users. In summary, through the above series of abnormal pattern recognition and causal relationship inference on the function state transition graph, we can not only comprehensively and deeply understand the running status of the application but also provide strong support for subsequent function optimization and fault diagnosis. This method makes full use of the advantages of modern information technology to achieve an effective transformation from abnormal patterns to a list of fault points, which is of great significance for improving the quality and reliability of mobile applications. Especially in a complex social network environment, this method greatly enhances our ability to discover potential problems and propose targeted solutions, ultimately promoting the continuous development of the application and the continuous improvement of the user experience.

[0038] In a specific embodiment, through a preset causal relationship mining algorithm, based on the probability distribution of the occurrence of the abnormal pattern, abnormal causal relationship exploration and inference are performed on the target mobile phone to obtain a preliminary causal relationship chain, including: Temporal feature extraction is performed on the probability distribution of the occurrence of the abnormal pattern through a preset bidirectional long short-term memory network to obtain an abnormal temporal feature vector; wherein, the abnormal temporal feature vector includes abnormal duration, abnormal occurrence frequency, and abnormal severity; Causal relationship modeling is performed on the abnormal temporal feature vector to obtain a causal variable association graph; A preset graph attention network is used to calculate dynamic weights for the causal variable association graph to obtain a weighted causal graph; Based on the weighted causal graph, causal path exploration is performed on the target mobile phone to obtain a set of candidate causal chains; Based on a conditional random field model, probability dependence analysis is performed on the set of candidate causal chains to obtain a causal chain probability distribution; The entropy weight method is used to rank and infer the importance of the causal chains for the causal chain probability distribution to obtain a preliminary causal relationship chain.

[0039] Specifically, first, we extract the temporal features of the abnormal pattern occurrence probability distribution through a pre-set bidirectional long short-term memory network (BiLSTM) to obtain an abnormal temporal feature vector. BiLSTM is a neural network structure particularly suitable for processing sequence data. It can capture long-term dependencies in time series while retaining forward and backward time information. At this stage, we use the abnormal pattern occurrence probability distribution as the input and utilize BiLSTM to extract an abnormal temporal feature vector containing key information such as abnormal duration, abnormal occurrence frequency, and abnormal severity. For example, in a social network application, if a user frequently encounters the problem of "failed picture upload", BiLSTM can help us determine the frequency of these problems, the duration of each upload attempt, and the impact of these failures on the user experience. This temporal feature extraction provides a solid foundation for subsequent causal relationship analysis. Next, to further understand the internal connections between these abnormal temporal features, we perform causal relationship modeling on the abnormal temporal feature vector to obtain a causal variable association graph. Causal relationship modeling aims to find the causal connections between events and help us explain why certain abnormalities occur. In our case, this means we need to explore the interactions between different abnormal features, such as whether the abnormal duration affects the abnormal occurrence frequency, or how the abnormal severity is related to specific operations. By constructing a causal variable association graph, we can visually see the direct and indirect relationships between various abnormal features, providing a clear map for further analysis. For example, in a social network application, we may find that "failed upload of large-sized pictures in a mobile data environment" is often accompanied by longer waiting times and higher failure rates, and there is a significant correlation between these features. Then, to improve the accuracy of causal relationship analysis, we use a pre-set graph attention network (GAT) to perform dynamic weight calculation on the causal variable association graph to obtain a weighted causal graph. The graph attention network is a neural network model that can automatically learn the importance between nodes. It can emphasize the connections that play a key role in causal relationships by assigning different weights. In this process, GAT will dynamically adjust the weights of each connection according to the nodes and their adjacency relationships in the causal variable association graph, thus forming a more accurate weighted causal graph. For a social network application, this means we can more clearly identify which factors play a decisive role in abnormal behavior, such as insufficient mobile data bandwidth or server response delay. The weighted causal graph not only shows the strength relationship between various abnormal features but also reveals their relative importance in the entire system, providing guidance for our subsequent causal path exploration. Based on this, we explore the causal path of the target mobile phone based on the weighted causal graph to obtain a set of candidate causal chains. The goal of this step is to find all possible paths from the initial abnormal state to the final result.By traversing the nodes and edges in the weighted causal graph, we can construct a series of candidate causal chains, each of which represents a possible abnormal propagation path. For example, in a social network application, we may find a causal chain: "Mobile data environment -> Large-size picture upload -> Long waiting time -> Upload failure". These candidate causal chains not only reflect the development trajectory of the abnormal pattern but also provide us with multiple perspectives to understand and solve problems. Next, in order to evaluate the effectiveness and reliability of these candidate causal chains, we conduct a probability dependence analysis on the set of candidate causal chains based on the conditional random field model (CRF) to obtain the causal chain probability distribution. The conditional random field model is a probabilistic graphical model for modeling sequence data, which can effectively capture the dependencies in the sequence and make predictions. At this stage, we regard the candidate causal chains as input sequences and use CRF to calculate the likelihood of each chain occurring. For example, in a social network application, CRF can help us determine which causal chain is most likely to lead to "picture upload failure" and the relative probabilities between each chain. The causal chain probability distribution provides us with a quantitative evaluation criterion, enabling us to more rationally select the most reasonable causal explanation. Finally, in order to further optimize the selection of the causal relationship chain, we use the entropy weight method to rank and infer the importance of the causal chains in the causal chain probability distribution to obtain the preliminary causal relationship chain. The entropy weight method is a method based on information theory, which can determine the importance of each factor according to the uncertainty of the data. By applying the entropy weight method, we can score each chain in the causal chain probability distribution and select those causal relationship chains with the most explanatory power and influence. For example, in a social network application, if we find that the causal chain "Large-size picture upload failure in the mobile data environment" has the highest score, then this chain will be given priority as the preliminary causal relationship chain. This method not only improves the accuracy of causal relationship inference but also provides clear improvement suggestions for the development team to ensure that each release version can meet the needs and expectations of users. In summary, through the above series of causal relationship explorations and inferences on the abnormal pattern occurrence probability distribution, we can not only comprehensively and deeply understand the running status of the application but also provide strong support for subsequent function optimization and fault diagnosis. This method makes full use of the advantages of modern information technology to achieve an effective transformation from the abnormal pattern to the preliminary causal relationship chain, which is of great significance for improving the quality and reliability of mobile applications. Especially in a complex social network environment, this method greatly enhances our ability to discover potential problems and propose targeted solutions, ultimately promoting the continuous development of the application and the continuous improvement of the user experience.

[0040] In a specific embodiment, a plurality of mobile phone function modules are provided on the target mobile phone, and the static analysis of each potential fault in the fault point list to obtain function optimization suggestions includes: Extract keywords for each potential fault in the fault point list through a feature matching algorithm to obtain a fault type mapping table; Segmentedly scan the application program code in the target mobile phone based on the fault type mapping table to obtain a set of fault code blocks; Restore the scene of the set of fault code blocks through a preset code execution path analysis technology to obtain a list of fault trigger conditions; wherein, the list of fault trigger conditions includes a user operation sequence, a system state condition, and a resource occupancy threshold; Classify and label multiple mobile phone function modules based on the list of fault trigger conditions to obtain a function risk level table; Input the function risk level table into a preset similarity analysis algorithm for historical problem comparison and static analysis to obtain a set of problem solution knowledge; Generate repair suggestions for each potential fault based on the set of problem solution knowledge to obtain function optimization suggestions.

[0041] Specifically, first, we extract keywords for each potential fault in the fault point list through a feature matching algorithm to obtain a fault type mapping table. The feature matching algorithm aims to quickly locate and extract representative keywords or phrases from a large amount of text data, which can help us understand and classify different fault types. For example, in a social network application, if the fault point list mentions "failed to upload a picture", the feature matching algorithm can help us identify keywords related to "upload" and "failure", and map them to specific fault categories, such as "network connection problem" or "server response timeout". The fault type mapping table provides a clear direction for our subsequent code scanning, enabling us to search for relevant code segments more targeted. Next, based on the fault type mapping table, we perform segmented scanning on the application code in the target mobile phone to obtain a set of fault code blocks. At this stage, we will conduct a detailed inspection in the source code of the application according to the keyword guidance in the mapping table. For example, for the problem of "failed to upload a picture", we will focus on scanning the code parts related to file upload, network request processing, error capture, etc. In this way, we can find the specific code segments that may cause the fault and form a set of fault code blocks. This step not only helps to narrow down the problem scope but also provides direct evidence support for subsequent scenario restoration. Then, to understand the performance of these fault code blocks in actual operation more deeply, we perform scenario restoration on the set of fault code blocks through a preset code execution path analysis technique to obtain a list of fault trigger conditions. Code execution path analysis is a technique for tracing the program execution flow, which can help us reproduce the specific conditions that lead to the fault. For example, in a social network application, we can find through code execution path analysis that "failed to upload a picture" often occurs in a mobile data environment when uploading large-sized pictures and the network bandwidth is insufficient. This scenario restoration can not only reveal the background of the fault but also help us identify which user operation sequences (such as starting the upload), system state conditions (such as the current network type), and resource occupancy thresholds (such as available bandwidth) jointly cause the problem. On this basis, we classify and label multiple mobile phone function modules based on the list of fault trigger conditions to obtain a functional risk level table. A function module refers to an independent unit that implements different functions on a mobile phone, such as a camera module, a message push module, etc. By analyzing the fault trigger conditions, we can determine which function modules are most vulnerable to being affected and classify and label them accordingly. For example, if "failed to upload a picture" mainly involves the camera module and the network module, then these two modules may be marked as high risk. The functional risk level table not only provides us with an intuitive risk assessment tool but also helps the development team to prioritize the problems that have the greatest impact on the user experience.Subsequently, to further optimize the solution, we input the functional risk rating table into a preset similarity analysis algorithm for historical problem comparison and static analysis, obtaining a knowledge set of problem solutions. The similarity analysis algorithm aims to find existing solutions by comparing the similarity between new and old problems. For example, in a social network application, if we find that the current problem is very similar to a similar problem in a past version, we can draw on the solution at that time and quickly propose countermeasures. At the same time, static analysis can also help us identify potential problems in the code, such as unhandled exceptions or unreasonable resource management. In this way, we can construct a knowledge set containing various solutions, providing rich reference materials for the subsequent generation of repair suggestions. Finally, based on the knowledge set of problem solutions, repair suggestions are generated for each of the potential faults, obtaining functional optimization suggestions. The goal of this step is to comprehensively consider all known information and provide specific repair solutions for each potential fault. For example, for the problem of "failed picture upload", we can suggest improving the client code by adding an intelligent chunked upload mechanism, or optimizing the server-side processing logic to increase support for weak network environments. By implementing these suggestions, not only can the current problems be solved, but the robustness of the application under various network conditions can also be improved, thereby enhancing the overall user experience. In summary, through the above series of static analyses of each potential fault in the fault point list, we can not only comprehensively and deeply understand the running status of the application program, but also provide strong support for subsequent functional optimization and fault diagnosis. This method makes full use of the advantages of modern information technology, realizing the effective transformation from the fault point list to functional optimization suggestions, which is of great significance for improving the quality and reliability of mobile applications. Especially in a complex social network environment, this method greatly enhances our ability to discover potential problems and propose targeted solutions, ultimately promoting the continuous development of the application and the continuous improvement of the user experience.

[0042] The mobile phone function testing method in the embodiment of the present invention has been described above. Next, the mobile phone function testing system in the embodiment of the present invention will be described. Please refer to Figure 2 , an embodiment of the mobile phone function testing system in the embodiment of the present invention includes: A parsing module 21, configured to parse the features of the running log in the application program of the target mobile phone to obtain an original feature data set; An extraction module 22, configured to perform multi-dimensional feature extraction on the original feature data set to obtain an application program behavior feature sequence; A first analysis module 23, configured to perform dynamic function status analysis on the target mobile phone based on the application program behavior feature sequence to obtain a function status transition map; An inference module 24, configured to identify whether there is an abnormal pattern in the functional state transition map through a recurrent neural network. If so, perform causal relationship inference on the functional state transition map based on the abnormal pattern to obtain a list of fault points; A second analysis module 25, configured to perform static analysis on each potential fault in the list of fault points to obtain functional optimization suggestions.

[0043] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to that described in the above method embodiment, and details are not described herein again.

[0044] Refer to Figure 3 , this embodiment of the present invention further provides a computer device, and its internal structure may be as Figure 3 shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements the above method.

[0045] Those skilled in the art can understand that Figure 3 the structure shown in

[0046] is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0047] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0048] It should be noted that in this article, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, device, article or method comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, device, article or method. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of additional identical elements in the process, device, article or method comprising the element.

[0049] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A method for testing mobile phone functions, characterized in that: The following steps are involved: Perform feature analysis on the running logs in the target mobile phone's application to obtain the original feature data set; Performing multi-dimensional feature extraction on the original feature data set to obtain an application behavior feature sequence; Performing dynamic functional state analysis on the target mobile phone based on the application behavior feature sequence to obtain a functional state transition graph; Identify whether there is an abnormal pattern in the functional state transfer map through a recursive neural network, and if so, perform causal relationship inference on the functional state transfer map based on the abnormal pattern to obtain a list of fault points; Perform static analysis on each potential fault in the fault point list to obtain function optimization suggestions.

2. The mobile phone function testing method according to claim 1, characterized in that: The feature analysis of the running log in the application of the target mobile phone is performed to obtain the original feature data set, including: Use the log segmentation algorithm to segment the application running log into sessions and obtain a log segment sequence. Based on the abstract syntax tree technology, the grammatical structure of the log fragment sequence is parsed to obtain a grammatical feature vector; Performing pattern matching on the grammatical feature vector through a regular expression engine to obtain a log template set; Performing topic extraction on the log template set to obtain log topic distribution; wherein the log topic distribution includes function category and state transition information; The log topic distribution is indexed and constructed by a dynamic dictionary tree technology to obtain a hierarchical feature index; wherein the hierarchical feature index includes a feature path, a keyword weight, and an association metric; The hierarchical feature index is processed in parallel through the preset MapReduce framework to obtain the original feature data set.

3. The mobile phone function testing method according to claim 1, characterized in that: The multi-dimensional feature extraction is performed on the original feature data set to obtain an application behavior feature sequence, including: Performing user interaction event analysis on the original feature data set to obtain a user behavior sequence; Using a preset wavelet transform technology to perform multi-scale decomposition on the user behavior sequence to obtain a time feature vector; Performing multi-dimensional extraction on the time feature vector to obtain behavior pattern features; Classifying the behavior pattern features to obtain behavior classification features; wherein the behavior classification features include click frequency features, sliding mode features, long press behavior features, browsing depth features, and interaction density features; The behavior classification features are mined using a preset time series algorithm to obtain an application behavior feature sequence.

4. The mobile phone function testing method according to claim 1, characterized in that: The step of performing dynamic functional state analysis on the target mobile phone based on the application behavior feature sequence to obtain a functional state transition graph includes: Performing state space mapping on the application behavior feature sequence to obtain a discrete representation of the functional state; The random walk theory and the transfer entropy algorithm are used to perform transfer probability modeling on the state discretization representation to obtain a state transfer probability graph; By using a preset graph convolutional network, node feature encoding is performed on the state transition probability graph to obtain an enhanced state representation vector; Performing dynamic functional state analysis on the target mobile phone based on the enhanced state representation vector to obtain a weighted state transition diagram; Capturing the potential correlation between the discretized representations of the functional states of the target mobile phone based on the weighted state transition diagram to obtain a state correlation vector; Performing spectral clustering analysis on the state association vector to obtain a functional state clustering set; Using the preset small-world network model, the functional state clustering set is topologically reconstructed to obtain a state transition network; Dynamic path analysis is performed on the state transition network to obtain a functional state transition graph; wherein the functional state transition graph includes nodes and edges, the nodes represent functional states, and the edges represent transfer relationships between functional states.

5. The mobile phone function testing method according to claim 1, characterized in that: The method of identifying whether an abnormal pattern exists in the functional state transition graph by using a recursive neural network, and if so, performing abnormal causal relationship inference on the target mobile phone based on the abnormal pattern to obtain a list of fault points includes: Using a deep neural network algorithm to perform a deep scan on the functional state transition map to obtain a deep scan result; Determine whether there is an abnormality in the deep scanning result. If so, use a preset Bayesian network inference algorithm to perform abnormal pattern inference on the deep scanning result to obtain a probability distribution of the abnormal pattern; By using a preset causal relationship mining algorithm, based on the probability distribution of the abnormal pattern occurrence, the abnormal causal relationship of the target mobile phone is explored and inferred to obtain a preliminary causal relationship chain; Retrieving the historical abnormal causal chain and the historical abnormal nodes on the historical abnormal causal chain from the database, and performing key path screening on the preliminary causal relationship chain based on the historical abnormal causal chain to obtain a screened causal relationship path; The fault nodes of the filtered causal relationship path are located through the historical abnormal nodes to obtain a fault point list.

6. The method for testing mobile phone functions according to claim 5, characterized in that: The preset causal relationship mining algorithm is used to explore and infer the abnormal causal relationship of the target mobile phone based on the probability distribution of the abnormal pattern, and a preliminary causal relationship chain is obtained, including: The time series feature extraction of the probability distribution of the abnormal pattern is performed through a preset bidirectional long short-term memory network to obtain an abnormal time series feature vector; wherein the abnormal time series feature vector includes the duration of the abnormality, the frequency of the abnormality and the severity of the abnormality; Performing causal relationship modeling on the abnormal time series feature vector to obtain a causal variable association graph; Using a preset graph attention network to dynamically calculate the weight of the causal variable association graph to obtain a weighted causal graph; Performing causal path exploration on the target mobile phone based on the weighted causal graph to obtain a set of candidate causal chains; Performing probability dependency analysis on the candidate causal chain set based on a conditional random field model to obtain a causal chain probability distribution; The entropy weight method is used to sort and infer the causal chain importance of the causal chain probability distribution to obtain a preliminary causal relationship chain.

7. The method for testing mobile phone functions according to claim 1, characterized in that: The target mobile phone is provided with multiple mobile phone function modules. The static analysis of each potential fault in the fault point list is performed to obtain function optimization suggestions, including: The key words of each potential fault in the fault point list are extracted by a feature matching algorithm to obtain a fault type mapping table; Scan the application code in the target mobile phone in sections based on the fault type mapping table to obtain a set of fault code blocks; The fault code block set is restored by using a preset code execution path analysis technology to obtain a fault triggering condition list; wherein the fault triggering condition list includes a user operation sequence, a system status condition, and a resource occupancy threshold; Classify and label the plurality of mobile phone function modules based on the fault trigger condition list to obtain a function risk level table; Input the functional risk level table into a preset similarity analysis algorithm to perform historical problem comparison and static analysis to obtain a problem solution knowledge set; Based on the problem solution knowledge set, repair suggestions are generated for each of the potential faults to obtain function optimization suggestions.

8. A mobile phone function testing device, characterized in that: include: The parsing module is used to perform feature parsing on the running logs in the application of the target mobile phone to obtain the original feature data set; An extraction module, used to perform multi-dimensional feature extraction on the original feature data set to obtain an application behavior feature sequence; A first analysis module is used to perform dynamic functional state analysis on the target mobile phone based on the application behavior feature sequence to obtain a functional state transition graph; An inference module, used to identify whether there is an abnormal pattern in the functional state transfer map through a recursive neural network, and if so, perform causal relationship inference on the functional state transfer map based on the abnormal pattern to obtain a fault point list; The second analysis module is used to perform static analysis on each potential fault in the fault point list to obtain function optimization suggestions.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images