Software improvement suggestion generation method based on fault feature matching and role orientation
By standardizing and analyzing software fault data, and combining vertical causal relationships with horizontal relationships, role-oriented improvement suggestions are generated. This solves the problem of insufficient fault pattern mining and targeted suggestions in existing technologies, and achieves efficient software improvement and fault handling.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINESE PEOPLES LIBERATION ARMY UNIT 93184
- Filing Date
- 2025-10-17
- Publication Date
- 2026-05-22
AI Technical Summary
Existing technologies lack the ability to extract and analyze the multi-dimensional features of software fault data, and cannot take into account the differences in knowledge background, division of responsibilities and core needs of different user roles. This makes it difficult to discover fault patterns, and the resulting improvement suggestions lack pertinence and practicality.
By acquiring raw fault data and analytical description data, and after normalizing them, a pre-trained domain natural language model is used for word segmentation and semantic understanding. Combined with bidirectional long short-term memory network analysis, the contextual dependencies between features are analyzed to construct vertical causal relationships and horizontal relationships, generating role-oriented improvement suggestions.
It enables the systematic reuse of fault data, improves the pertinence and practicality of improvement suggestions, reduces testing costs and manpower input, ensures that the content of suggestions matches the role requirements, and improves the accuracy and systematicness of fault handling.
Smart Images
Figure CN121233477B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of data retrieval and software automation, and in particular to a method for generating software improvement suggestions based on fault feature matching and role orientation. Background Technology
[0002] With the development of equipment and the continuous improvement of informatization, the role of software is becoming increasingly important, and the quality requirements for software are also rising. Rapid repair of software faults and comprehensive testing are crucial aspects of ensuring software quality. Meanwhile, in practical applications, as software complexity increases, software defects become more subtle, and their impact and consequences become more severe. Software-related quality problems are becoming increasingly prominent, and homogeneous problems are constantly occurring. There is an urgent need to study software fault data exposed during testing and use, discover fault patterns, and propose corresponding fault prevention and quality assurance methods. Developing methods capable of systematically analyzing fault data, automatically mining fault feature correlations, and intelligently generating improvement suggestions has become an urgent technical requirement for improving the efficiency of software quality engineering.
[0003] Chinese Patent Publication No. CN114048882A discloses a method, apparatus, equipment, and computer-readable storage medium for railway fault handling and maintenance decision-making suggestions. The method includes acquiring status data; inputting the status data into a fault detection model to obtain fault data from the status data; generating corresponding scene graph features based on the fault data; fusing the fault data and the corresponding scene graph features to obtain fused features; and matching the fused features with maintenance decision-making suggestion templates in an maintenance scheme database to obtain corresponding maintenance decision-making suggestions. This approach improves the efficiency and accuracy of rail transit data analysis and fault diagnosis.
[0004] However, the following problems still exist in the existing technology:
[0005] 1. Existing technologies lack the ability to extract and correlate multi-dimensional features of software fault data, and fail to transform isolated fault cases into structured networks with causal relationships, making it difficult to discover fault patterns and develop common improvement strategies for similar problems.
[0006] 2. Existing technologies fail to consider the differences in knowledge background, division of responsibilities and core needs among different user roles, and cannot dynamically adjust the analysis dimensions and output content according to the role. The resulting improvement suggestions lack pertinence, practicality and guidance. Summary of the Invention
[0007] To address these issues, this invention provides a software improvement suggestion generation method based on fault feature matching and role orientation. This method overcomes the problems in existing technologies, such as the lack of ability to extract and correlate multi-dimensional features of software fault data, failure to consider the differences in knowledge background, division of responsibilities, and core demands of different user roles leading to difficulty in discovering fault patterns, inability to formulate common improvement strategies for similar problems, and lack of pertinence, practicality, and guidance in the generated improvement suggestions.
[0008] To achieve the above objectives, this invention provides a method for generating software improvement suggestions based on fault feature matching and role-oriented approach, comprising:
[0009] The uploaded raw fault data and analysis description data are obtained and standardized to obtain standardized fault events.
[0010] Based on a pre-trained domain natural language model, the normalized fault events are segmented and semantically understood to identify and extract preliminary feature information.
[0011] The preliminary feature information is analyzed for contextual dependency by a bidirectional long short-term memory network to capture internal dependencies, and association rules are defined based on the preliminary feature information and internal dependencies.
[0012] Based on the association rules, a vertical causal relationship is established among the internal features of the standardized fault events, similarity calculation is performed to determine the horizontal association value, and the existence of a horizontal association is determined based on the horizontal threshold.
[0013] In response to software improvement requests initiated by different user roles, based on the key input content corresponding to each role, the corresponding association chain is triggered in the vertical causal association, and other potential faults are identified based on the horizontal association.
[0014] Based on trigger-based association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated.
[0015] The vertical causal relationship includes the transmission chain, the responsibility chain, and the correction chain.
[0016] Furthermore, the uploaded raw fault data and analysis / description data are obtained, including:
[0017] Obtain raw fault data including fault occurrence time, fault location, fault device type, fault software type, fault type, and fault symptoms;
[0018] Obtain analytical descriptive data that includes possible causes of the fault and fault prevention suggestions.
[0019] Furthermore, standardization processes will be implemented, including...
[0020] The integrity of the original fault data and the analysis and description data is verified.
[0021] If the integrity check passes, the original fault data and analysis description data will be formatted in a standardized manner.
[0022] If the integrity check fails, a data completion request is triggered, sending a notification containing the missing data to the corresponding data source.
[0023] Further, preliminary feature information is identified and extracted, including,
[0024] The pre-trained domain natural language model is used to perform word segmentation and named entity recognition on the normalized fault raw data and analysis description data in order to identify entities and keywords related to predefined feature categories;
[0025] Based on the identified entities and keywords, the entities and keywords are mapped to corresponding preliminary feature categories according to predefined feature classification rules;
[0026] The preliminary feature categories include scene features, equipment features, fault features, root cause features, and verification features.
[0027] Furthermore, contextual dependency analysis is performed on the preliminary feature information to capture internal dependencies. Association rules are then defined based on the preliminary feature information and internal dependencies, including:
[0028] The initial feature information is input into a bidirectional long short-term memory network. By analyzing the contextual information of the initial feature information, the semantic dependencies between and within each feature are captured, forming an internal dependency network.
[0029] Based on the semantic association strength and directionality revealed by the internal dependency network, association rules for constructing vertical causal associations are defined.
[0030] Among them, the association rules include the transmission chain rules, the responsibility chain rules, and the correction chain rules.
[0031] Furthermore, based on the association rules, vertical causal relationships are established among the internal features of the normalized fault events; similarity calculations are performed to determine horizontal association values; and a horizontal threshold is used to determine whether a horizontal association exists, including...
[0032] Based on the transmission chain rule, a causal transmission path is constructed from the fault feature to the root cause feature;
[0033] Based on the chain of responsibility rules, construct an attribution path from the fault characteristics to the responsible module;
[0034] Based on the correction chain rules, a fault correction path is constructed that points from the root cause feature to the improved feature and from the improved feature to the verification feature;
[0035] Calculate the similarity between each pair of faults, including equipment fault similarity, software defect similarity, and introduction phase similarity;
[0036] The weighted sum of equipment fault similarity, software defect similarity, and introduction stage similarity is used as the horizontal correlation value;
[0037] If the horizontal correlation value is greater than the preset horizontal threshold, it is determined that there is a horizontal correlation between faults.
[0038] Furthermore, in response to software improvement requests initiated by different user roles, based on the key input content corresponding to each role, corresponding association chains are triggered in the vertical causal relationship, and other potential faults are identified based on the horizontal association, including...
[0039] In response to software improvement requests initiated by different user roles, identify the user role identifier;
[0040] Based on the predefined role feature mapping relationship, determine the key input feature type corresponding to the user role identifier;
[0041] The actual content input by the user is matched with the key input feature type. Based on the matching result, the corresponding association chain in the vertical causal association is activated, and based on the horizontal relationship of the matching result, other potential faults are determined.
[0042] Furthermore, based on triggered association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated, including...
[0043] If the user role is identified as a user unit, its key input feature type is determined to be a description of the fault phenomenon;
[0044] The system matches the user-input description of the fault phenomenon with the fault features in the fault database to identify associated standardized fault events.
[0045] Activate the transmission chain in the associated normalized fault events to obtain causal analysis information from fault characteristics to root cause characteristics;
[0046] Based on the causal analysis information, user suggestions for troubleshooting are derived, including specific operational steps and potential other faults.
[0047] Furthermore, based on triggered association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated, which also includes...
[0048] If the user role is identified as a development unit, its key input feature type is determined to be fault root cause features and responsible module information;
[0049] The root cause characteristics and responsible module information input by the user are matched with the corresponding characteristics in the fault database to identify the associated standardized fault events.
[0050] Activate the correction chain in the associated normalized fault events to obtain a suggested solution path from root cause features to improvement features;
[0051] Based on the proposed solution path, an improved and optimized user suggestion containing the corresponding code location is generated using a pre-trained natural language model.
[0052] Furthermore, based on triggered association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated, which also includes...
[0053] If the user role is identified as an evaluation unit, its key input feature types are determined to be device features, scene features, root cause features, and verification features.
[0054] The user-input feature combination is matched with the feature network in the fault database to identify associated normalized fault events.
[0055] Based on the correction chain in the associated normalized fault events, and based on equipment characteristics and fault characteristics, other related fault events are retrieved, and software evaluation suggestions containing fault reproduction test cases are generated.
[0056] Compared to existing technologies, this invention standardizes the raw fault data and analytical description data to form standardized fault events. It then utilizes a pre-trained domain-specific natural language model to segment and semantically understand the event content, extracting preliminary feature information. Furthermore, it analyzes the contextual dependencies between features using a bidirectional long short-term memory network, constructing vertical causal relationships and horizontal relationships including transmission chains, responsibility chains, and correction chains. Based on this, it can trigger corresponding association chains according to key content input by the user's role, and output targeted improvement suggestions using natural language generation technology. This invention transforms fragmented fault cases into a structured knowledge network, enabling the systematic reuse of fault data. This improves the targeting and efficiency of software improvements while effectively reducing testing costs and manpower investment.
[0057] In particular, this invention considers a role-oriented improvement suggestion generation mechanism. In practice, user units, development units, and evaluation units differ in their knowledge background and core needs, making it difficult for a single form of improvement suggestion to meet the requirements of each stage. This invention identifies user role identifiers and predefines their key input feature types, enabling targeted triggering of corresponding association chains in a vertical causal relationship network. This mechanism ensures that the generated improvement methods match the role requirements in both content and form, realizing different output strategies from emergency operations to code modifications, improving the practicality and guidance of the suggestions, and solving the problem of insufficient targeting in existing technologies.
[0058] In particular, this invention considers the structured association and reuse mechanism of fault data features. In reality, software fault data is mostly fragmented, with cases existing in isolation. This invention extracts multi-dimensional features through natural language models and long short-term memory networks, and defines association rules such as transmission chains, responsibility chains, and correction chains to construct a vertical association network that reveals the inherent causal logic of faults. This structured knowledge network transforms discrete cases into a searchable and reasonable experience system, providing a data foundation for the localization and analysis of similar faults and reducing the cost of repeated testing.
[0059] In particular, this invention considers the construction and triggering mechanism of vertical causal relationship chains between fault features. In reality, the intrinsic connection between the root cause, manifestation, and proposed solutions to a fault is difficult to capture. This invention defines three causal paths: a transmission chain, a responsibility chain, and a correction chain, respectively depicting the complete logical chains from fault features to root causes, from fault phenomena to responsible modules, and from root causes to improvement and verification methods. The system can automatically identify and trigger the corresponding chains in the constructed relationship network based on key features input by the user. This mechanism connects discrete feature points into causal relationships with clear directions, ensuring that the generation of improvement suggestions is not based on simple matching but on reasoning about the inherent logic of the fault, thereby improving the technical rationality and reliability of the improvement suggestions.
[0060] In particular, this invention considers a fault analysis mechanism that combines vertical causal relationships with horizontal correlations. In reality, software faults often involve complex causal chains, and different faults may exhibit potential correlation patterns. This invention reveals the intrinsic mechanism of faults by establishing transmission chains, responsibility chains, and correction chains, while constructing a horizontal correlation network based on the similarity calculation of equipment characteristics, fault characteristics, and introduction stage characteristics. This combined vertical and horizontal architecture can deeply analyze the causal path of individual faults and identify potential related faults, achieving an analytical closed loop from problem localization to risk identification, thus improving the accuracy of fault handling and the system's reusability. Attached Figure Description
[0061] Figure 1This is a schematic diagram illustrating the steps of the software improvement suggestion generation method based on fault feature matching and role orientation according to an embodiment of the present invention;
[0062] Figure 2 This is a logic block diagram for integrity verification in an embodiment of the invention.
[0063] Figure 3 This is a logic block diagram illustrating the determination of a horizontal relationship between faults according to an embodiment of the invention.
[0064] Figure 4 A logic block diagram for determining the corresponding key input feature types for different user roles in the embodiments of the invention. Detailed Implementation
[0065] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.
[0066] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.
[0067] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.
[0068] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0069] Please see Figure 1 As shown, Figure 1 This is a schematic diagram illustrating the steps of a software improvement suggestion generation method based on fault feature matching and role orientation according to an embodiment of the present invention. The method includes:
[0070] Step S1: Obtain the uploaded original fault data and analysis description data for normalization processing to obtain normalized fault events;
[0071] Step S2: Based on a pre-trained domain natural language model, the normalized fault events are segmented and semantically understood to identify and extract preliminary feature information;
[0072] Step S3: Context dependency analysis is performed on the preliminary feature information using a bidirectional long short-term memory network to capture internal dependencies, and association rules are defined based on the preliminary feature information and internal dependencies.
[0073] Step S4: Based on the association rules, establish vertical causal relationships among the internal features of the standardized fault events, perform similarity calculations to determine the horizontal association value, and determine whether a horizontal association exists based on the horizontal threshold.
[0074] Step S5: In response to software improvement requests initiated by different user roles, trigger the corresponding association chain in the vertical causal association based on the key input content corresponding to each role, and determine other potential faults based on the horizontal association.
[0075] Step S6: Based on the triggered association chain, generate role-oriented user suggestions by combining the pre-trained natural language model;
[0076] The vertical causal relationship includes the transmission chain, the responsibility chain, and the correction chain.
[0077] Specifically, this involves acquiring the uploaded raw fault data and analysis / description data, including:
[0078] Obtain raw fault data including fault occurrence time, fault location, fault device type, fault software type, fault type, and fault symptoms;
[0079] Obtain analytical descriptive data that includes possible causes of the fault and fault prevention suggestions.
[0080] Specifically, the analytical description data is structured diagnostic and improvement information pre-analyzed and judged by those skilled in the art based on a summary of historical typical failures. Its content includes not only failure prevention recommendations (GA), but also analytical results from multiple dimensions such as failure causes (GR), test error causes (CR), failure detection methods (GF), development and improvement methods (JF), and failure introduction phase (JD).
[0081] Understandably, in practice, the original fault data and the analysis and description data are considered as a complete fault event, and the relevant data are stored in the fault database on a unit basis.
[0082] Specifically, this involves standardization processes, including...
[0083] The integrity of the original fault data and the analysis and description data is verified.
[0084] Please see Figure 2 As shown, Figure 2 The logical block diagram for integrity verification is as follows: if the integrity verification passes, the original fault data and analysis description data are standardized and formatted.
[0085] If the integrity check fails, a data completion request is triggered, sending a notification containing the missing data to the corresponding data source.
[0086] Specifically, the integrity verification should first check the completeness of necessary fields in the original fault data and analysis description data to verify whether key data items such as fault occurrence time, fault equipment type, fault phenomenon, and fault cause are missing. Secondly, the data format standardization should be verified by checking whether the value format of each field in the original fault data and analysis description data conforms to the standard according to the predefined data dictionary and format rules.
[0087] Specifically, identifying and extracting preliminary feature information includes,
[0088] The pre-trained domain natural language model is used to perform word segmentation and named entity recognition on the normalized fault raw data and analysis description data in order to identify entities and keywords related to predefined feature categories;
[0089] Based on the identified entities and keywords, the entities and keywords are mapped to corresponding preliminary feature categories according to predefined feature classification rules;
[0090] The preliminary feature categories include scene features, equipment features, fault features, root cause features, and verification features.
[0091] In implementation, a locally deployed DeepSeek-R1 natural language model is used, and it is trained with relevant knowledge from fields such as software development, testing, and engineering applications to form a pre-trained natural language model that incorporates knowledge from these fields.
[0092] For example, in some possible implementations, pre-trained natural language models identify and segment technical terms such as “1394 bus” and “CRC check error”.
[0093] It is understood that no specific limitations are placed on the form of the natural language model; the model is primarily used to understand fault data terminology and generate subsequent specialized language. Those skilled in the art can choose or train a natural language model capable of achieving this purpose.
[0094] It is understandable that those skilled in the art can set up corresponding fault feature libraries based on equipment characteristics to facilitate the establishment of a mapping relationship between equipment characteristics and fault features. In implementation, equipment characteristics are first identified, and then fault features are identified and classified according to the fault feature library corresponding to the equipment characteristics, thereby reducing the workload of comparison and improving the accuracy and efficiency of fault feature matching. For example, for a certain model of monitor, a fault feature library with fault features such as "screen tearing," "screen distortion," "display flickering," and "black screen" can be pre-set.
[0095] Specifically, context dependency analysis is performed on the preliminary feature information to capture internal dependencies, and association rules are defined based on the preliminary feature information and internal dependencies, including:
[0096] The initial feature information is input into a bidirectional long short-term memory network. By analyzing the contextual information of the initial feature information, the semantic dependencies between and within each feature are captured, forming an internal dependency network.
[0097] Based on the semantic association strength and directionality revealed by the internal dependency network, association rules for constructing vertical causal associations are defined.
[0098] Among them, the association rules include the transmission chain rules, the responsibility chain rules, and the correction chain rules.
[0099] Specifically, the bidirectional long short-term memory network is a special type of recurrent neural network used to capture contextual dependencies. In this invention, after a pre-trained natural language model segments the fault text, the resulting preliminary feature information is input into the bidirectional long short-term memory network. The bidirectional long short-term memory network processes the sequence of preliminary feature information in two directions: forward and backward. This captures the contextual relationships between technical terms in the fault description, thereby deeply mining the complex intrinsic relationships between features such as fault phenomena, equipment environment, and root causes, laying the foundation for subsequent association rule construction.
[0100] In some possible implementations, after obtaining text such as "the right main display screen is distorted when the ground is powered on, but the recorded video is displayed normally", the dependencies between terms such as "display screen distortion", "right main display", and "recorded video" are identified through bidirectional long short-term memory network processing. This helps to understand the specific scenario and phenomenon association of the fault, providing support for accurately extracting "equipment characteristics" and "fault characteristics".
[0101] Specifically, the internal dependency network is a feature association logic representation formed through bidirectional long short-term memory network analysis. The internal dependency network uses the extracted preliminary feature information as nodes and the semantic dependencies between features as edges.
[0102] Specifically, the association rules for constructing vertical causal relationships are three types of inference rules defined based on the semantic relationships revealed by the internal dependency relationship network, including the transmission chain rule, the responsibility chain rule, and the correction chain rule.
[0103] In some possible implementations, the conduction chain rule is used to construct paths such as "display screen flickering and system crash → chip power supply interference leading to false reset"; the responsibility chain rule is used to construct paths such as "screen flickering → hardware module"; and the correction chain rule is used to construct paths such as "chip power supply interference leading to false reset → detection mechanism and active reset conditions → hardware application environment testing". Association rules link nodes in different internal dependency networks in a logical order, forming a vertical causal chain with engineering significance.
[0104] Specifically, based on the association rules, vertical causal relationships are established among the internal features of the normalized fault events; similarity calculations are performed to determine horizontal association values; and a horizontal threshold is used to determine whether a horizontal association exists.
[0105] Based on the transmission chain rule, a causal transmission path is constructed from the fault feature to the root cause feature;
[0106] Based on the chain of responsibility rules, construct an attribution path from the fault characteristics to the responsible module;
[0107] Based on the correction chain rules, a fault correction path is constructed that points from the root cause feature to the improved feature and from the improved feature to the verification feature;
[0108] Calculate the similarity between each pair of faults, including equipment fault similarity, software defect similarity, and introduction phase similarity;
[0109] The weighted sum of equipment fault similarity, software defect similarity, and introduction stage similarity is used as the horizontal correlation value;
[0110] The vertical causal relationship includes three core paths: the transmission path represents the causal transmission relationship from fault features to root cause features; the responsibility path represents the attribution relationship from fault features to responsible modules; and the correction path represents the correction and verification relationship from root cause features through improved features to verification features. Together, these three constitute a complete fault causal analysis chain.
[0111] Understandably, while establishing only vertical causal relationships can provide in-depth analysis of the underlying mechanisms of individual failures, it cannot achieve cross-failure knowledge reuse; conversely, relying solely on horizontal relationships can identify similar failures, but it is difficult to reveal the root cause and corrective path of the failure. This invention combines vertical causal relationships with horizontal relationships, enabling in-depth analysis of current failures through transmission chains, responsibility chains, and correction chains, while also leveraging horizontal relationships to discover potential related failures, thus achieving the systematic reuse of historical experience.
[0112] Please see Figure 3 As shown, Figure 3 This is a logic block diagram for determining whether there is a horizontal relationship between faults according to an embodiment of the invention. If the horizontal relationship value is greater than a preset horizontal threshold, it is determined that there is a horizontal relationship between faults.
[0113] Specifically, the calculation of equipment fault similarity, software defect similarity, and introduction stage similarity is achieved by converting the feature terms corresponding to each fault event into feature vectors using a pre-trained domain natural language model, and then calculating the cosine similarity between the vectors. For equipment fault similarity, the similarity between the equipment feature vector and the fault feature vector of two fault events is calculated separately, and the higher value is taken as the equipment fault similarity; for software defect similarity, the similarity between the improved feature vectors of two fault events is calculated; for introduction stage similarity, the similarity between the stage feature vectors of two fault events is calculated.
[0114] Specifically, similarity .in, For equipment fault similarity, For software defect similarity, To introduce stage similarity.
[0115] Specifically, in implementation, the weighted weight of equipment failure similarity The weighted average of software defect similarity is 0.4. The weighted average of stage similarity is set to 0.4. It is 0.2.
[0116] Specifically, based on historical fault datasets, statistical analysis is performed, and those skilled in the art select a horizontal correlation value that balances recall and precision as the horizontal threshold; in implementation, the horizontal threshold is set as follows: .
[0117] Please see Figure 4 As shown, Figure 4A logical block diagram for determining corresponding key input feature types for different user roles in embodiments of the invention; in response to software improvement requests initiated by different user roles, triggering corresponding association chains in the vertical causal association based on the key input content corresponding to each role; and identifying potential other faults based on horizontal association, including...
[0118] In response to software improvement requests initiated by different user roles, identify the user role identifier;
[0119] Based on the predefined role feature mapping relationship, determine the key input feature type corresponding to the user role identifier;
[0120] The actual content input by the user is matched with the key input feature type. Based on the matching result, the corresponding association chain in the vertical causal association is activated, and based on the horizontal relationship of the matching result, other potential faults are determined.
[0121] Calculate the horizontal correlation value between the matched fault events and other events in the database, and filter out fault events that exceed the preset potential threshold, identifying them as potential other faults.
[0122] Specifically, the preset potential threshold is determined in the following way: first, the arithmetic mean of the horizontal correlation values of all valid fault pairs is calculated based on the historical fault dataset as a baseline threshold; then, the baseline threshold is multiplied by an amplification factor greater than 1 to obtain the final preset potential threshold.
[0123] In practice, the amplification factor ranges from 1.05 to 1.15, and the specific value is determined by those skilled in the art based on the requirements for correlation accuracy. By appropriately amplifying the benchmark threshold, weakly correlated fault events can be filtered out, ensuring that the identified potential other faults have high reference value.
[0124] Specifically, the key input feature type corresponding to the user unit is a description of the fault phenomenon, which triggers the transmission chain; the key input feature type corresponding to the development unit is the root cause feature and responsibility module of the fault, which triggers the responsibility chain; and the key input feature type corresponding to the evaluation unit is equipment feature, scenario feature, root cause feature and verification feature, which triggers the correction chain.
[0125] Specifically, based on triggered association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated, including:
[0126] If the user role is identified as a user unit, its key input feature type is determined to be a description of the fault phenomenon;
[0127] The system matches the user-input description of the fault phenomenon with the fault features in the fault database to identify associated standardized fault events.
[0128] Activate the transmission chain in the associated normalized fault events to obtain causal analysis information from fault characteristics to root cause characteristics;
[0129] Based on the causal analysis information, user suggestions for troubleshooting are derived, including specific operational steps and potential other faults.
[0130] Specifically, the derived fault-solving user suggestions, which include specific operational steps and potential other faults, refer to generating temporary avoidance measures or emergency operation steps with clear operational guidance through a pre-trained natural language model based on the causal analysis information provided by the transmission chain. A standardized fault avoidance template library is pre-constructed by those skilled in the art, establishing a structured mapping relationship between typical fault phenomena and corresponding emergency response measures for reference by the pre-trained natural language model.
[0131] Specifically, based on triggered association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated, which also includes:
[0132] If the user role is identified as a development unit, its key input feature type is determined to be fault root cause features and responsible module information;
[0133] The root cause characteristics and responsible module information input by the user are matched with the corresponding characteristics in the fault database to identify the associated standardized fault events.
[0134] Activate the correction chain in the associated normalized fault events to obtain a suggested solution path from root cause features to improvement features;
[0135] Based on the proposed solution path, an improved and optimized user suggestion containing the corresponding code location is generated using a pre-trained natural language model.
[0136] In some possible implementations, improved and optimized user suggestions can also be based on other pre-trained models to generate code modification, algorithm optimization, and fault prevention suggestions for development units and users to refer to.
[0137] Specifically, based on triggered association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated, which also includes:
[0138] If the user role is identified as an evaluation unit, its key input feature types are determined to be device features, scene features, root cause features, and verification features.
[0139] The user-input feature combination is matched with the feature network in the fault database to identify associated normalized fault events.
[0140] Based on the correction chain in the associated normalized fault events, and based on equipment characteristics and fault characteristics, other related fault events are retrieved, and software evaluation suggestions containing fault reproduction test cases are generated.
[0141] In some possible implementations, the software improvement recommendations also include test environment optimization and test completeness analysis. By comparing the test environment configuration and test coverage of related failure events, the shortcomings of the current test plan are identified, and targeted test environment improvement measures and test case supplementation suggestions are proposed, thereby forming a systematic test improvement plan.
[0142] It is understood that the software evaluation recommendations may also include potential fault warnings and verification suggestions. Based on potential fault events identified through horizontal correlations, other potential fault modes that may occur under the current fault environment are inferred, and corresponding detection methods and verification steps are generated for these potential faults, forming predictive maintenance recommendations.
[0143] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.
Claims
1. A method for generating software improvement suggestions based on fault feature matching and role-oriented approach, characterized in that, include: The uploaded raw fault data and analysis description data are obtained and standardized to obtain standardized fault events. Based on a pre-trained domain natural language model, the normalized fault events are segmented and semantically understood to identify and extract preliminary feature information. include, The pre-trained domain natural language model is used to perform word segmentation and named entity recognition on the normalized fault raw data and analysis description data to identify entities and keywords related to predefined feature categories; based on the identified entities and keywords, the entities and keywords are mapped to the corresponding preliminary feature categories according to predefined feature classification rules. Contextual dependency analysis is performed on the preliminary feature information using a bidirectional long short-term memory network to capture internal dependencies. Association rules are then defined based on the preliminary feature information and these internal dependencies, including... The initial feature information is input into a bidirectional long short-term memory network. By analyzing the contextual information of the initial feature information, the semantic dependencies between and within each feature are captured, forming an internal dependency network. Based on the semantic association strength and directionality revealed by the internal dependency network, association rules for constructing vertical causal associations are defined. Based on the association rules, a vertical causal relationship is established among the internal features of the standardized fault events, similarity calculation is performed to determine the horizontal association value, and the existence of a horizontal association is determined based on the horizontal threshold. In response to software improvement requests initiated by different user roles, based on the key input content corresponding to each role, the corresponding association chain is triggered in the vertical causal association, and other potential faults are identified based on the horizontal association. Based on trigger-based association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated. The vertical causal relationship includes a transmission chain, a responsibility chain, and a correction chain. The preliminary feature categories include scenario features, equipment features, fault features, root cause features, and verification features. The association rules include transmission chain rules, responsibility chain rules, and correction chain rules.
2. The software improvement suggestion generation method based on fault feature matching and role orientation according to claim 1, characterized in that, Obtain the uploaded raw fault data and analysis / description data, including: Obtain raw fault data including fault occurrence time, fault location, fault device type, fault software type, fault type, and fault symptoms; Obtain analytical descriptive data that includes possible causes of the fault and fault prevention suggestions.
3. The software improvement suggestion generation method based on fault feature matching and role orientation according to claim 1, characterized in that, Standardization processes, including, The integrity of the original fault data and the analysis and description data is verified. If the integrity check passes, the original fault data and analysis description data will be formatted in a standardized manner. If the integrity check fails, a data completion request is triggered, sending a notification containing the missing data to the corresponding data source.
4. The software improvement suggestion generation method based on fault feature matching and role orientation according to claim 1, characterized in that, Based on the association rules, vertical causal relationships are established among the internal features of the normalized fault events. Similarity calculations are performed to determine horizontal association values, and a horizontal threshold is used to determine whether a horizontal association exists. include, Based on the transmission chain rule, a causal transmission path is constructed from the fault feature to the root cause feature; Based on the aforementioned chain of responsibility rules, an attribution path is constructed from the fault characteristics to the responsible module; Based on the correction chain rules, a fault correction path is constructed that points from the root cause feature to the improved feature and from the improved feature to the verification feature; Calculate the similarity between each pair of faults, including equipment fault similarity, software defect similarity, and introduction phase similarity; The weighted sum of equipment fault similarity, software defect similarity, and introduction stage similarity is used as the horizontal correlation value; If the horizontal correlation value is greater than the preset horizontal threshold, it is determined that there is a horizontal correlation between faults.
5. The software improvement suggestion generation method based on fault feature matching and role orientation according to claim 1, characterized in that, In response to software improvement requests initiated by different user roles, based on the key input content corresponding to each role, corresponding association chains are triggered in the vertical causal relationship, and other potential faults are identified based on the horizontal association, including... In response to software improvement requests initiated by different user roles, identify the user role identifier; Based on the predefined role feature mapping relationship, determine the key input feature type corresponding to the user role identifier; The actual content input by the user is matched with the key input feature type. Based on the matching result, the corresponding association chain in the vertical causal association is activated, and based on the horizontal relationship of the matching result, other potential faults are determined.
6. The software improvement suggestion generation method based on fault feature matching and role orientation according to claim 1, characterized in that, Based on trigger-based association chains, and combined with a pre-trained natural language model, role-oriented user suggestions are generated, including: If the user role is identified as a user unit, its key input feature type is determined to be a description of the fault phenomenon; The system matches the user-input description of the fault phenomenon with the fault features in the fault database to identify associated standardized fault events. Activate the transmission chain in the associated normalized fault events to obtain causal analysis information from fault characteristics to root cause characteristics; Based on the causal analysis information, user suggestions for troubleshooting are derived, including specific operational steps and potential other faults.
7. The software improvement suggestion generation method based on fault feature matching and role orientation according to claim 1, characterized in that, Based on trigger-based association chains, and combined with pre-trained natural language models, role-oriented user suggestions are generated, which also include: If the user role is identified as a development unit, its key input feature type is determined to be fault root cause features and responsible module information; The root cause characteristics and responsible module information input by the user are matched with the corresponding characteristics in the fault database to identify the associated standardized fault events. Activate the correction chain in the associated normalized fault events to obtain a suggested solution path from root cause features to improvement features; Based on the proposed solution path, an improved and optimized user suggestion containing the corresponding code location is generated using a pre-trained natural language model.
8. The software improvement suggestion generation method based on fault feature matching and role orientation according to claim 1, characterized in that, Based on trigger-based association chains, and combined with pre-trained natural language models, role-oriented user suggestions are generated, which also include: If the user role is identified as an evaluation unit, its key input feature types are determined to be device features, scene features, root cause features, and verification features. The user-input feature combination is matched with the feature network in the fault database to identify associated normalized fault events. Based on the correction chain in the associated normalized fault events, and based on device characteristics and fault characteristics, other related fault events are retrieved, and software evaluation suggestions containing fault reproduction test cases are generated.