Mobile terminal software information intelligent matching method and system

Abstract

The present invention provides a method and system for intelligent matching of mobile software information, which relates to the field of intelligent matching technology, including extracting feature vectors by deep learning processing of software requirement text, constructing a software requirement knowledge graph and generating a requirement dependency matrix; analyzing historical test environment data based on the matrix, extracting test environment feature vectors by using a graph neural network; inputting the requirement feature vector and the environment feature vector into a bidirectional long short-term memory network to construct an environment matching model; when there are new requirements, extracting features and calculating matching scores, selecting the optimal environment template for deployment, and continuously optimizing the model through online learning. The present invention improves the accuracy of software test environment matching, reduces the cost of environment configuration, and shortens the test cycle.

Description

Technical Field

[0001] The present invention relates to intelligent matching technology, and in particular to a method and system for intelligent matching of mobile terminal software information. Background Art

[0002] The rapid development of mobile Internet has spawned a large number of mobile software applications. The functions of these applications are becoming increasingly complex, and the requirements for the test environment are becoming higher and higher. In order to ensure the quality of software, a large number of tests are required during the software development process, and the construction and maintenance of the test environment is a time-consuming and complex task. The test environment usually includes various software and hardware resources such as operating systems, databases, middleware, container images, etc. How to quickly and accurately match the corresponding test environment according to software requirements has become the key to improving testing efficiency.

[0003] Traditional test environment matching methods mainly rely on manual experience, and testers need to manually select and configure the test environment according to software requirements. This method is inefficient, error-prone, and difficult to adapt to the rapid iteration development model. Some automated testing tools can match environments according to pre-defined rules, but these rules are usually difficult to cover all scenarios and have poor flexibility. Reliance on manual experience and low efficiency: Traditional environment matching methods require testers to manually select and configure the environment based on experience, which is time-consuming, labor-intensive, inefficient, and prone to human errors.

[0004] Difficult to adapt to complex requirements: For complex software requirements, it is difficult to match the appropriate test environment manually or through simple rules, especially when multiple dependencies and configuration parameters are involved. Lack of adaptive learning ability: Most existing automation tools are based on fixed rules or templates and cannot be dynamically adjusted and optimized according to actual usage, making it difficult to adapt to changing software requirements and test environments. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for intelligently matching software information on a mobile terminal, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention, Provides a method for intelligently matching mobile software information, including: After word segmentation, removal of stop words and entity annotations on the software requirement text information, a deep learning model is constructed, wherein the deep learning model extracts the software requirement feature vector through the text encoding layer, feature extraction layer and attention mechanism layer; a software requirement knowledge graph is constructed according to the association relationship between the software requirement feature vector and the entity, a mapping relationship between the function requirement node and the technology dependency node is established in the software requirement knowledge graph, and a requirement dependency matrix is ​​generated; Analyze historical test environment data based on the requirement dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data, and use a graph neural network to feature encode the container image configuration information, the runtime dependency information, and the middleware version information to generate a test environment feature vector; input the software requirement feature vector and the test environment feature vector into an environment matching model constructed by a bidirectional long short-term memory network, calculate the cosine similarity in vector space to obtain an environment matching score, and establish an association mapping model between requirement features and environment features; When receiving new software requirements, the deep learning model is used to extract new requirement feature vectors, the new requirement feature vectors are input into the environment matching model, and the matching score with the existing test environment is calculated based on the association mapping model; the test environment template with the highest matching score is selected for instantiation and deployment, and the environment operation status data is recorded; the environment operation status data is fed back to the environment matching model, and the parameter weights of the association mapping model are updated through online learning to achieve continuous optimization of the environment matching accuracy.

[0007] After word segmentation, removal of stop words and entity annotations on the software requirement text information, a deep learning model is constructed. The deep learning model extracts the software requirement feature vector through a text encoding layer, a feature extraction layer, and an attention mechanism layer, including: Acquire software requirement text information, use a word segmentation processing module to segment the software requirement text information, use a stop word filtering module to remove stop words in the software requirement text information, use an entity annotation module to identify professional terms and technical terms in the software requirement text information, and generate a preprocessed requirement text; Construct a deep learning model based on a hierarchical architecture, wherein the deep learning model includes a text encoding layer, a feature extraction layer, and an attention mechanism layer. The text encoding layer adopts a BERT pre-trained model, the feature extraction layer adopts a multi-head self-attention network, and the attention mechanism layer adopts a demand-oriented attention allocation strategy; Input the preprocessed demand text into the text encoding layer of the deep learning model, the text encoding layer semantically encodes the demand text based on pre-trained language knowledge, and outputs text semantic features; input the text semantic features into the feature extraction layer, the feature extraction layer generates a first attention weight by calculating the product of the query matrix, the key matrix and the value matrix, and extracts a multi-level text representation; The multi-level text representation is input into the attention mechanism layer, and the attention mechanism layer constructs an attention allocation matrix according to the demand domain knowledge, and allocates differentiated weights to different types of demand information; the multi-level text representation is weightedly fused based on the attention allocation matrix to generate a software requirement feature vector.

[0008] A software requirement knowledge graph is constructed according to the software requirement feature vector and the association relationship between entities, a mapping relationship between function requirement nodes and technology dependency nodes is established in the software requirement knowledge graph, and a requirement dependency matrix is ​​generated, including: Identify entities in the software requirement feature vector, extract associations between the entities, and construct a software requirement knowledge graph based on the associations between the software requirement feature vector and the entities, wherein the software requirement knowledge graph includes a set of functional requirement nodes, a set of technical dependency nodes, and a set of relationship edges between nodes, wherein the set of relationship edges includes dependency relationships, inheritance relationships, combination relationships, call relationships, and implementation relationships between entities; Establishing a mapping relationship between the function requirement node set and the technology dependency node set in the software requirement knowledge graph, inputting each function requirement node in the function requirement node set into a graph attention network, the graph attention network assigning a second attention weight to a neighboring node of each function requirement node, performing feature aggregation on the function requirement node based on the second attention weight, and generating a structured feature of the function requirement node; The structured features of the functional requirement nodes are spliced ​​with the corresponding technology dependency node features, and the spliced ​​features are input into a mapping strength calculation module, which outputs the association strength value between the functional requirement nodes and the technology dependency nodes; the association strength value is organized into a requirement dependency relationship matrix.

[0009] Analyzing historical test environment data based on the demand dependency matrix, extracting container image configuration information, runtime dependency information, and middleware version information from the historical test environment data, and using a graph neural network to feature encode the container image configuration information, the runtime dependency information, and the middleware version information to generate a test environment feature vector, including: Analyze historical test environment data based on the demand dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data; construct an environment information heterogeneous graph, wherein the nodes of the environment information heterogeneous graph include configuration instruction nodes in the container image configuration information, dependency package nodes in the runtime dependency information, and middleware nodes in the middleware version information, and the edges of the environment information heterogeneous graph represent associations between nodes; The environmental information heterogeneous graph is feature encoded by using a graph neural network, the graph neural network includes a feature extraction module, a type-aware attention module, and a feature aggregation module; the feature extraction module extracts features of the basic image information and the environmental configuration information of the configuration instruction node, the version information and constraint information of the dependency package node, and the version information and configuration information of the middleware node to generate heterogeneous node features; The heterogeneous node features are input into the type-aware attention module. The type-aware attention module assigns third attention weights to different types of nodes according to the node types, and performs weighted aggregation on the heterogeneous node features based on the third attention weights. The feature aggregation module performs intra-component aggregation and inter-component aggregation on the node features after weighted aggregation to generate a test environment feature vector.

[0010] The software requirement feature vector and the test environment feature vector are input into the environment matching model constructed by the bidirectional long short-term memory network, and the environment matching score is obtained by calculating the cosine similarity in the vector space, and an association mapping model between the requirement feature and the environment feature is established, including: Constructing a bidirectional long short-term memory network as an environment matching model, the bidirectional long short-term memory network includes an input layer, a bidirectional long short-term memory layer, a feature fusion layer, and a similarity calculation layer, the bidirectional long short-term memory layer includes a forward long short-term memory unit and a backward long short-term memory unit, the forward long short-term memory unit and the backward long short-term memory unit respectively include an input gate, a forget gate, and an output gate; The software requirement feature vector and the test environment feature vector are input into the input layer of the environment matching model, and the forward long short-term memory unit and the backward long short-term memory unit process the software requirement feature vector and the test environment feature vector in the forward and reverse directions respectively to obtain bidirectional feature information; the feature fusion layer fuses the hidden state vector output by the forward long short-term memory unit and the backward long short-term memory unit to generate a fused feature vector; A vector space cosine similarity calculation module is constructed, and the vector space cosine similarity calculation module includes a feature weight allocation unit and a similarity calculation unit; the feature weight allocation unit allocates a weight coefficient to the fused feature vector based on the feature importance to generate a weighted feature vector; the similarity calculation unit calculates the cosine similarity of the weighted feature vector in the vector space to obtain an environment matching score, and constructs an association mapping model between demand features and environment features based on the environment matching score.

[0011] The test environment template with the highest matching score is selected for instantiation and deployment, and the environment operation status data is recorded; the environment operation status data is fed back to the environment matching model, and the parameter weights of the association mapping model are updated through online learning to achieve continuous optimization of the environment matching accuracy, including: Selecting the test environment template with the highest environment matching score, building a deployment plan based on the test environment template, the deployment plan including resource allocation strategy, component installation sequence, and configuration parameter settings; instantiating the deployment test environment according to the deployment plan to generate an operating environment instance; Collecting the running status data of the running environment instance, the running status data includes infrastructure layer monitoring data, system layer monitoring data, application layer monitoring data, and business layer monitoring data; performing time alignment and data standardization processing on the running status data to generate environment running characteristics; The environmental operation characteristics are input into an environmental status assessment module, and the environmental status assessment module calculates environmental operation stability indicators, performance indicators, and resource utilization indicators through multi-dimensional time series analysis to generate an environmental health score; Constructing an online learning optimization unit, wherein the online learning optimization unit uses the difference between the environmental health score and the original environmental matching score as a loss function, and calculates gradient information based on the loss function; dynamically adjusting the parameter weights of the environmental matching model using an adaptive learning rate according to the gradient information; The updated environment matching model is used for a new test environment matching score to calculate the environment matching accuracy; when the environment matching accuracy is higher than a preset threshold, the parameter weight is recorded as the optimal weight; when the environment matching accuracy is lower than the preset threshold, the parameter optimization module is triggered to perform weight fine-tuning until the environment matching accuracy meets the optimization target.

[0012] According to a second aspect of the embodiments of the present invention, Provides a mobile software information intelligent matching system, including: The first unit is used to perform word segmentation processing on the software requirement text information, remove stop words and entity annotations, and then construct a deep learning model, wherein the deep learning model extracts the software requirement feature vector through the text encoding layer, the feature extraction layer, and the attention mechanism layer; construct a software requirement knowledge graph according to the association relationship between the software requirement feature vector and the entity, establish a mapping relationship between the function requirement node and the technology dependency node in the software requirement knowledge graph, and generate a requirement dependency matrix; The second unit is used to analyze the historical test environment data based on the requirement dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data, and use a graph neural network to feature encode the container image configuration information, the runtime dependency information, and the middleware version information to generate a test environment feature vector; input the software requirement feature vector and the test environment feature vector into an environment matching model constructed by a bidirectional long short-term memory network, calculate the cosine similarity of the vector space to obtain an environment matching score, and establish an association mapping model between requirement features and environment features; The third unit is used to extract a new requirement feature vector using the deep learning model when receiving a new software requirement, input the new requirement feature vector into the environment matching model, and calculate a matching score with the existing test environment based on the association mapping model; select the test environment template with the highest matching score for instantiation and deployment, and record the environment operation status data; feed back the environment operation status data to the environment matching model, and update the parameter weights of the association mapping model through online learning to achieve continuous optimization of the environment matching accuracy.

[0013] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0015] The beneficial effects of this application are as follows: 1. Improve the accuracy of test environment matching: This invention uses a deep learning model to extract software requirement features, and combines historical test environment data to build a correlation mapping model between requirement features and environment features. Through online learning, the model parameters are continuously optimized to achieve continuous improvement in environment matching accuracy, and can more accurately match the appropriate test environment for new requirements.

[0016] 2. Improve the efficiency of test environment deployment: The present invention calculates the matching score between the new demand feature vector and the existing test environment, selects the test environment template with the highest matching degree for instantiation deployment, and avoids the tedious process of manually configuring the test environment, thereby improving the deployment efficiency of the test environment.

[0017] 3. Optimize resource utilization: The present invention avoids unnecessary test environment construction and resource waste through intelligent matching, and realizes optimal utilization of test resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the process of the method for intelligent matching of mobile terminal software information according to an embodiment of the present invention; Figure 2 The figure is a schematic diagram of the structure of the mobile terminal software information intelligent matching system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0021] Figure 1 FIG. 1 is a flow chart of a method for intelligently matching software information on a mobile terminal according to an embodiment of the present invention. Figure 1 As shown, the method includes: S11. After word segmentation, removal of stop words and entity annotations on the software requirement text information, a deep learning model is constructed, wherein the deep learning model extracts the software requirement feature vector through a text encoding layer, a feature extraction layer, and an attention mechanism layer; a software requirement knowledge graph is constructed according to the association relationship between the software requirement feature vector and the entity, a mapping relationship between the function requirement node and the technology dependency node is established in the software requirement knowledge graph, and a requirement dependency matrix is ​​generated; S12. Analyze historical test environment data based on the requirement dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data, and use a graph neural network to feature encode the container image configuration information, the runtime dependency information, and the middleware version information to generate a test environment feature vector; input the software requirement feature vector and the test environment feature vector into an environment matching model constructed by a bidirectional long short-term memory network, calculate the cosine similarity in vector space to obtain an environment matching score, and establish an association mapping model between requirement features and environment features; S13. When receiving new software requirements, use the deep learning model to extract new requirement feature vectors, input the new requirement feature vectors into the environment matching model, and calculate the matching score with the existing test environment based on the association mapping model; select the test environment template with the highest matching score for instantiation and deployment, and record the environment operation status data; feed the environment operation status data back to the environment matching model, update the parameter weights of the association mapping model through online learning, and achieve continuous optimization of the environment matching accuracy.

[0022] In an optional implementation, after word segmentation processing is performed on the software requirement text information, stop words and entity annotations are removed, a deep learning model is constructed, and the deep learning model extracts the software requirement feature vector through a text encoding layer, a feature extraction layer and an attention mechanism layer, including: Acquire software requirement text information, use a word segmentation processing module to segment the software requirement text information, use a stop word filtering module to remove stop words in the software requirement text information, use an entity annotation module to identify professional terms and technical terms in the software requirement text information, and generate a preprocessed requirement text; Construct a deep learning model based on a hierarchical architecture, wherein the deep learning model includes a text encoding layer, a feature extraction layer, and an attention mechanism layer. The text encoding layer adopts a BERT pre-trained model, the feature extraction layer adopts a multi-head self-attention network, and the attention mechanism layer adopts a demand-oriented attention allocation strategy; Input the preprocessed demand text into the text encoding layer of the deep learning model, the text encoding layer semantically encodes the demand text based on pre-trained language knowledge, and outputs text semantic features; input the text semantic features into the feature extraction layer, the feature extraction layer generates a first attention weight by calculating the product of the query matrix, the key matrix and the value matrix, and extracts a multi-level text representation; The multi-level text representation is input into the attention mechanism layer, and the attention mechanism layer constructs an attention allocation matrix according to the demand domain knowledge, and allocates differentiated weights to different types of demand information; the multi-level text representation is weightedly fused based on the attention allocation matrix to generate a software requirement feature vector.

[0023] The software requirement feature extraction method aims to extract key features from the software requirement text for subsequent software development tasks. Based on the deep learning model, this method can effectively capture the semantic information of the requirement text and perform feature weighting based on the requirement domain knowledge to improve the accuracy and effectiveness of feature representation.

[0024] First, preprocess the software requirement text information. Specifically, use a word segmentation tool (such as jieba) to segment the requirement text into individual words. Then, use a stop word list to filter out meaningless words, such as "de" (的), "a". Commonly used stop word lists can refer to open-source stop word libraries, such as the Chinese stop word list cn_stopwords. After filtering, the remaining words are "user", "need", "upload", "picture", "function". Finally, use an entity annotation tool (such as the BERT-CRF model) to identify professional terms and technical nouns in the requirement text. For example, identify "upload picture" as a functional entity and "user" as a role entity.

[0025] Next, build a deep learning model. This model adopts a hierarchical architecture, including a text encoding layer, a feature extraction layer, and an attention mechanism layer. The text encoding layer uses the BERT pre-trained model. Input the preprocessed requirement text into the BERT model, and the BERT model will convert each word into a corresponding word vector and generate text semantic features containing semantic information in combination with context information. For example, the word vector of "user" may contain its meaning as a system user, and the word vector of "upload picture" may contain semantics related to file transfer. The feature extraction layer uses a multi-head self-attention network. This network will calculate the product of the query matrix, key matrix, and value matrix to generate the first attention weight. Through the multi-head mechanism, text features can be extracted from different perspectives and these features can be combined into multi-level text representations. It can be understood that each "head" focuses on different aspects of the text. For example, one "head" focuses on entity relationships, and another "head" focuses on action descriptions. The attention mechanism layer adopts a requirement-oriented attention allocation strategy. According to requirement domain knowledge, such as learning from the structure of the requirement document, keywords, or expert experience, construct an attention allocation matrix. This matrix assigns different weights to different types of requirement information. For example, requirement information related to core functions will be given higher weights, while requirement information related to interface styles may have lower weights. Finally, based on the attention allocation matrix, perform weighted fusion on the multi-level text representations to generate the final software requirement feature vector. For example, assume that the multi-level text representation of "upload picture" contains information such as "file transfer" and "user interaction", and the attention allocation matrix pays more attention to the "file transfer" aspect, then the final feature vector will focus more on the features of "file transfer".

[0026] Taking "users need a function to upload pictures" as an example, after preprocessing, we get "user", "need", "upload", "picture", and "function". After BERT encoding, each word is converted into a corresponding word vector. The multi-head self-attention mechanism of the feature extraction layer will focus on the relationship between different words, such as the relationship between "upload" and "picture", and generate multi-level text representations. The attention mechanism layer assigns higher weights to "upload", "picture", and "function" according to the pre-set demand-oriented attention allocation strategy, such as paying more attention to functional requirements, and finally generates a software requirement feature vector focusing on the "upload picture function".

[0027] The solution of this application can: Improved the accuracy of demand feature extraction. Using the BERT pre-trained model and multi-head self-attention network, the semantic information of the demand text can be effectively captured, avoiding the excessive reliance on keywords in traditional methods. Enhanced the effectiveness of demand feature representation. The demand-oriented attention mechanism can assign differentiated weights to different types of demand information based on demand domain knowledge, making the extracted features more in line with actual needs. Improved the efficiency of subsequent software development tasks. Accurate and effective demand feature vectors can provide strong support for subsequent tasks such as demand analysis, software design and testing, thereby improving software development efficiency and quality.

[0028] In an optional implementation, a software requirement knowledge graph is constructed based on the association relationship between the software requirement feature vector and the entity, a mapping relationship between the function requirement node and the technology dependency node is established in the software requirement knowledge graph, and a requirement dependency matrix is ​​generated, including: Identify entities in the software requirement feature vector, extract associations between the entities, and construct a software requirement knowledge graph based on the associations between the software requirement feature vector and the entities, wherein the software requirement knowledge graph includes a set of functional requirement nodes, a set of technical dependency nodes, and a set of relationship edges between nodes, wherein the set of relationship edges includes dependency relationships, inheritance relationships, combination relationships, call relationships, and implementation relationships between entities; Establishing a mapping relationship between the function requirement node set and the technology dependency node set in the software requirement knowledge graph, inputting each function requirement node in the function requirement node set into a graph attention network, the graph attention network assigning a second attention weight to a neighboring node of each function requirement node, performing feature aggregation on the function requirement node based on the second attention weight, and generating a structured feature of the function requirement node; The structured features of the functional requirement nodes are spliced ​​with the corresponding technology dependency node features, and the spliced ​​features are input into a mapping strength calculation module, which outputs the association strength value between the functional requirement nodes and the technology dependency nodes; the association strength value is organized into a requirement dependency relationship matrix.

[0029] The software requirement knowledge graph construction and requirement dependency matrix generation method are specifically implemented as follows: First, preprocess the software requirement feature vector. The software requirement feature vector is usually described in text form, such as user stories, use case descriptions, etc. These texts are subjected to natural language processing operations such as word segmentation, part-of-speech tagging, and named entity recognition to extract key entities, such as functional modules, technical components, and performance indicators. For example, for the user story "As a user, I hope to be able to pay online so that I can complete my shopping more quickly", the functional module entity "online payment" and the performance indicator entity "faster" can be extracted.

[0030] Next, build a software requirement knowledge graph. Based on the extracted entities and the associations between them, build a software requirement knowledge graph. The knowledge graph consists of nodes and edges. Nodes represent entities, and edges represent the relationships between entities. The relationships between entities can be dependency relationships, inheritance relationships, combination relationships, call relationships, implementation relationships, etc. For example, the "online payment" functional module depends on the "payment gateway" technical component, then a dependency edge from the "online payment" node to the "payment gateway" node is established in the knowledge graph. Assuming that the "user authentication" functional module and the "security module" technical component are also identified, and "user authentication" depends on the "security module", and "online payment" also depends on the "security module", the corresponding nodes and edges will be added to the knowledge graph.

[0031] Then, the structured features of the function requirement nodes are generated based on the graph attention network. The function requirement nodes are input into the graph attention network. The graph attention network learns the attention weights between nodes based on the features of the node and its neighbor nodes. The degree of influence of the neighbor nodes on the central node is determined by the attention weights. The features of the neighbor nodes are weighted and summed according to the attention weights to obtain the structured features of the central node. For example, the neighbor nodes of the "online payment" node include "payment gateway" and "security module". The graph attention network learns the influence weights of these two neighbor nodes on the "online payment" node, and aggregates the neighbor node features based on the weights to generate the structured features of the "online payment" node.

[0032] After that, the association strength between the function requirement node and the technology dependency node is calculated. The structural features of the function requirement node are spliced ​​with the corresponding technology dependency node features. For example, the structural features [0.17, 0.59, 0.24] of the "online payment" node and the feature vector [0.2, 0.5, 0.3] of the technology node "payment gateway" on which it depends are spliced ​​into [0.17, 0.59, 0.24, 0.2, 0.5, 0.3]. The spliced ​​feature vector is input into the mapping strength calculation module, for example, a multi-layer perceptron model can be used. The mapping strength calculation module outputs the association strength value between the function requirement node and the technology dependency node, for example, the output is 0.85, indicating that the association strength between "online payment" and "payment gateway" is high.

[0033] Finally, generate the requirement dependency matrix. Organize the association strength values ​​between all functional requirement nodes and technical dependency nodes into a matrix form, namely the requirement dependency matrix. The rows of the matrix represent functional requirement nodes, the columns represent technical dependency nodes, and the matrix elements represent the association strength between the corresponding functional requirement nodes and technical dependency nodes. For example, assuming that the calculated result of the association strength between "user authentication" and "security module" is 0.92, then in the requirement dependency matrix, the value of the "online payment"-"payment gateway" position is 0.85, the value of the "user authentication"-"security module" position is 0.92, and the values ​​of other positions are also filled according to the calculation results.

[0034] The solution of this application can: Improve the efficiency and accuracy of requirements analysis. By building a software requirements knowledge graph, the correlation between requirements can be clearly displayed, which helps to deeply understand the requirements, reduce requirements ambiguity and omissions, and improve the efficiency and accuracy of requirements analysis. Support requirements change impact analysis. When requirements change, the impact of the change can be quickly assessed based on the requirements dependency matrix, and the affected functional modules and technical components can be identified, thereby reducing the risks brought by the requirements change. Promote requirements tracking and management. The requirements dependency matrix can serve as a basis for requirements tracking and management, helping the development team to better understand the dependencies between requirements and ensure the integrity and consistency of requirements.

[0035] In an optional implementation, historical test environment data is analyzed based on the demand dependency matrix, container image configuration information, runtime dependency information, and middleware version information are extracted from the historical test environment data, and the container image configuration information, the runtime dependency information, and the middleware version information are feature encoded using a graph neural network to generate a test environment feature vector, including: Analyze historical test environment data based on the demand dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data; construct an environment information heterogeneous graph, wherein the nodes of the environment information heterogeneous graph include configuration instruction nodes in the container image configuration information, dependency package nodes in the runtime dependency information, and middleware nodes in the middleware version information, and the edges of the environment information heterogeneous graph represent associations between nodes; The environmental information heterogeneous graph is feature encoded by using a graph neural network, the graph neural network includes a feature extraction module, a type-aware attention module, and a feature aggregation module; the feature extraction module extracts features of the basic image information and the environmental configuration information of the configuration instruction node, the version information and constraint information of the dependency package node, and the version information and configuration information of the middleware node to generate heterogeneous node features; The heterogeneous node features are input into the type-aware attention module. The type-aware attention module assigns third attention weights to different types of nodes according to the node types, and performs weighted aggregation on the heterogeneous node features based on the third attention weights. The feature aggregation module performs intra-component aggregation and inter-component aggregation on the node features after weighted aggregation to generate a test environment feature vector.

[0036] A test environment feature vector generation method based on graph neural network can effectively capture the complex dependencies of the test environment, so as to more accurately evaluate the similarity and compatibility of the test environment and improve the test efficiency and quality. The core of this method is to use graph neural network to encode various information in the test environment and generate environment feature vectors.

[0037] First, the historical test environment data is analyzed, and three types of key information are extracted from it based on the demand dependency matrix: container image configuration information, runtime dependency information, and middleware version information.

[0038] Next, a heterogeneous graph of environmental information is constructed. The node types of the graph include configuration instruction nodes, dependency package nodes, and middleware nodes. The edges in the graph represent the association between nodes. For example, if a container image uses a dependency package, an edge is established between the corresponding configuration instruction node and the dependency package node; if a middleware depends on a specific dependency package, an edge is established between the corresponding middleware node and the dependency package node. The type of edge can be defined based on the relationship between nodes, such as "use", "dependency", etc.

[0039] Then, the constructed environment information heterogeneous graph is feature encoded using the graph neural network. The graph neural network consists of three modules: feature extraction module, type-aware attention module and feature aggregation module. The feature extraction module extracts features from different types of nodes respectively. For example, for configuration instruction nodes, the basic image information and environment configuration information are extracted; for dependency package nodes, the version information and constraint information are extracted; for middleware nodes, the version information and configuration information are extracted. Through feature extraction, the node information is converted into a vector representation, such as converting "Tomcat9.0.58" into a digital vector. The type-aware attention module assigns attention weights to different types of nodes according to the node type. For example, if a certain test scenario pays more attention to the middleware version, a higher weight is assigned to the middleware node. Based on these weights, the node features are weighted aggregated to obtain the aggregated features of each node. The feature aggregation module performs intra-component aggregation and inter-component aggregation on the weighted aggregated node features to finally generate a test environment feature vector. Intra-component aggregation refers to aggregating node features within the same component, such as aggregating all node features within the same container image. Inter-component aggregation refers to aggregating the features of different components together to ultimately generate a feature vector for the entire test environment.

[0040] The solution of this application can: Improve the accuracy of test environment assessment: By capturing the complex dependencies of the test environment, the similarity and compatibility of the test environment can be more accurately assessed, thereby reducing the errors caused by differences in the test environment and improving the reliability of the test results. Improve test efficiency: By encoding the features of the test environment, the most suitable test environment can be quickly matched to avoid testing in incompatible environments, thereby saving test time and resources. Enhance test environment management: By representing the test environment as a feature vector, the test environment can be more easily classified, retrieved and managed, improving the reuse rate of the test environment and reducing the maintenance cost of the test environment.

[0041] In an optional implementation, the software requirement feature vector and the test environment feature vector are input into an environment matching model constructed by a bidirectional long short-term memory network, and an environment matching score is obtained by calculating the cosine similarity in the vector space, and an association mapping model between the requirement feature and the environment feature is established, including: Constructing a bidirectional long short-term memory network as an environment matching model, the bidirectional long short-term memory network includes an input layer, a bidirectional long short-term memory layer, a feature fusion layer, and a similarity calculation layer, the bidirectional long short-term memory layer includes a forward long short-term memory unit and a backward long short-term memory unit, the forward long short-term memory unit and the backward long short-term memory unit respectively include an input gate, a forget gate, and an output gate; The software requirement feature vector and the test environment feature vector are input into the input layer of the environment matching model, and the forward long short-term memory unit and the backward long short-term memory unit process the software requirement feature vector and the test environment feature vector in the forward and reverse directions respectively to obtain bidirectional feature information; the feature fusion layer fuses the hidden state vector output by the forward long short-term memory unit and the backward long short-term memory unit to generate a fused feature vector; A vector space cosine similarity calculation module is constructed, and the vector space cosine similarity calculation module includes a feature weight allocation unit and a similarity calculation unit; the feature weight allocation unit allocates a weight coefficient to the fused feature vector based on the feature importance to generate a weighted feature vector; the similarity calculation unit calculates the cosine similarity of the weighted feature vector in the vector space to obtain an environment matching score, and constructs an association mapping model between demand features and environment features based on the environment matching score.

[0042] The intelligent matching method between software requirements and test environment aims to more accurately match software requirements with appropriate test environments and improve test efficiency and accuracy. The core of this method is to build an environment matching model based on a bidirectional long short-term memory network (BiLSTM) and use vector space cosine similarity to calculate the environment matching score.

[0043] First, extract features from software requirements and test environments. For example, software requirement features may include: operating system type, required software dependencies, network configuration, hardware requirements, etc. Test environment features may include: installed operating system, installed software, network bandwidth, hardware configuration, etc. Convert these features into numerical vector representations. For example, the operating system type can be represented by one-hot encoding. Assuming that a software requirement requires a Windows operating system and a Java runtime environment, its feature vector can be represented as [1,0,1,0,...], where 1 represents the presence of Windows and Java, and 0 represents the absence of other operating systems and runtime environments. Similarly, if a test environment has a Windows operating system and a Python runtime environment, its feature vector can be represented as [1,0,0,1,...].

[0044] Next, a bidirectional long short-term memory network (BiLSTM) is constructed as an environment matching model. The model includes an input layer, a bidirectional long short-term memory layer, a feature fusion layer, and a similarity calculation layer. The software requirement feature vector and the test environment feature vector are input into the input layer of the model. The bidirectional long short-term memory layer includes a forward LSTM unit and a backward LSTM unit. The forward LSTM unit processes the input vector sequence from left to right to capture the forward and backward dependencies between the requirements and the environment features; the backward LSTM unit processes the input vector sequence from right to left to capture the reverse dependencies. For example, the software requirement feature vector is input into the forward LSTM unit in sequence, and each unit updates its internal state according to the current input and the output of the previous unit, and outputs a hidden state vector. The backward LSTM unit also performs similar processing, and finally obtains a hidden state vector sequence in two directions.

[0045] The feature fusion layer fuses the hidden state vectors output by the forward and backward LSTM units to generate a fused feature vector. A simple fusion method is to concatenate the hidden state vectors at corresponding positions. For example, the hidden state vector output by the forward LSTM unit at a certain position is [0.1, 0.2, 0.3], and the hidden state vector output by the backward LSTM unit at the same position is [0.4, 0.5, 0.6], then the fused feature vector is [0.1, 0.2, 0.3, 0.4, 0.5, 0.6].

[0046] Then, the environmental matching score is calculated using the vector space cosine similarity calculation module. This module includes a feature weight assignment unit and a similarity calculation unit. The feature weight assignment unit assigns weight coefficients to the fused feature vector based on the feature importance to generate a weighted feature vector. For example, if the operating system type is more important than the network bandwidth, a higher weight can be assigned to the feature dimension corresponding to the operating system type. The similarity calculation unit calculates the cosine similarity of the weighted feature vector in the vector space to obtain the environmental matching score.

[0047] Finally, a correlation mapping model between requirement characteristics and environment characteristics is constructed based on the environment matching score. This model can be used to quickly retrieve the test environment that matches the specific requirement.

[0048] The solution of this application can: Improve test efficiency: By accurately matching requirements and environment, reduce unnecessary test environment preparation time and improve test efficiency. Improve test coverage: This method can more comprehensively consider the characteristics of requirements and environment, thereby improving test coverage and discovering more potential defects. Reduce test costs: Reduce test costs by reducing unnecessary test environment construction and maintenance.

[0049] In an optional implementation, the test environment template with the highest matching score is selected for instantiation and deployment, and the environment operation status data is recorded; the environment operation status data is fed back to the environment matching model, and the parameter weights of the association mapping model are updated through online learning to achieve continuous optimization of the environment matching accuracy, including: Selecting the test environment template with the highest environment matching score, building a deployment plan based on the test environment template, the deployment plan including resource allocation strategy, component installation sequence, and configuration parameter settings; instantiating the deployment test environment according to the deployment plan to generate an operating environment instance; Collecting the running status data of the running environment instance, the running status data includes infrastructure layer monitoring data, system layer monitoring data, application layer monitoring data, and business layer monitoring data; performing time alignment and data standardization processing on the running status data to generate environment running characteristics; The environmental operation characteristics are input into an environmental status assessment module, and the environmental status assessment module calculates environmental operation stability indicators, performance indicators, and resource utilization indicators through multi-dimensional time series analysis to generate an environmental health score; Constructing an online learning optimization unit, wherein the online learning optimization unit uses the difference between the environmental health score and the original environmental matching score as a loss function, and calculates gradient information based on the loss function; dynamically adjusting the parameter weights of the environmental matching model using an adaptive learning rate according to the gradient information; The updated environment matching model is used for a new test environment matching score to calculate the environment matching accuracy; when the environment matching accuracy is higher than a preset threshold, the parameter weight is recorded as the optimal weight; when the environment matching accuracy is lower than the preset threshold, the parameter optimization module is triggered to perform weight fine-tuning until the environment matching accuracy meets the optimization target.

[0050] The test environment intelligent matching and online optimization method is used to efficiently select and deploy the most suitable test environment, and continuously optimize the accuracy of environment matching through online learning.

[0051] First, an environment matching model is constructed. The model inputs the required feature vector of the test task, such as the required CPU architecture, operating system type, memory size, network bandwidth, etc., and outputs a matching score with each available test environment template. The model can use machine learning algorithms, such as support vector machines, random forests, or deep neural networks. The initial model parameters can be set based on expert experience or historical data.

[0052] Assume that there are three test environment templates: Template A (4-core CPU, 8GB memory, Linux operating system), Template B (8-core CPU, 16GB memory, Windows operating system), Template C (16-core CPU, 32GB memory, Linux operating system). The requirement feature vector of a test task is: [4-core CPU, Linux operating system, 8GB memory]. The environment matching model calculates the matching scores of the task and the three templates respectively, for example: [Template A: 0.9, Template B: 0.2, Template C: 0.7].

[0053] Then, select the test environment template with the highest matching score. Based on the above case, select Template A for deployment. The deployment plan includes: allocating virtual machine resources with 4 CPU cores and 8GB memory; installing the Linux operating system; configuring network parameters, etc. According to the deployment plan, instantiate the deployment test environment and generate a running environment instance.

[0054] Next, collect the running status data of the running environment instance. The collected data includes: infrastructure layer monitoring data (CPU utilization, memory usage, disk IO, etc.), system layer monitoring data (number of processes, system load, etc.), application layer monitoring data (response time, error rate, etc.), and business layer monitoring data (transaction success rate, user visits, etc.). Assume that some of the collected data are as follows: CPU utilization 70%, memory usage 60%, response time 100ms, transaction success rate 98%.

[0055] Process the collected operation status data. First, perform time series alignment to ensure that the data of different indicators are aligned in time. Then perform data standardization, such as using the Z-score standardization method to convert data of different dimensions into standardized data with a mean of 0 and a standard deviation of 1. The processed data, for example: CPU utilization 0.8, memory usage 0.5, response time -0.2, transaction success rate 0.9.

[0056] The processed environment operation characteristics are input into the environment status assessment module. This module calculates environment operation stability indicators (such as the fluctuation degree of CPU utilization), performance indicators (such as average response time), and resource utilization indicators (such as the average value of CPU utilization) through multi-dimensional time series analysis. Finally, an environment health score is generated, such as 0.85.

[0057] Construct an online learning optimization unit. Use the difference between the environmental health score (0.85) and the original environmental matching score (0.9) as the loss function. Calculate the gradient information based on the loss function and use an adaptive learning rate, such as the Adam optimizer, to dynamically adjust the parameter weights of the environmental matching model.

[0058] The updated environment matching model is used for the new test environment matching score calculation, and the environment matching accuracy is calculated. For example, a new test task and its required feature vector are used for testing, and the ratio of the best matching template predicted by the model to the actual best matching template is calculated, which is the environment matching accuracy.

[0059] Set a preset threshold for the environment matching accuracy, such as 95%. When the environment matching accuracy is higher than the preset threshold, the current parameter weight is recorded as the optimal weight. When the environment matching accuracy is lower than the preset threshold, the parameter optimization module is triggered to fine-tune the weight, and the above online learning process is repeated until the environment matching accuracy meets the optimization target.

[0060] The solution of this application can: Improve the accuracy of environment matching: Through online learning, the environment matching model is continuously optimized, which improves the model's ability to understand the complex relationship between test tasks and test environments, so that it can more accurately select the most suitable test environment and reduce the loss of test efficiency caused by environment mismatch. Optimize resource utilization: By selecting the most matching test environment, it avoids over-allocation or idleness of resources, improves the utilization efficiency of test resources, and reduces test costs. Enhance environmental adaptability: The online learning mechanism enables the model to adapt to changing test requirements and environmental configurations, maintains the effectiveness of environmental matching, and enhances the robustness and adaptability of the system.

[0061] Figure 2 FIG. 1 is a schematic diagram of the structure of the mobile terminal software information intelligent matching system according to an embodiment of the present invention. Figure 2 As shown, the system comprises: The first unit is used to perform word segmentation processing on the software requirement text information, remove stop words and entity annotations, and then construct a deep learning model, wherein the deep learning model extracts the software requirement feature vector through the text encoding layer, the feature extraction layer, and the attention mechanism layer; construct a software requirement knowledge graph according to the association relationship between the software requirement feature vector and the entity, establish a mapping relationship between the function requirement node and the technology dependency node in the software requirement knowledge graph, and generate a requirement dependency matrix; The second unit is used to analyze the historical test environment data based on the requirement dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data, and use a graph neural network to feature encode the container image configuration information, the runtime dependency information, and the middleware version information to generate a test environment feature vector; input the software requirement feature vector and the test environment feature vector into an environment matching model constructed by a bidirectional long short-term memory network, calculate the cosine similarity of the vector space to obtain an environment matching score, and establish an association mapping model between requirement features and environment features; The third unit is used to extract a new requirement feature vector using the deep learning model when receiving a new software requirement, input the new requirement feature vector into the environment matching model, and calculate a matching score with the existing test environment based on the association mapping model; select the test environment template with the highest matching score for instantiation and deployment, and record the environment operation status data; feed back the environment operation status data to the environment matching model, and update the parameter weights of the association mapping model through online learning to achieve continuous optimization of the environment matching accuracy.

[0062] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0063] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0064] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligently matching software information on a mobile terminal, characterized in that: include: After word segmentation, removal of stop words and entity annotations on the software requirement text information, a deep learning model is constructed, wherein the deep learning model extracts the software requirement feature vector through the text encoding layer, feature extraction layer and attention mechanism layer; a software requirement knowledge graph is constructed according to the association relationship between the software requirement feature vector and the entity, a mapping relationship between the function requirement node and the technology dependency node is established in the software requirement knowledge graph, and a requirement dependency matrix is ​​generated; Analyze historical test environment data based on the requirement dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data, and use a graph neural network to feature encode the container image configuration information, the runtime dependency information, and the middleware version information to generate a test environment feature vector; input the software requirement feature vector and the test environment feature vector into an environment matching model constructed by a bidirectional long short-term memory network, calculate the cosine similarity in vector space to obtain an environment matching score, and establish an association mapping model between requirement features and environment features; When receiving a new software requirement, extracting a new requirement feature vector using the deep learning model, inputting the new requirement feature vector into the environment matching model, and calculating a matching score with the existing test environment based on the association mapping model; The test environment template with the highest matching score is selected for instantiation and deployment, and the environment operation status data is recorded; the environment operation status data is fed back to the environment matching model, and the parameter weights of the association mapping model are updated through online learning to achieve continuous optimization of the environment matching accuracy.

2. The method according to claim 1, characterized in that After word segmentation, removal of stop words and entity annotations on the software requirement text information, a deep learning model is constructed. The deep learning model extracts the software requirement feature vector through the text encoding layer, feature extraction layer and attention mechanism layer, including: Acquire software requirement text information, use a word segmentation processing module to segment the software requirement text information, use a stop word filtering module to remove stop words in the software requirement text information, use an entity annotation module to identify professional terms and technical terms in the software requirement text information, and generate a preprocessed requirement text; Construct a deep learning model based on a hierarchical architecture, wherein the deep learning model includes a text encoding layer, a feature extraction layer, and an attention mechanism layer. The text encoding layer adopts a BERT pre-trained model, the feature extraction layer adopts a multi-head self-attention network, and the attention mechanism layer adopts a demand-oriented attention allocation strategy; Input the preprocessed demand text into the text encoding layer of the deep learning model, the text encoding layer semantically encodes the demand text based on pre-trained language knowledge, and outputs text semantic features; input the text semantic features into the feature extraction layer, the feature extraction layer generates a first attention weight by calculating the product of the query matrix, the key matrix and the value matrix, and extracts a multi-level text representation; The multi-level text representation is input into the attention mechanism layer, and the attention mechanism layer constructs an attention allocation matrix according to the demand domain knowledge, and allocates differentiated weights to different types of demand information; the multi-level text representation is weightedly fused based on the attention allocation matrix to generate a software requirement feature vector.

3. The method according to claim 1, characterized in that Constructing a software requirement knowledge graph based on the software requirement feature vector and the association relationship between entities, establishing a mapping relationship between functional requirement nodes and technical dependency nodes in the software requirement knowledge graph, and generating a requirement dependency matrix includes: Identify entities in the software requirement feature vector, extract associations between the entities, and construct a software requirement knowledge graph based on the associations between the software requirement feature vector and the entities, wherein the software requirement knowledge graph includes a set of functional requirement nodes, a set of technical dependency nodes, and a set of relationship edges between nodes, wherein the set of relationship edges includes dependency relationships, inheritance relationships, combination relationships, call relationships, and implementation relationships between entities; Establishing a mapping relationship between the function requirement node set and the technology dependency node set in the software requirement knowledge graph, inputting each function requirement node in the function requirement node set into a graph attention network, the graph attention network assigning a second attention weight to a neighboring node of each function requirement node, performing feature aggregation on the function requirement node based on the second attention weight, and generating a structured feature of the function requirement node; The structured features of the functional requirement nodes are spliced ​​with the corresponding technology dependency node features, and the spliced ​​features are input into a mapping strength calculation module, which outputs the association strength value between the functional requirement nodes and the technology dependency nodes; the association strength value is organized into a requirement dependency relationship matrix.

4. The method according to claim 1, characterized in that: Analyzing historical test environment data based on the demand dependency matrix, extracting container image configuration information, runtime dependency information, and middleware version information from the historical test environment data, and using a graph neural network to feature encode the container image configuration information, the runtime dependency information, and the middleware version information to generate a test environment feature vector, including: Analyze historical test environment data based on the demand dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data; construct an environment information heterogeneous graph, wherein the nodes of the environment information heterogeneous graph include configuration instruction nodes in the container image configuration information, dependency package nodes in the runtime dependency information, and middleware nodes in the middleware version information, and the edges of the environment information heterogeneous graph represent associations between nodes; The environmental information heterogeneous graph is feature encoded by using a graph neural network, the graph neural network includes a feature extraction module, a type-aware attention module, and a feature aggregation module; the feature extraction module extracts features of the basic image information and the environmental configuration information of the configuration instruction node, the version information and constraint information of the dependency package node, and the version information and configuration information of the middleware node to generate heterogeneous node features; The heterogeneous node features are input into the type-aware attention module. The type-aware attention module assigns third attention weights to different types of nodes according to the node types, and performs weighted aggregation on the heterogeneous node features based on the third attention weights. The feature aggregation module performs intra-component aggregation and inter-component aggregation on the node features after weighted aggregation to generate a test environment feature vector.

5. The method according to claim 1, characterized in that Inputting the software requirement feature vector and the test environment feature vector into the environment matching model constructed by the bidirectional long short-term memory network, calculating the cosine similarity in the vector space to obtain the environment matching score, and establishing the association mapping model between the requirement feature and the environment feature includes: Constructing a bidirectional long short-term memory network as an environment matching model, the bidirectional long short-term memory network includes an input layer, a bidirectional long short-term memory layer, a feature fusion layer, and a similarity calculation layer, the bidirectional long short-term memory layer includes a forward long short-term memory unit and a backward long short-term memory unit, the forward long short-term memory unit and the backward long short-term memory unit respectively include an input gate, a forget gate, and an output gate; The software requirement feature vector and the test environment feature vector are input into the input layer of the environment matching model, and the forward long short-term memory unit and the backward long short-term memory unit process the software requirement feature vector and the test environment feature vector in the forward and reverse directions respectively to obtain bidirectional feature information; the feature fusion layer fuses the hidden state vector output by the forward long short-term memory unit and the backward long short-term memory unit to generate a fused feature vector; A vector space cosine similarity calculation module is constructed, and the vector space cosine similarity calculation module includes a feature weight allocation unit and a similarity calculation unit; the feature weight allocation unit allocates a weight coefficient to the fused feature vector based on the feature importance to generate a weighted feature vector; the similarity calculation unit calculates the cosine similarity of the weighted feature vector in the vector space to obtain an environment matching score, and constructs an association mapping model between demand features and environment features based on the environment matching score.

6. The method according to claim 1, characterized in that The test environment template with the highest matching score is selected for instantiation and deployment, and the environment operation status data is recorded; the environment operation status data is fed back to the environment matching model, and the parameter weights of the association mapping model are updated through online learning to achieve continuous optimization of the environment matching accuracy, including: Selecting the test environment template with the highest environment matching score, building a deployment plan based on the test environment template, the deployment plan including resource allocation strategy, component installation sequence, and configuration parameter settings; instantiating the deployment test environment according to the deployment plan to generate an operating environment instance; Collecting the running status data of the running environment instance, the running status data includes infrastructure layer monitoring data, system layer monitoring data, application layer monitoring data, and business layer monitoring data; performing time alignment and data standardization processing on the running status data to generate environment running characteristics; The environmental operation characteristics are input into an environmental status assessment module, and the environmental status assessment module calculates environmental operation stability indicators, performance indicators, and resource utilization indicators through multi-dimensional time series analysis to generate an environmental health score; Constructing an online learning optimization unit, wherein the online learning optimization unit uses the difference between the environmental health score and the original environmental matching score as a loss function, and calculates gradient information based on the loss function; dynamically adjusting the parameter weights of the environmental matching model using an adaptive learning rate according to the gradient information; The updated environment matching model is used for a new test environment matching score to calculate the environment matching accuracy; when the environment matching accuracy is higher than a preset threshold, the parameter weight is recorded as the optimal weight; when the environment matching accuracy is lower than the preset threshold, the parameter optimization module is triggered to perform weight fine-tuning until the environment matching accuracy meets the optimization target.

7. A mobile terminal software information intelligent matching system, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to perform word segmentation processing on the software requirement text information, remove stop words and entity annotations, and then construct a deep learning model, wherein the deep learning model extracts the software requirement feature vector through the text encoding layer, the feature extraction layer, and the attention mechanism layer; construct a software requirement knowledge graph according to the association relationship between the software requirement feature vector and the entity, establish a mapping relationship between the function requirement node and the technology dependency node in the software requirement knowledge graph, and generate a requirement dependency matrix; The second unit is used to analyze the historical test environment data based on the requirement dependency matrix, extract container image configuration information, runtime dependency information, and middleware version information from the historical test environment data, and use a graph neural network to feature encode the container image configuration information, the runtime dependency information, and the middleware version information to generate a test environment feature vector; input the software requirement feature vector and the test environment feature vector into an environment matching model constructed by a bidirectional long short-term memory network, calculate the cosine similarity of the vector space to obtain an environment matching score, and establish an association mapping model between requirement features and environment features; A third unit is used for, when receiving a new software requirement, extracting a new requirement feature vector using the deep learning model, inputting the new requirement feature vector into the environment matching model, and calculating a matching score with the existing test environment based on the association mapping model; The test environment template with the highest matching score is selected for instantiation and deployment, and the environment operation status data is recorded; the environment operation status data is fed back to the environment matching model, and the parameter weights of the association mapping model are updated through online learning to achieve continuous optimization of the environment matching accuracy.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images