940 results about "Granularity" patented technology

The invention provides a multi-feature fusion rumor detection method, system and device based on knowledge distillation, and mainly solves the problems that an existing model is high in calculation overhead, insufficient in feature fusion and insufficient in emotion utilization. The method comprises the steps of firstly obtaining multi-dimensional data such as social media original texts and comments; extracting deep semantic representation by using a pre-training model, and analyzing comment emotion features in combination with a hybrid neural network; then, features such as semantics, emotions, emoticons and populations are input into a hierarchical gating interactive fusion network (GIFN), and weights are dynamically adjusted to achieve effective fusion of multi-granularity features; in order to reduce complexity, a knowledge distillation framework is designed: a deep GIFN is used as a teacher network to generate a soft label, and a lightweight student network (LSTM) is guided to perform training. According to the trained student model, the parameter quantity is remarkably reduced, meanwhile, good detection performance is kept, the student model can be conveniently deployed in an actual content auditing system or edge equipment, and social content rumors can be efficiently recognized and judged.

The invention discloses a scene granularity interface test case generation method based on multiple agents, and the method comprises a plurality of large language model agents and related components, namely context memory, an observer, a decision maker, an executor, a supervisor, a recorder and a to-be-tested application. An observer identifies the interface control, detection and attribute identification of the control are effectively focused, and it is ensured that accurate data input is provided for subsequent testing steps. Based on the output of an observer, a decision maker generates a detailed plan of the next operation according to the current page state, and an operation strategy can be independently formulated, so that test generation is supported by business scene logic. And an executor analyzes and executes the generated plan operation, carries out test operation strictly according to the plan, and ensures the traceability and repeatability of the test process. Furthermore, the supervisor verifies whether the operation is executed according to the plan and meets the requirements of the target scene, and makes corresponding feedback according to the check result, thereby ensuring that the wrong operation is corrected in time, avoiding the continuous accumulation of errors, and ensuring the effectiveness of each test step. The recorder is responsible for storing the execution instruction and the context information of the test process to the context memory, and the operation history and result in the test process can be traced through the test context information, so that a better decision can be made in the subsequent iteration, and the result export after the test is completed can be effectively supported. And after the recording is finished, the test generation enters the next iteration until the coverage of the test scene is finished.

The invention discloses an auto-filling method and device of a northbound FTP (File Transfer Protocol) performance file for a telecommunication network management system. The method comprises the following steps of: setting the granularity state of granularity of performance data according to the reported time of the performance data from a foreground network element by a performance loading module of the telecommunication network management system; when the performance loading module receives information reported subsequently by the foreground network element, wherein the information contains the performance data, if judging that the granularity of the performance data is in a completed state according to the set granularity state, then sending filling information containing the performance data to a northbound FTP performance module; and carrying out the processing that the performance data is filled to a previously generated northbound FTP performance file with the granularity by the northbound FTP performance module according to the filling information. Through the auto-filling method, the information reported by the foreground network element is analyzed to judge whether the northbound FTP performance file is complete or not, and the auto-filling of the performance data is carried out when the judgment result is that the northbound FTP performance file is incomplete.

The invention discloses an information resource efficient collaborative optimization method for a distributed computer system, which belongs to the technical field of information resource optimization control and comprises the following steps: executing a transaction according to a target lock element information set, monitoring popularity information of each resource segment, generating self-adaptive lock granularity decision information of each resource segment, and executing dynamic adjustment of lock granularity. And obtaining load sensing optimization parameters to complete lock competition optimization, and completing real-time iterative updating of lock element information and strategies. According to the method, the problems of resource competition and deadlock in the distributed system can be effectively solved, and the system performance and the resource utilization efficiency are improved. Transaction pre-allocation resources are reasonably allocated, the deadlock risk is reduced, lock granularity is dynamically adjusted according to resource segmentation popularity, lock competition optimization is completed based on load awareness optimization parameters, an adjustment result and the optimization parameters are synchronized to a distributed lock management library for real-time iteration updating, and it is ensured that a system always operates with an optimal strategy; and the distributed computer system efficiently and stably processes information resources in a cooperative manner.

The invention relates to the technical field of natural language processing, knowledge engineering and artificial intelligence, and particularly discloses a method and system for enhancing understanding of professional domain knowledge by a large model. The method comprises the steps that a professional domain entity classification system composed of a core entity, an auxiliary entity and a relation entity is constructed, attributes are expressed in a layered labeling and multi-granularity modeling mode, and semantic vectors are generated through ontology modeling and an embedding algorithm; based on a mixed extraction framework fusing expert rules and a neural network model, high-quality extraction of professional domain knowledge is realized; the method comprises the following steps: integrating multi-source heterogeneous data, and constructing a dynamically updated domain knowledge graph through semantic mapping, entity normalization and metadata weighting strategies; a knowledge graph is embedded into a Transform architecture, a knowledge perception attention mechanism and a multi-hop inference engine driven by reinforcement learning are introduced, and the knowledge fusion and inference ability of a large model is improved; and meanwhile, a triple check mechanism is designed to ensure entity consistency, relation logicality and numerical reasonability of the generated content. According to the method, the knowledge understanding and reasoning capability of a large model in professional scenes such as water conservancy is effectively improved, and the method has good universality and engineering application prospects.

The invention discloses a GPU computing power resource scheduling method and device based on load awareness and a medium, and relates to the technical field of computing power scheduling. The method comprises the steps that according to the video memory capacity and the number of calculation cores of the GPU, the video memory capacity is divided into a plurality of continuous fragments, the calculation cores are divided into a plurality of logic calculation groups, and sub-resource units are obtained; recording occupation states and load indexes of the sub-resource units in real time to obtain a resource pool; analyzing a job submitted by a user, and collecting an execution period of a kernel function, a video memory access mode and an instruction pipeline blocking rate when the job runs; and on the basis of a reinforcement learning algorithm, predicting a resource demand inflection point of the job according to the load fingerprint model, performing hierarchical scheduling on the sub-resource units based on virtualization isolation and multi-dimensional resource quantitative evaluation, and generating a preemptive allocation strategy of the sub-resource units. According to the method, efficient utilization and secure sharing of GPU resources are realized through collaborative design of dynamic granularity segmentation, load aware scheduling and hardware-level security isolation.

The invention discloses a noise text open intention classification method and system based on a large model, and belongs to the technical field of natural language processing and artificial intelligence, and the method mainly comprises the steps: extracting text features through a pre-trained large language model, carrying out the clustering of the text features through unsupervised granular ball clustering, and constructing intra-class structure representation in a feature space; dividing the training sample into a clean sample, an internal distribution noise sample, an external distribution noise sample and an uncertain sample based on a global particle-ball-level OOD tendency index and a sample-level consistency index; constructing a joint loss function guided by beneficial noise, and performing differentiation training on the model by using a sample division result; and constructing a multi-granularity decision boundary based on a granular ball structure, wherein the multi-granularity decision boundary is used for judging the category of the input sample. According to the method, multi-granularity structure information, a noise pattern recognition mechanism and a joint representation learning strategy are fused, and the recognition robustness and generalization ability of the model in a complex open environment are effectively improved.

Action and / or process determination and recommendations for Robotic Process Automation (RPA) using semantic action graphs is disclosed. Semantic action graphs are graphs that store individual actions, and potentially graphical elements and / or text associated with the actions, as nodes, as well as the relationships between nodes as edges. Metadata to develop the semantic action graphs may be derived from task mining applications that can monitor the interactions of users with computing systems, workforce intelligence, etc. The semantic action graphs may be for a user, an organization, an industry, product-wide, etc. At their lowest level of granularity, the recommendations may be for mouse clicks, key presses, Application Programming Interface (API) calls, system events, etc. At higher levels of granularity, the recommendations may be for opening an order, creating a lead, approving a work item, etc.

Systems and methods for training and using large language models to respond to queries and cascading to new models when their performance falls below a threshold during the response generation cycle are described. The methods involve receiving and analyzing a query using heuristics to determine the query's categories and a level of granularity at which its response is to be evaluated. Selecting a language model based on the query analysis and during the generation of the response, evaluating the model's performance for its ability to predict the next segment in a response. The methods score the evaluation and use the score to determine whether the language model's performance exceeds a confidence threshold. If it does not, then during the response is being generated, e.g., at a point before the response generation is completed, cascading from the model to a different model that can perform at a higher confidence level.

The invention provides a power big data-based sparsity feature adaptive compression method, which comprises the steps of adaptively optimizing a sensing matrix through a dictionary learning method aiming at a screened multi-granularity feature subset, introducing a sparse regularization item to enhance the sparsity of data reconstruction, and according to a missing mode of a missing value and context information, carrying out adaptive compression on the sparse feature of the power big data. Establishing a mapping relationship among the compression ratio, the reconstruction error and the feature subset, and determining a self-adaptive compression strategy; based on a self-adaptive compression strategy, a compressed sensing method is adopted to carry out self-adaptive compression on the multi-time-granularity power data, the compression ratio is dynamically adjusted according to reconstruction error feedback, the data compression and reconstruction quality is balanced, and the reconstructed multi-granularity power data is synchronized according to timestamp precision.

Computer-implemented systems and methods implement semantic search in high-dimensional vector spaces, specifically tailored for use with large language models (LLMs). In particular, clustering is combined with Euclidean distance measurements to facilitate real-time vector searches. By implementing clustering, the invention reduces the computational complexity and costs associated with Euclidean distance calculations, which are typically more resource-intensive than other methods such as cosine similarity. This reduction is achieved by limiting the scope of distance calculations to within clusters, thereby avoiding the inefficiencies and diminished accuracy otherwise encountered by existing systems when using Euclidean distance in high-dimensional spaces. As a result, the invention retains the benefits of Euclidean distance, such as its superior granularity and precision in measuring semantic relevance, without succumbing to the usual drawbacks of high computational demands and poor scalability.

The invention discloses a Token level cache matching method and system of a large language model and a storage medium. The method comprises the following steps: constructing a local context fragment; generating a context embedding vector of the current Token, and calculating a context entropy value, a semantic consistency index and a semantic change gradient; for the target Token, determining a semantic category of the target Token; inputting the semantic category to which the target Token belongs into a Hash decision maker to obtain a Hash granularity level; if the Hash is the first-level Hash, executing fixed-length Hash and mapping the Hash to a semantic cache bucket based on a semantic theme or a context range of the first-level Hash; if the first-level hash is the second-level hash, dynamically adjusting the hash length according to the semantic similarity with the adjacent Token; if the Hash granularity level of the target Token is a third-level Hash, constructing a high-dimension context representation and executing a fine Hash operation; and matching the hash result with the KV in the pre-stored cache, and executing corresponding operation according to the matching result.

The invention discloses a reasoning optimization method and device for a code generation large model, equipment and a medium, and relates to the technical field of model reasoning, and the method comprises the steps: in the reasoning process of a code generation task of a target code generation large model, executing a multi-granularity uncertainty quantification step in parallel every time a new Token is generated, obtaining a multi-granularity uncertainty score; constructing a state space vector, and utilizing a preset reinforcement learning strategy network to evaluate the selection probability of a plurality of preset reasoning optimization strategies based on a preset smooth decision mechanism so as to determine a target reasoning optimization strategy; if the strategy is a preset reasoning acceleration strategy, optimization processing is carried out through speculation decoding; if the strategy is a preset exploration optimization strategy, performing optimization processing by using a preset multi-path sampling technology and a preset knowledge enhancement technology; if the strategy is the preset fuzzy processing strategy, taking the plurality of candidate outputs as target reasoning outputs for optimization processing; and evaluating the decision effect according to the reasoning result to optimize the preset reinforcement learning strategy network.

The invention discloses a processor, a chip product, computer equipment and a tensor calculation method, and relates to the technical field of processors. A first tensor is stored in a first storage unit of the processor; a plurality of scaling factors corresponding to the first tensor are stored in a second storage unit of the processor, the scaling factor set comprises a plurality of scaling factors, each scaling factor corresponds to one sub-block in the first tensor, each sub-block comprises at least one element in the first tensor, and each sub-block comprises at least one element in the first tensor; the scaling factor is used for adjusting the dynamic range of each element in the same sub-block; a tensor calculation engine of the processor is used for reading a first tensor from the first storage unit and reading a scaling factor set from the second storage unit; and for any sub-block in the first tensor, executing logical operation on elements in the sub-block according to the scale factor corresponding to the sub-block to obtain an operation result. According to the method and the device, the tensor calculation engine can realize accurate operation of the low-precision tensor through the scale factor of the sub-block granularity.

The invention relates to the technical field of engineering process control, and discloses a ball mill granularity soft measurement method based on a time sequence large model. Constructing a multi-feature fusion module for extracting multi-scale features based on Convld K3, Convld K5 and Convld K1, and constructing a soft measurement model in combination with multi-head attention and a large language module with a fixed weight; the soft measurement model generates a first feature and a query matrix, generates a key matrix and a value matrix according to a fixed weight of the large language module, generates a second feature based on the query matrix, the key matrix and the value matrix, and fuses the second feature with the first feature to obtain a fused feature; and then a granularity prediction result corresponding to the field data is obtained through a large language module, so that the problems that an existing soft measurement model is too high in dependence on large-scale sample data, insufficient in modeling capability for complex nonlinear process parameters and low in prediction precision are solved.

The invention discloses an equipment operation and maintenance decision-making method based on large-scale knowledge graph multi-granularity learning reasoning, which comprises the following steps of: firstly, extracting a multi-granularity sub-graph of a subject entity in an equipment operation and maintenance knowledge graph, and integrating multilevel information such as an equipment state, a fault mode and an environmental factor to form a relation path set; secondly, performing joint coding on a question text and a relation path through an attention enhanced word embedding module based on a graph convolutional network, forming a word embedding vector, calculating a similarity score between a relation in the path and the question, pruning a sub-graph, and reducing the calculation amount; and finally, fusing a plurality of particle size sub-graphs after pruning processing, inputting the fused particle size sub-graphs into a reasoning decision module, extracting feature information from graph structure data by the reasoning decision module through graph convolution operation, dynamically adjusting a GCN learning process in combination with LSTM, finally realizing question reasoning and answer generation, and ensuring the accuracy of an equipment operation and maintenance decision.

The invention provides a method and a device for constructing a three-dimensional random geologic model based on machine learning. The construction method comprises the following steps: constructing a geologic model area range and discretizing the geologic model area range into n voxels; constructing an initial geologic model based on the label value of the known voxel determined by the drilling data and the initial label value of the unknown voxel determined by the cubic learning domain; constructing a neighborhood system of each voxel, and setting spatial constraints in 13 directions; based on the label value of each voxel in the initial geologic model, obtaining a prediction result according to a label probability prediction formula and an updating iteration formula of a granularity coefficient; based on the prediction result, obtaining the marginal probability of each unknown voxel and the prediction label value of each unknown voxel; and updating the initial geologic model based on the predicted label value of each unknown voxel, and constructing a three-dimensional random geologic model. The three-dimensional random geologic model constructed by adopting the construction method disclosed by the invention is accurate in simulation result, and meanwhile, the uncertainty of the stratum can be represented.

The invention belongs to the technical field of time sequence data streams, and particularly relates to a multi-granularity anomaly detection system for a time sequence data stream, which comprises a data acquisition and preprocessing module used for firstly performing multi-source data acquisition, then cleaning and standardizing the data and then generating data with different time granularities; the multi-granularity feature extraction module is used for firstly performing time sequence feature extraction, then performing window feature extraction and then performing multi-granularity feature fusion; and the anomaly detection model module is used for detecting whether the data fused by the multi-granularity feature extraction module is abnormal or not, and then improving the accuracy and stability of detection according to an integrated learning model. Compared with a traditional detection method, the method has the advantages that by arranging the anomaly detection model module, the detection accuracy and timeliness of complex mode anomalies can be greatly improved, and economic losses caused by the fact that enterprises do not find anomalies in time are avoided.

The invention relates to a question and answer method and system fusing a multi-granularity knowledge graph and a self-adaptive agent. The question and answer method comprises the steps of knowledge governance and extraction, knowledge generation and expansion, knowledge retrieval and management, and knowledge rule mining and self-adaptive retrieval. Through cooperation of the multi-granularity knowledge graph and the self-adaptive agent, questions of the user can be more comprehensively understood, and more relevant knowledge can be retrieved from the knowledge base, so that the question and answer accuracy is remarkably improved, and wrong answers are reduced. Meanwhile, the multi-granularity knowledge graph can support more complex knowledge application requirements such as reasoning and prediction, so that the application capability of the question and answer system is enhanced, and the question and answer system can process more complex question and answer tasks such as multi-round dialogues and multi-task question and answer.

The invention discloses a multi-modal open intention recognition method and system based on pellet characterization, and belongs to the technical field of artificial intelligence and multi-modal intention understanding, and the method comprises the steps: carrying out the feature extraction and modal fusion of multi-modal input data; carrying out structural modeling on the feature representation of each mode and the fusion mode through an adaptive particle and ball clustering method, and generating a multi-granularity particle and ball set; the mass centers of the pellets serve as multi-granularity anchor points, and the pellets with the same labels in different modalities are aligned; introducing a weighting mechanism based on purity and sample scale into the fusion mode; generating a boundary-constrained pseudo-distribution outer sample in the fusion modal space; and constructing a self-adaptive decision boundary based on the fusion modal particle ball obtained by training, and carrying out known class classification and unknown class detection. According to the invention, by introducing the multi-granularity anchor point and the structure perception particle-ball representation mode, the joint recognition of the known category and the unknown category in the multi-modal scene is realized, and the accuracy and robustness of intention recognition are remarkably improved.

The invention relates to safety compliance evaluation, in particular to a safety compliance evaluation system and method based on a multi-modal large model, and a functional module cluster, which are used for managing each functional module in the system, supporting cross-modal consistency detection and comparing answer consistency under different modals. The self-developed security system is responsible for execution of an assessment task, including resistance test enhancement and built-in multi-granularity data variation strategy, improves assessment robustness by simulating a real attack scene, realizes end-to-end automation from data generation, tested model test to risk assessment, supports configurable parameters and reproducible results, and has the advantages of high reliability and high reliability. The self-iteration and version management of the evaluation model are realized, and the self-iteration mechanism is deeply combined with a compliance label system, so that the risk identification capability under legal, ethical and industrial specifications can be continuously optimized; according to the technical scheme provided by the invention, the limitation that evaluation can be carried out only under a single mode and a single rule in the prior art can be effectively overcome.

Certain aspects provide techniques and apparatus for executing queries in a computing system using machine learning models. An example method generally includes receiving a plan to satisfy a request in the computing system and event log data associated with execution of the plan. The plan generally specifies a first plurality of function calls at a first level of granularity. Using a plan refinement machine learning model, a refined plan is generated when the event log data indicates that execution of the generated plan results in one or more execution errors and the one or more execution errors are solvable. Generally, the refined plan specifies a second plurality of function calls at a second level of granularity, the second level of granularity being finer than the first level of granularity.

The invention provides a multi-granularity open vocabulary query method based on an object-level lossless Gaussian field, and the method comprises the steps: introducing an object-level Gaussian field with a global consistency codebook, rendering a learnable semantic tag vector in the Gaussian field back to a corresponding object tag; the direct mapping between the label and the corresponding uncompressed high-dimensional feature is established through the code book, so that the semantic feature of any dimension is supported, additional compression is not needed, and the understanding capability on an object is remarkably improved; according to the method, wide quantitative and qualitative evaluation is carried out in a plurality of scenes, excellent performance in the aspects of object level zero sample segmentation and open vocabulary understanding is shown, the highest precision is particularly achieved in object-component hierarchical retrieval, and meanwhile multi-granularity scene editing is supported.

The invention discloses a human-intelligent collaborative domain increment target detection method. According to the invention, an increment strategy and an active learning sampling method are arranged in a flexible increment training loop; reuse and combination of knowledge are realized by using a sparse hybrid expert network, the generalization ability of model increment training is improved, data playback is performed on an old domain sample based on distributed sampling, and knowledge forgetting in the model increment training is relieved; a part of new domain samples with representativeness and diversity are screened through clustering and maximum mean value difference calculation, on one hand, the data size difference of new and old samples is reduced, and on the other hand, the stability and plasticity of the model are balanced by controlling the number proportion of the new and old samples and the round fine granularity of iterative training. According to the method, the adaptability of the team in complex and diversified environments is improved, fine-grained balance of model stability and plasticity is supported, and the method can be applied to a human-intelligent cooperation domain increment target detection scene and has certain practical significance and theoretical value.

The present invention generally relates to the field of robotics, and, more particularly, to a method and system for task anticipation by integrating large language models and classical planning. Conventional methods for task anticipating use data-driven deep network architectures and Large Language Models (LLMs) for task estimation but they do so at the level of high-level tasks and require a large number of training examples. Thus, embodiments of present disclosure provide a method and system for task anticipation by integrating large language models and classical planning. The disclosed method and system leverages the generic knowledge of LLMs through a small number of prompts to perform high-level task anticipation, using the anticipated tasks as joint goals in a classical planning system to compute a sequence of finer granularity actions that jointly achieve these goals.

A system and method for hierarchical PCM-controlled traversal across nested latent hyperspaces. Input data, including video, is encoded into coupled various granularity subspaces. A goal-conditioned controller computes geodesic routes within levels and defines cross-level lifts and projections to maintain semantic continuity. Symbolic anchors provide durable reentry and audit, while strategy caching abstracts recurrent decision motifs for reuse. A kernel-adaptation subsystem derives motion / recurrence / frequency / semantic features to reshape local metrics and traversal costs, enabling level-aware, reversible updates. During execution the system dynamically switches levels, records checkpoints for backtracking, and commits salient results to persistent memory. For video embodiments, a Lorentzian structure preserves temporal causality and supports continuous zoom, multiview alignment, and cross-temporal analysis. The architecture transforms navigation from frame- or token-based stepping to structured, goal-aligned movement through shaped latent space, improving efficiency, fidelity, and explainability across tasks.

The invention relates to the technical field of information, and discloses a multi-machine collaborative operation control method and system based on a common inductance calculation control technology, and the method comprises the steps: obtaining task granularity distribution and a resource demand matrix, and carrying out the classification of the task granularity distribution and the resource demand matrix through a classification algorithm, and obtaining an initial task segmentation feature set; based on the optimal segmentation point set, a preliminary task allocation scheme is generated in combination with node calculation heterogeneity, and a task segmentation feature set is dynamically updated through a real-time sensing technology; a real-time task state set is generated through a group information sharing mechanism, a task allocation scheme is optimized through a dynamic programming algorithm, and it is ensured that tasks are reasonably allocated according to priorities and node resources; and the cooperation state matrix is updated through a real-time communication protocol, so that the accuracy and timeliness of a multi-node cooperation execution result are ensured. According to the invention, task allocation can be dynamically adjusted in a complex environment, resource utilization is optimized, the cooperation efficiency of a multi-robot system is improved, and the method can be widely applied to the fields of industrial production, warehouse logistics, disaster rescue and the like.

The invention discloses an abnormal data monitoring method and device based on artificial intelligence, and the method comprises the steps: 1, dividing an original data stream through a sliding window, extracting statistics, time sequence and change rate features, and dynamically screening features adaptive to data distribution based on an SHAP value; 2, constructing a double-flow model, capturing a global isolated mode by adopting an improved isolated forest in a static flow, capturing time sequence dependence on the basis of LSTM-AE in a dynamic flow, and fusing two-flow scores through performance-driven dynamic weight distribution; 3, combining a density peak value algorithm with historical density attenuation weighting, and dynamically adjusting an abnormal threshold value; 4, realizing low-delay incremental learning through a double-trigger mechanism and experience playback; 5, multi-granularity interpretation is generated, manual annotation feedback is supported, feature engineering and model training are integrated, and a'detection-interpretation-feedback-optimization 'closed loop is formed; high-adaptability anomaly monitoring is realized through dynamic feature screening, double-flow fusion detection, threshold value self-adaption and man-machine collaborative optimization.

The invention discloses a single-stage generative sequence recommendation method, which belongs to the field of personalized recommendation systems, and comprises the following steps: S1, item code representation: performing formalized representation on items according to attribute granularity, and generating code representation of the items through a large language model and a product quantization technology; s2, constructing a user interaction sequence: splitting the user interaction sequence into a plurality of attributes to form an interaction sequence consisting of project attributes; s3, generative sequence recommendation: improving an encoder architecture of Transform, including a sequence projection layer, a spatial aggregation layer and a conversion layer, so as to adapt to a sequence recommendation task; s4, recommendation item prediction: based on the user interaction sequence, learning probability distribution, and generating prediction of a next interaction item; according to the method, prediction of the next item possibly interacted by the user can be directly generated, similarity query does not need to be carried out in the candidate set, and the sequence recommendation effect is effectively improved.

The invention relates to the technical field of software security, in particular to a vulnerability mining system and device based on source code similarity. The vulnerability mining system comprises a vulnerability feature library which comprises vulnerability code snippets and feature information associated with the vulnerability code snippets; the coarse granularity positioning module is used for matching the vulnerability code snippets in the vulnerability feature library with the to-be-analyzed source code aiming at the to-be-analyzed source code, determining suspicious code snippets and forming similar code pairs; the feature representation module is used for acquiring feature information of the suspicious code snippets; the fine granularity positioning module is used for obtaining the semantic similarity, the structural similarity and the subgraph matching degree of the similar code pair according to the feature information associated with the two code snippets in the similar code pair, and determining the comprehensive similarity of the similar code pair; and the judgment module is used for determining whether the suspicious code snippets in the similar code pairs have vulnerabilities or not according to the comprehensive similarity. The method has the beneficial effects that the processing efficiency can be improved while the detection precision can be ensured.