160 results about "Dependency tree" patented technology

The invention belongs to the technical field of natural language processing. According to the zero resource framework-based fine-grained large model illusion processing method and system, multi-dimensional anomaly detection is carried out on a preprocessed text through a large language model, and a multi-dimensional anomaly detection result is obtained; performing anomaly correction priority ranking according to a multi-dimensional anomaly detection result; dynamically adjusting a context window range according to a syntactic analysis result and the generated syntactic dependency tree; performing priority correction on high-risk vocabularies and sentences according to an abnormal correction priority ranking result in combination with the adjusted context window range; and evaluating the corrected content, and obtaining a corrected final text when a termination condition is met. According to the method, the excessive confidence, subtle semantics and grammar deviation illusion detection of the detection content is realized, and the illusion detection processing capability of a large language model, especially the detection processing capability of a fine-grained illusion problem, is improved.

The invention discloses a multi-modal classification method fusing a graph convolutional neural network and a capsule graph neural network, and solves the problem of how to apply an artificial intelligence technology to an appearance patent. The method comprises the following steps: coding a patent text sequence by utilizing BERT, segmenting semantic graph convolution by utilizing PCNN, constructing a dependency tree by utilizing grammar graph convolution, carrying out interactive attention and relationship classification, and finally carrying out image feature extraction through a capsule network. According to the method, the appearance patent classification technology fusing heterogeneous multi-modal information is researched, the feature extraction and fusion problems of multi-modal information such as characters, images and networks are solved, a heterogeneous multi-modal appearance design patent classification model is constructed, and the classification accuracy is greatly improved relative to pure text content. And meanwhile, an appearance design patent AI standard data set is constructed on the basis of the intellectual property hypergraph network, so that the blank of patent types in a traditional Lankyo classification system is filled, and the bottleneck of a traditional appearance patent classification method is broken through.

The invention discloses a cross-modal alignment method and device for a multi-modal large language model, and relates to the technical field of multi-modal alignment, and the method comprises the steps: obtaining an image, carrying out the target detection, and generating a local image fine-grained feature; obtaining a text, extracting text features, extracting a text entity relationship, and obtaining an entity relationship dependency tree as a text entity relationship graph; performing local fine-grained semantic alignment on the local image fine-grained features and the text features; using the scene graph model to obtain a visual relation graph between detection targets in the image; performing graph structure alignment and graph semantic alignment on the text entity relation graph and the visual relation graph; and fusing the semantic alignment features and the graph alignment features to obtain semantic alignment features with rich fine grit, global semantic alignment features and richer relationship alignment features, and inputting the semantic alignment features, the global semantic alignment features and the richer relationship alignment features into a multi-modal large model. According to the method, alignment of different modal data on different feature dimensions is realized, so that the different modal data are accurately associated.

The invention belongs to the technical field of software engineering, and particularly relates to a project dependency library vulnerability step-by-step utilization method based on LLM enhancement. The method comprises the following steps: carrying out lightweight dependency tree and call graph reachability analysis on a project, and identifying all reachability call chains from the project to a vulnerability method; decomposing the fuzz test on the whole call chain into a directional fuzz test on each step of method call in the call chain; in combination with static analysis and prompt engineering technologies, seed generation is assisted by semantic understanding of LLM on a current test function and utilization information generated by a previous test. And performing directional fuzzy testing by utilizing the generated seeds, analyzing execution paths of the seeds through an execution feedback mechanism, guiding selection and variation of the seeds, and finally generating effective input capable of successfully utilizing the project dependent library vulnerabilities. According to the method, developers can be helped to quickly position and timely repair vulnerabilities, so that risks and potential loss of projects caused by dependency on the vulnerabilities are reduced.

The invention discloses a multi-layer graph multi-mode sentiment analysis method and device based on meta-information, and the method comprises the steps: firstly, extracting the high-level features of text, video and audio data through BERT, OpenFace 2.0 and Librosa, carrying out the time sequence coding through BiLSTM, and then mapping each mode feature to a unified space through MLP; then, respectively capturing modal invariant and specific features by using a shared multi-modal encoder and a private multi-modal encoder; for modal specific features, adopting GNN to construct a modal graph, and combining a grammar dependency tree and self-attention to calculate a graph edge weight so as to align meta-information in a modal; video and audio features are aligned with a text through Cross-Attention, a new multi-modal graph is constructed, and multi-level features are extracted; then, applying convolution, BiLSTM and GNN to extract local, time sequence and global features, and realizing multi-level feature fusion through a cross-modal attention mechanism; and after each modal feature graph and each modal invariant feature are input into a graph knowledge distillation module, dimension reduction processing is carried out, features are further fused through Transform and a multi-head attention mechanism, and finally an emotion prediction value is obtained.

The invention discloses a software test automation report generation optimization method which comprises the following steps: automatically registering and authenticating a multi-source test data access endpoint, uniformly collecting and normalizing test data with a multi-dimensional attribute tag, and forming a high-quality grouped data set through structured modeling and automatic clustering; further extracting event and result feature nodes, and establishing an interpretable tree rule dependency relationship; according to the method, the minimum redundancy and reusable customized rule path is generated by applying a dependency tree traversal and optimization algorithm and automatic reasoning, and the continuous adaptive evolution of the rule base among multiple projects is realized by combining conflict detection and visual interaction, so that the efficiency and expandability of data processing in a complex test scene are improved, and the test efficiency is improved. And the flexibility and automation level of report customization rule configuration are enhanced.

According to the software configuration method and device provided by the embodiment of the invention, the traceability and real-time synchronization of configuration change are realized by constructing the distributed configuration database and the message publishing and subscribing center. A machine learning classifier is innovatively introduced to intelligently classify configuration features, and a configuration dependency tree is established based on a configuration project template, so that a complex dependency relationship among key configuration projects is effectively managed. Accurate tracking of the configuration state is achieved through the configuration version linked list and the historical version library, and the reliability of configuration updating is ensured in combination with a syntactic analyzer and integrity verification. According to the method, the limitation of traditional static configuration management is broken through, and an intelligent and reliable configuration management solution is provided for a complex software system.

Disclosed herein are systems and methods for automatic detection, fixing, and monitoring of vulnerabilities. In one aspect, an exemplary method comprises receiving as an input an artifact, when the received artifact is vulnerable, building the artifact for a new release, scanning a library containing the new release of the artifact to determine whether the library is safe for publishing, and publishing the library to an accessible artifact depository when the library is safe for publishing. In one aspect, the building of the artifact for the new release includes, scanning a dependency tree, replacing the vulnerable version by a safe version of the artifact when a safe version is available, fixing the vulnerable version of the artifact using known patches when a safe version of the artifact is not available, adding a postfix to a group ID of the artifact for distinguishing the fixed version from prior versions of the artifact.

The invention relates to a large law model fine tuning data screening method and device and a medium, and belongs to the technical field of artificial intelligence. Generating and forming a preference data set, training a legal quality scoring model by using the preference data set, and dynamically adjusting weights of a simple compliance sample and a complex dispute sample by using a balance focus loss function; performing quality scoring on the candidate law fine tuning data set, calculating a complexity score of the candidate law fine tuning data set, and generating a quality-complexity scoring data set of the candidate law fine tuning data set; and constructing a dependency tree of a grammar dependency relationship, performing clustering analysis on vectorized dependency tree features, and performing cluster division and sorting on candidate law fine adjustment data sets so as to merge the candidate law fine adjustment data sets into final fine adjustment sub-data sets and complete screening. The problems of semantic distortion and training deviation caused by a traditional screening method are effectively solved, and a systematic solution is provided for efficient training of a large model in the legal vertical field.

The invention relates to the technical field of LINUX operating systems, and discloses a software customization method and system in an ISO mirror image of the LINUX operating system, and the method comprises the steps: recursively analyzing a construction dependency tree and an operation dependency tree of a target component through a pre-analysis module, recognizing a missing component and / or a conflicting component, creating an isolation task for the conflicting component, and generating a customized work task set; through a construction module, deploying a dependency environment of a to-be-constructed component, executing an isolation task, constructing an independent directory for a conflict component, and generating a temporary finished product warehouse; executing a test on the target component in the temporary finished product warehouse through a test module; and through a distribution module, constructing a customized ISO and distributing the customized ISO to a product warehouse. According to the embodiment of the invention, manual demand analysis and dependency relationship processing are replaced by an automatic mode in a software customization system, the dependence on high-skill research and development personnel is reduced, a complex customization process is standardized, and the research and development and implementation cost is reduced.

The invention belongs to the technical field of software management, and particularly relates to a cross-platform repair method and system based on a large language model.The method comprises the steps that a cross-platform package manager interface is used for conducting static analysis on a dependency list of a target software project, a cross-platform dependency tree is obtained, and the target software project is repaired based on the cross-platform dependency tree; respectively injecting lightweight monitoring probes into at least two operating system platforms, and collecting dynamic tracking data of each dependency component in the dependency tree; fusing the dependency tree and the dynamic tracking data, and establishing a cross-platform security map comprising a dependency calling edge, a version conflict edge and a platform vulnerability association edge; according to the method, the lightweight monitoring probe is injected into the multi-operating system platform, the loading sequence, the function call sequence and the memory layout difference of the dependent components are dynamically collected, and the cross-platform security map is constructed in combination with the static dependency tree, so that the specific potential safety hazard of the platform can be accurately identified, and the comprehensiveness and the accuracy of cross-platform risk perception are remarkably improved.

The invention discloses a threat intelligence triple combined extraction method based on data enhancement, comprising a data enhancement method which is responsible for performing multi-granularity high-quality expansion on a small amount of data in the field of threat intelligence through a data enhancement method based on a large model. And the syntactic information fusion method is responsible for analyzing an input sentence into a component tree and a dependency tree, extracting information in the component tree and the dependency tree by combining a Tree-GRU syntactic fusion algorithm to obtain a syntactic dependency matrix, and fusing the syntactic dependency matrix into an extraction model to enhance the understanding of the syntactic dependency matrix on an explicit text structure and semantics. And the form filling extraction method is responsible for simultaneously representing and extracting entities and relationships in sentences in the same form through a joint extraction algorithm based on a form filling method so as to enhance interaction between the entities and the relationships, and introducing an adversarial training mechanism based on an MDA method in training so as to enhance robustness and generalization ability of the model. The invention provides a new design thought for solving the problem of insufficient labels in the network threat intelligence triple extraction field and extracting complex, long and difficult sentences.

The invention belongs to the technical field of data processing, and particularly relates to a fingerprint generation method of a virtual evaluation model based on Hash operation, which comprises the following steps of: firstly, constructing a dependency tree by taking a main model file specified by a user as a root node, and initializing an access record table; establishing a static keyword protection area containing a core parameter rule base, scanning a dynamic file operated by a user, and comparing parameter definitions to generate an early warning; reading the binary data of the current node file, and generating a character string through SHA-256 after standardization; recursively processing the sub-nodes to obtain a node tree hash value, sorting the sub-nodes, splicing the sub-nodes according to a fixed format, and generating a character string through SHA-256 operation; then, a current node standardized path is spliced and generated through SHA-256 operation; and finally, by taking the root node as a starting point and the DFS as a sequential serialization tree structure, executing SHA-256 operation after adding a global collision salt value, and outputting a global unique fingerprint of the model. The problems of poor stability, low efficiency and high generation risk in a model fingerprint generation technology in the prior art can be solved.

The invention discloses a WebGL-based Unity asset cross-platform preview method and system, and a storage medium. The preview method comprises the following steps: 1) preprocessing an editor plug-in: scanning a target asset through the Unity editor plug-in, extracting a resource GUID,. Meta metadata and a dependency relationship, generating a lightweight AB packet and adding a dynamic thumbnail; step 2) cloud dependency analysis: the server side recursively constructs a dependency tree based on GUID, filters redundant resources, and generates a resource loading list which can be identified by the browser side; and step 3) a WebGL rendering engine: loading an AB packet in a browser, realizing particle effect regeneration, animation control and scene interaction by simulating a Unity Runtime environment, and adapting to a WebGL 1.0 / 2.0 rendering pipeline. The invention aims to provide a WebGL-based Unity asset cross-platform preview method and system and a storage medium, aims to solve the problem that a browser cannot directly preview Unity resource files, and provides a set of resource management workflow based on Unity-cloud storage-browser preview for the Unity resource files.

The invention relates to a security risk detection method and device for a software repository, computer equipment, a storage medium and a computer program product. The method comprises the following steps: establishing a dependency tree model of a software package according to version information of the software package in a software repository; the dependency tree model represents the dependency relationship among the software packages through a tree structure; vulnerability information in a vulnerability library is matched with the software packages and the version information, and a target software package matched with the vulnerability information is obtained; in the dependency tree model, vulnerability marking processing is carried out on the target software package, and risk marking processing is carried out on nodes associated with the target software package; and obtaining a security risk detection result of the software repository according to the software package with the risk mark in the dependency tree model. By adopting the method, the timeliness, accuracy and comprehensiveness of security risk monitoring of the software repository can be improved.

The invention relates to the technical field of computers, and provides a cross-platform installation package automatic adaptation method and system, and the method comprises the steps: obtaining a unified description file and an operating system fingerprint of each target platform; based on the operating system fingerprints of the target platforms and the installation package adaptation parameters, matching is carried out, and dependency trees of the target platforms are generated; and generating an adaptive installation package based on the dependency tree of each target platform and the installation package adaptation parameter. According to the cross-platform installation package automatic adaptation method, specific parameters and shared parameters of the platform are managed in a centralized mode through unified description files, environment intelligent perception is achieved and the dependency tree of the platform is generated in combination with dynamically obtained operating system fingerprints, and the adaptive installation package is generated through analysis of the installation package adaptation parameters of the platform and the dependency tree. The problems of low adaptation efficiency, high configuration error rate, high maintenance cost and the like are solved; in addition, a unified construction management mechanism realizes cross-platform centralized management and control, and the accuracy and efficiency of cross-platform installation package adaptation are improved.

System and method for aspect-level sentiment classification. The system includes a computing device, the computing device has a processer and a storage device storing computer executable code. The computer executable code is configured to: receive an aspect term-sentence pair; embed the aspect term-sentence pair; parse the sentence using multiple parsers to obtain dependency trees, and perform edge union to obtain a merged graph; combine the embedding and the merged graph to obtain a relation graph; perform a relation graph neural network on the relation graph; extract hidden representation of the aspect term from updated relation neural network; and classify the aspect term based on the extracted representation to obtain a predicted classification label of the aspect term. During training, the computer executable code is further configured to calculate a loss function based on the predicted label and the ground truth label, and adjust parameters of models.

The invention provides a language translation data processing method and system based on a large model, and the method comprises the steps: carrying out the coding segmentation of language translation data, and obtaining a plurality of text segments; performing dependency syntactic analysis on each text fragment, determining a dependency tree structure of each text fragment, and determining an explicit semantic dependency relationship between every two text fragments according to the semantic similarity between every two text fragments and the dependency tree structure of the corresponding text fragment; performing dependency analysis on an implicit semantic relationship between every two text segments according to the implicit semantic association degree between the text segments and the segmentation loss of the text segments to obtain an implicit semantic dependency relationship between every two text segments; and constructing a paragraph tag of the language translation data through a dependency relationship between explicit semantics and implicit semantics between every two text fragments, and translating the to-be-translated data based on the paragraph tags. By the adoption of the scheme, cross-segment semantic guidance translation of the long complex text can be achieved.

The invention provides an applet permission security detection method and system, and relates to software engineering.The method comprises the steps that API test code snippets of a super application are generated based on a large language model, the large language model is guided according to an API dependency tree to generate a plurality of API test code snippets, and executable test codes are constructed; dynamically executing the executable test code in the simulation environment, and obtaining the actual permission configuration of the applet according to the dynamic execution result; obtaining standard permission configuration of the super application, comparing the actual permission configuration with the standard permission configuration, and if the situation that the actual permission configuration does not accord with the standard permission configuration occurs, determining that the super application has an API permission exception problem; the method comprises the following steps: constructing an abstract syntax tree according to an applet source code, performing granularity optimization on the basis of the abstract syntax tree to construct an abstract method tree, searching an abnormal API (Application Program Interface) with a permission defect in the abstract method tree, determining a function body name where the abnormal API is located, performing front-end and rear-end dual confirmation according to the function body name, and determining a calling condition and a binding condition of the abnormal API. Accurate detection of the permission defect in the super application is realized.

The invention discloses an open source software supply chain security intelligent monitoring and risk assessment method, relates to the technical field of software supply chain security, and solves the problem that only existence of indirect dependence can be identified generally in the prior art, but whether a vulnerability code can be called on an actual execution path cannot be accurately judged. According to the method, through dependence locking and Hash fingerprint embedding during construction, cryptographic-level accurate traceability of a component version is realized, and version ambiguity is thoroughly eliminated; in combination with deep interception of a JVM probe for method calling and class loading events during operation, a loading source and a call stack context of each class in a real execution path can be captured, bidirectional association mapping is performed on dynamic behavior data and a static dependency tree, and whether a vulnerability component is located on an executable call chain or not is accurately identified. Compared with the defect that coarse-grained warning is carried out only through version matching in a traditional tool, the method has the advantage that by means of dynamic-static data fusion, it is avoided that a large number of unavailable indirect dependency vulnerabilities are misjudged to be high in risk.

A method and a system for assessing sentence segmenting in subtitles of a digital content is disclosed. The method comprises of acquiring a source text of the digital content and identifying linguistic boundary within sentences by assigning parts of speech (POS) tags and dependency tags to each word using natural language processing (NLP) libraries. Further, head information is assigned for each word to form a dependency tree structure. And, then, cohesiveness scores are assigned based at least on the parts of speech (POS) tags and the dependency tree structure. The method further includes identifying incorrect lines which violate the linguistic boundary and a set of static rules, and thereby assessing the sentence segmentation in subtitles of a digital content.

The invention relates to the technical field of software migration, in particular to a Linux executable program migration method and device.The method comprises the steps that compatibility detection is conducted on system calling of a target ELF file, and gasket adaptation processing is conducted on incompatible system calling; performing dependency tree total analysis on the target ELF file, and identifying a dependency library required by the operation of the target ELF file; extracting an actual calling symbol list of each dependency library, comparing the actual calling symbol list with symbols of corresponding libraries on the target system, and collecting incompatible dependency libraries of the original system to an isolation directory; modifying a library search path when the target ELF file runs, so as to preferentially load an original system dependency library in the isolation directory; and finally, constructing a migration packet deployed to the target system, and carrying out adaptive migration verification on the migration packet. Compatibility and operation stability of the program in the target system are improved, only the software and the dependency library thereof are migrated, migration difficulty is reduced, and the migration process is accelerated.

The invention relates to a software project vulnerability detection method, device and equipment and a medium, a dependency tree of a software project is generated according to configuration information in a configuration file of the software project, and the dependency tree comprises direct dependency items and indirect dependency items in the software project. Furthermore, according to the attribute information of the direct dependency item and the attribute information of the indirect dependency item, querying vulnerability data obtained from a vulnerability database to determine whether vulnerability exists in the direct dependency item and the indirect dependency item, and according to the direct dependency item and / or the indirect dependency item with the vulnerability, generating a vulnerability detection report. According to the method, the vulnerability detection is performed on the direct dependency item and the indirect dependency item, so that comprehensive detection of the software project is realized, efficient matching with the vulnerability data is performed, a detailed vulnerability detection report is generated, the vulnerability detection accuracy of the software project is improved, and the security of a software project supply chain is improved.

The invention discloses a dynamic SQL (Structured Query Language) optimization method based on consanguinity dependency, and relates to the technical field of database query optimization. The method comprises the following steps: analyzing all indexes in a database, and dividing the indexes into atomic indexes and composite indexes; a blood relationship dependency tree is constructed according to the mathematical relationship between the indexes, each node represents one index, and each directed edge represents the dependency relationship; receiving a to-be-queried index set, and recursively searching all dependent target atomic indexes based on the graph structure; grouping the target atomic indexes according to data sources, and generating corresponding SQL query statements; and executing the SQL statement, and calculating a final query index result after obtaining an atomic index value. According to the invention, through unified management of the index dependence structure, repeated calculation across indexes is eliminated, the query performance is improved, the resource consumption is reduced, and the index development and maintenance efficiency is improved.

Systems and methods for extracting information from documents and constructing corresponding knowledge maps with respect to defined knowledge models. Deep-learning-based models for Natural Language Processing (NLP) are applied to tokenize words, tag, parse, and lemmatize sentences of input documents. Then an information extractor traverses the dependency tree of NLP object to recursively extract the entities of interest to the knowledge models. Finally, a knowledge map constructor traverses the dependency tree of NLP object to determine the relationships among the extracted entities and construct knowledge maps recursively following the defined knowledge models.

The invention discloses an artificial intelligence-based interactive question-answering method and system, and relates to the technical field of interactive question-answering, and the method comprises the steps: converting a user input text question into vector representation, constructing a relation graph by combining a dependency relation tree of a text entity, capturing a dependency relation in the question as an edge weight, forming an adjacent matrix, and constructing a node degree matrix; and determining a standardized graph Laplacian matrix, performing feature decomposition, determining an optimal clustering number, clustering language phrases of the text question, and determining task fragment sets of different clusters. According to the method, semantic association between language phrases is achieved by constructing the dependency relationship graph, meanwhile, syntactic relationships are fused, the integrity of text structure information is ensured, the degree of nodes serves as a quantitative index of local strength in the graph, the information directly supports subsequent Laplacian matrix normalization processing, a unified scale is provided, and the method is suitable for being applied to the field of text processing. And the imbalance problem of the data range is avoided.

Disclosed herein are systems and methods for automatic detection, fixing, and monitoring of vulnerabilities. In one aspect, an exemplary method comprises receiving as an input an artifact, when the received artifact is vulnerable, building the artifact for a new release, scanning a library containing the new release of the artifact to determine whether the library is safe for publishing, and publishing the library to an accessible artifact depository when the library is safe for publishing. In one aspect, the building of the artifact for the new release includes, scanning a dependency tree, replacing the vulnerable version by a safe version of the artifact when a safe version is available, fixing the vulnerable version of the artifact using known patches when a safe version of the artifact is not available, adding a postfix to a group ID of the artifact for distinguishing the fixed version from prior versions of the artifact.

The invention relates to an emotion detection method combining metaphor recognition and a syntactic structure, and belongs to the technical field of natural language processing and emotion analysis. The implementation method comprises the following steps: 1, tagging part-of-speech coarseness of a to-be-analyzed text by utilizing a natural language processing library; extracting and identifying a labeled result to obtain an emotion keyword; 2, performing fine-grained marking on the emotion keywords, and performing serial number marking on positions of the emotion keywords; 3, obtaining a feature vector, a segment vector and a mask vector; 4, screening to obtain a target word of the metaphor; 5, constructing a node relationship of the dependency tree; meanwhile, sorting and splicing are carried out, and then clauses are obtained; 6, performing emotion classification evaluation; the evaluation is used in emotional natural language text data processing; compared with the prior art, the method has the advantages that through cooperative work of the metaphor recognition model and the emotion detection model, the recognition precision of the expression emotion of the metaphor is improved, introduction of irrelevant information is effectively reduced, and high efficiency of emotion analysis is ensured.

The invention relates to the technical field of natural language processing and artificial intelligence, and discloses a large-model complex instruction-oriented fine-grained intention recognition method, which comprises the following steps of: constructing a candidate analysis forest of an input instruction, and executing entity alignment of each node and a domain knowledge graph; performing semantic compatibility verification on the candidate dependency trees by utilizing entity relationships in the atlas, and screening an optimal analytic tree in combination with syntactic probability and knowledge consistency scores; if the analytic tree meeting the threshold value does not exist, a semantic conflict edge is positioned, and candidate mounting points are searched by using a map neighborhood relation to reconstruct a dependency structure; and finally, generating standardized data containing a structured analytic path based on the optimal analytic tree, and performing fine adjustment on the large model. According to the method, logic constraint and dynamic repair are carried out on the syntactic structure by introducing the knowledge graph, analysis errors caused by multiple modification or structural ambiguity in a complex instruction are effectively solved, and the accuracy and robustness of intention recognition of a large model in the vertical field are improved.

The invention discloses an aspect-level sentiment analysis method fusing multi-granularity features and a multi-level fusion network, and the method comprises the steps: firstly obtaining context features and aspect words of a text through a RoBERTa pre-training model, and converting the context features and aspect words into word vectors for representation; secondly, introducing external knowledge by utilizing a graph convolutional neural network (GCN) in combination with a component dependency tree and a structure tree, and extracting a component structure, a dependency relationship, background knowledge and context emotion features of the text; then, a multi-level fusion filtering module is designed, and the features are filtered and simplified through a cross attention mechanism and an orthogonal projection technology; and finally, predicting the emotion polarity of the text by using a Softmax function. According to the method, the effectiveness of integrating various granularity information features in a complex natural language processing task is shown, the comprehensive task performance of the ABSA model is improved through combination of the various granularity features and an efficient feature fusion technology, and the accuracy of aspect-level sentiment analysis can be improved.