197 results about "Similarity computing" patented technology

The invention belongs to the technical field of air traffic control and conduction, and particularly relates to an intelligent question answering system method for air traffic control and conduction business knowledge. According to the method and the system, the refined and context-aware information retrieval capability of the air traffic control communication service knowledge and the expert-level air traffic control communication service knowledge question and answer capability are provided, the knowledge management and application efficiency is effectively improved, and powerful support is provided for digital transformation of related services. Comprising the following steps: S1, a document content extraction and storage process; s2, a content retrieval and generation process; s3, content slice optimization is carried out through extraction and similarity calculation of adjacent paragraphs or contents, complete contents are in the same slice, and the content retrieval efficiency is improved; according to the method, the label library and the business rule library are continuously iteratively optimized, so that the provided service is always in the optimal state.

The invention provides a scientific and technological achievement intelligent matching and transaction recommendation method and system based on a knowledge graph, and relates to the technical field of science and technology, and the method comprises the steps: extracting a technical gene sequence based on a scientific and technological achievement data set, generating a technical state vector, constructing a technical association intensity matrix and a technical iteration path graph, and calculating a technical potential energy distribution field; and forming an initial knowledge graph. And monitoring the fluctuation of the knowledge graph state feature set through the intelligent sensor, and dynamically optimizing the knowledge graph according to the fluctuation condition. Converting technical requirements of a user into a query state, performing state similarity calculation with the knowledge graph, identifying an optimal technical unit combination, predicting a state evolution trajectory, evaluating a regional adaptation degree, and generating a technical implementation scheme. By dynamically optimizing the knowledge graph and accurately matching the user demands, the efficiency and accuracy of scientific and technological achievement matching and transaction recommendation can be effectively improved, and scientific and technological achievement conversion is promoted.

The invention discloses a document link construction and evolution relation tracking method based on fragment-level semantic alignment, which relates to the technical field of document management, and comprises the following steps: carrying out structured analysis on a plurality of input documents, and dividing the document content into a plurality of semantic fragments; performing vectorization processing on each semantic fragment, and constructing a document-fragment-vector semantic mapping relationship; different document fragment-level semantic alignment is realized through cross-document semantic vector similarity calculation, a position offset tolerance algorithm and context association analysis, and a semantic link network is established; fusing the link relationship, the timestamp, the version information and the author information, and constructing a document evolution diagram taking the documents or fragments as nodes and evolution links as edges to record semantic flow paths between the documents; and outputting the semantic link network and the document evolution diagram in a structured form. The method can solve the problems that semantic relations between documents are missing, evolution paths are invisible, content reuse cannot be traced, and the structure induction capability is insufficient.

The invention relates to the technical field of administrative management, in particular to an NLP-based file approval method, which comprises the following steps: inputting an approval file, carrying out word segmentation processing on the content of the approval file, carrying out similarity calculation on a word segmentation result and a preset service type dictionary, and generating a file service classification result. According to the method, the file service classification result is formed through word segmentation processing of the approval file content, so that the approval process can be automatically classified based on text features, and dependence on fixed rules is reduced. In combination with semantic vector conversion and keyword semantic association calculation, the semantic understanding ability of approval contents is enhanced, accurate matching of approval files and organization structure responsibilities is ensured, and approval errors caused by insufficient keyword matching are avoided. By calculating the matching value, association is established between the approval file and the department responsibilities and between the approval file and the post responsibilities, so that allocation of the approval tasks is more targeted, and task execution difficulty caused by mismatching of the responsibilities of approvers is avoided.

The invention discloses a unified storage management method for station data in expressway data, which comprises the following steps of: S1, constructing a multi-source heterogeneous data acquisition channel, and processing various data sources by using edge computing nodes to generate a standardized metadata packet; s2, establishing a dynamic fragment storage matrix, and performing three-dimensional slicing on the data from time, space and business dimensions; s3, designing an event-driven data association engine, establishing a space-time mapping relation, and realizing automatic association storage through event ID chain type binding; s4, deploying a multi-modal storage cluster, storing different data by using different storage units, and tracking data consanguinity through a distributed index service; s5, implementing dynamic vectorization authority management, calculating an access authority level according to cosine similarity, and establishing a fine-grained data access control model; and step S6, constructing a service data interface gateway. The method has remarkable advantages in the aspects of highway data acquisition, storage, association, management, sharing and the like.

The invention relates to an intelligent expert extraction system based on deep learning. The intelligent expert extraction system comprises an expert database module, an expert information index module and an intelligent extraction and matching module, the expert database module is used for storing expert information from different fields and carrying out standardization processing; the expert information indexing module is used for intelligently indexing expert information by using a deep learning model and an information retrieval technology, receiving bid evaluation task requirements, screening out experts most conforming to conditions and providing matching degree scores; and the intelligent extraction and matching module converts the bid evaluation task into a task feature vector through an adaptive algorithm, performs similarity calculation on the task feature vector and the expert feature vector, and automatically extracts the most matched expert. According to the method, the standardized expert database is established, intelligent indexing and feature extraction are performed by using the deep learning model, and accurate matching between the experts and the bid evaluation task is realized through the adaptive algorithm, so that high efficiency, justice and transparency of the bid evaluation process are ensured.

The invention relates to the technical field of Web page intelligent arrangement, in particular to a Web page intelligent sensing method, which comprises the following steps of: acquiring the type, the content and the attribute of a Dom element of a current Web page, and performing intelligent analysis on acquired data by utilizing an AI (Artificial Intelligence) adaptation algorithm; cleaning by using the collected data; generating a target task pair; generating text embedding for the target text content by using an Embedding vector model; performing Lora task fine tuning on the large language model; embedding the fine-tuned large language model in the AI workflow; dynamic routing identification and logic control of multiple Web pages are realized through an automatic workflow; carrying out Web element similarity calculation by utilizing a multi-modal visual model; outputting a calculation result and coordinate positioning of the elements; according to the invention, the system can process complex Web page elements and user interaction more intelligently; by predicting and adapting to user behaviors, the system can operate more stably and reliably, and faults caused by human errors or system inadaptation are reduced.

The invention discloses a supplier matching method and system based on multi-source heterogeneous data fusion, and relates to the field of building engineering collection analysis, and the method comprises the steps: carrying out the feature analysis of a purchase demand text, and generating a structured purchase demand vector; dynamically adjusting the weight of the key index in the supplier portrait based on multi-source data fusion and historical time sequence prediction; selecting a feature mapping relation in combination with engineering types, extracting Top-K portrait indexes most relevant to demands, and performing weighted aggregation on key portrait indexes in the same supplier to generate exclusive representation vectors thereof; and finally, through similarity calculation of the vector and a purchase demand vector, high-precision candidate supplier screening is realized.

The invention provides a cattle face recognition method and system based on A-SH-BNet, and relates to the technical field of visual inspection. The method comprises the following steps: processing a to-be-recognized cattle face picture through a pre-trained cattle face feature extraction model to obtain a to-be-recognized cattle face feature vector; and finally, carrying out similarity calculation on the to-be-recognized cattle face feature vector and a cattle face feature vector in a database to obtain a recognition result. Wherein the pre-trained cattle face feature extraction model is formed by training an A-SH-BNet network, and the A-SH-BNet network is that an SE module and an HS Block module are embedded in an ArcFace framework; the ArcFace provides a ResNet and a MobileNet, an SE attention mechanism is introduced into a backbone network of a bottom layer of an ArcFace structure, and a Bottleneck structure at the end of a stage in the ResNet is replaced with an HS Block module. According to the method, an SE module in an A-SH-BNet network enhances feature representation by means of a channel attention mechanism; and the HS Block module can extract fine-grained features and global features at different levels through a hierarchical division mechanism, so that the feature extraction precision is further improved, and the cattle face recognition precision is improved.

The invention provides a multi-source heterogeneous data intelligent matching system applied to a department and wound supply chain online platform, and relates to the technical field of data matching, and the system comprises a data preprocessing module which is used for carrying out the standardization processing of department and wound resource data and generating a data quality score; the cross-modal semantic understanding module is used for receiving a user query, performing unified semantic feature representation on the user query and the medical and invasive resource data, and identifying a user query intention; the intelligent matching engine module is used for performing similarity calculation according to the user query intention and the semantic feature representation thereof and the semantic feature representation of the medical and invasive resource data, generating a matching score in combination with a data quality score, and outputting a matching list; and the feedback optimization module is used for collecting feedback data of the user to the matching list, feeding back the feedback data to the intelligent matching engine module, recalculating the matching score and updating the matching list until a preset stop condition is met. According to the invention, the matching precision and efficiency of the medical and invasive resources can be improved.

The invention discloses a building construction progress intelligent monitoring method and device and electronic equipment, and belongs to the technical field of intelligent monitoring, and the method comprises the steps: comparing a label of a building construction image with a building information model library, and obtaining a two-dimensional visual angle image data set; performing semantic segmentation on a bounding box of the building construction image to obtain a segmentation mask set; calculating the cosine similarity of the two-dimensional visual angle image data set and the segmentation mask set according to the feature vector cosine similarity, and determining the angle corresponding to the maximum value of the cosine similarity as the best matching angle; based on the optimal matching angle, a two-dimensional optimal matching view angle image and an optimal matching mask are determined, and the construction progress of the building construction image is determined; semantic segmentation is carried out on the bounding box, and parts related to the building are extracted in a targeted mode; and by calculating the cosine similarity between the two-dimensional visual angle image data set and the segmentation mask set, similarity matching is carried out on the building construction image and the building information model library, so that the monitoring accuracy is further ensured.

The invention discloses a position and knowledge similarity-based software crash error repair scheme recommendation method, which outputs a repair method for error codes according to generated crash stack tracking information and corresponding codes when software crashes, and can help developers to quickly and efficiently solve the problem of software crash. When the method works, firstly, a context knowledge base is constructed, problem posts containing crash stacks on Stack Overflow are preprocessed, a knowledge base and an abnormal dictionary are constructed, and a language model is trained; thirdly, a related question set is preliminarily obtained through stack tracking, the comprehensive similarity of the question and an input stack is calculated through the position related similarity and the additional knowledge similarity, and the most related question and answer set is returned; and finally, performing relevancy calculation on the answer paragraphs based on multiple features, selecting paragraphs related to the crash reason by applying an MMR algorithm, and outputting root cause analysis and repair code snippets related to the crash error.

The invention belongs to the technical field of industrial data processing, and provides an intelligent material matching method and system fusing multi-modal features and fuzzy matching, and the technical scheme is that data in obtained electronic component list data is analyzed, key characters in character strings are identified, and a constructed mapping table is called to carry out mapping replacement on the key characters, so that the matching accuracy of the electronic component list data is improved. Obtaining each field parameter of the mapped electronic component; performing matching based on the constructed manufacturer alias knowledge graph and the mapped manufacturer field parameters to obtain a manufacturer matching result; screening the mapped material number parameters to obtain a material number candidate set, performing semantic similarity calculation based on the material number candidate set and a constructed special word vector model in the field of electronic components, when the similarity is greater than a set threshold, performing accurate matching, otherwise, triggering fuzzy matching, calculating a service score according to a matching result, and obtaining a service result; and performing multi-objective optimization based on a service score result to obtain an optimal matching scheme. And the matching accuracy and the purchasing decision-making efficiency are obviously improved.

A dynamic preference perception online course recommendation method based on a Hawkes process and a TimeXer mechanism comprises the steps that user data and course data are acquired, and the data comprise user attributes, historical interaction sequences and course attributes; generating a context semantic embedding vector by adopting a pre-trained BERT language model; projecting an embedded vector to a unit hyperspherical space by using a spherical Gaussian model, and measuring similarity by using a spherical distance so as to relieve the problems of data sparsity and cold start; a user dynamic preference model integrating a Hawkes process and a TimeXer attention mechanism is constructed, the Hawkes process calculates event occurrence intensity to capture self-excitation characteristics, the TimeXer attention mechanism is adjusted through the intensity information, and weighted user representation reflecting preference dynamic changes is generated; and according to the similarity between the weighted user representation and the projected course embedding vector, calculating a recommendation score, and generating a personalized online course recommendation result. According to the method, the accuracy of online course recommendation is improved by accurately capturing the dynamic preference of the user and effectively processing the sparse cold start problem.

The invention discloses an intelligent operation and maintenance alarm causal relationship analysis method based on BERT. According to the method, the semantic similarity of core switch link interruption and access layer equipment offline is improved by 40% through semantic modeling upgrading and joint coding of equipment types and alarm contents, and compared with an independent coding scheme, the misjudgment rate is remarkably reduced. Through organic combination of deep semantic modeling and dynamic time sequence analysis, the crossing of the intelligent operation and maintenance alarm causal relationship from rule driving to data intelligent driving is realized, and an efficient analysis tool is provided for intelligent operation and maintenance of a complex system. According to the method, accurate causal modeling of complex system alarm data is realized through semantic similarity calculation and multi-dimensional weight dynamic fusion under time sequence constraint, and the method is widely applied to large-scale distributed system fault analysis and root cause positioning scenes in industries such as power systems, communication networks, internet platforms, cloud computing infrastructures and industrial Internet of Things.

The embodiment of the invention provides a network attack homologous analysis method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a plurality of preset internet protocol addresses and a preset weighted attack behavior graph; dividing a plurality of preset internet protocol addresses in the preset weighted attack behavior graph into a plurality of initial address clusters; for each initial address cluster, determining a connection weight associated with each preset internet protocol address contained in the initial address cluster, and calculating a corresponding modularity according to the connection weight; according to the modularity corresponding to each initial address cluster, a plurality of target address clusters are obtained again, and the feature centroid of each target address cluster is determined; and obtaining a to-be-analyzed internet protocol address, and performing similarity calculation on the target feature vector of the to-be-analyzed internet protocol address and the plurality of feature centroids of the plurality of target address clusters to obtain a network attack homologous analysis result of the to-be-analyzed internet protocol address. Therefore, the efficiency and accuracy of network attack homologous analysis can be improved.

The invention relates to the technical field of power system dispatching and maintenance management, discloses a word vector optimization equipment name intelligent matching method based on an equipment knowledge graph, and aims to solve the problem of low matching precision caused by nonstandard naming, OCR recognition errors and semantic ambiguity of power distribution network equipment names. The method comprises the following steps: constructing an equipment knowledge graph containing a triple structure; corpora are generated based on the atlas, a Word2Vec model is adopted to train equipment word vectors, and semantic features are captured; calculating the character similarity through the editing distance and the pinyin similarity; using word vector cosine similarity to match semantic association; and carrying out weighted fusion on the character and semantic similarity, and dynamically judging a matching result. According to the method, through multi-modal similarity fusion and knowledge reasoning, the matching accuracy can be improved, naming regionalization, abbreviation and topological association problems are effectively processed, scheduling instruction analysis, maintenance data standardization and cross-system data management are supported, and reliable technical support is provided for intelligent management of a power grid.

The invention discloses a script-free UI automatic testing system and method based on a swan monk system, and relates to the technical field of UI automatic testing. The method comprises the steps that a testing frame is deployed to a swan monk 5.0 development environment, a to-be-tested interface component tree is traversed, and if an element with the same unique identifier as a to-be-tested interface element component does not exist in an element library, the script-free UI automatic testing system is started; if yes, an element auxiliary analysis instruction is sent out; after an element auxiliary analysis instruction is received, obtaining an attribute library text of the interface element, converting the attribute library text into a vector, and calculating a second attribute similarity of the to-be-tested interface element according to a second attribute vector of the to-be-tested interface element and a second attribute vector of the interface element in the element library; according to the first attribute similarity and the second attribute similarity of the to-be-tested interface elements, the comprehensive matching coefficient of the to-be-tested interface elements is calculated, and manual calibration early warning is sent outwards; and after matching of all the interface elements is completed, an interface element operation flow chart is generated, parameterized testing is achieved, and the efficiency and quality of automatic testing can be improved.

The invention relates to an electricity consumption prediction method and system based on fusion RAG and personalized graph federated learning, and belongs to the technical field of electricity consumption prediction. The method comprises the following steps of: preprocessing historical electricity consumption data of each municipal district, converting the historical electricity consumption data into a data frame, and improving data quality by removing a full null value column, eliminating abnormal data, filling missing values by adopting linear interpolation and the like; carrying out normalization processing on the data by utilizing a Z-score and Min-Max standardization method; a graph neural network is adopted as a basic model structure, an adjacent matrix is generated in combination with geographic position information, and the space incidence relation between communities is captured; an RAG knowledge base is introduced, special scene structured external knowledge is integrated, an embedded vector is generated through Sension-BERT to achieve historical scene retrieval, a retrieval result and model prediction are fused in combination with a gating mechanism, under a federated learning framework, a server initializes a plurality of personalized models, and a user carries out training locally to upload updated model parameters. And the server calculates an aggregation weight between the users through the functional feature similarity.

The invention discloses a data mining system and method for mail data, relates to the technical field of data retrieval processing, solves the technical problem of inaccurate retrieval matching caused by single semantic analysis retrieval, and improves data relevance by unifying multi-source data formats and combining an enterprise address book to supplement metadata. Structured data standardization and unstructured text vectorization are carried out, a computable semantic space is constructed, keywords are screened through a matching proportion, a word frequency weight is calculated in combination with a BM25 algorithm, the keyword matching accuracy is improved, a semantic vector is generated by utilizing a pre-training language model, and semantic correlation is calculated through cosine similarity. Keyword matching scores and semantic similarity scores are dynamically fused, accuracy and comprehensiveness are balanced, accuracy and recall rate are calculated in real time, retrieval deviation types are automatically recognized, keyword weights and similarity threshold values are adjusted in a targeted mode or synonym lists are expanded, and self-adaptive optimization of the system is achieved.

The invention provides an OPC UA information model automatic construction method based on a knowledge graph and a large language model. The method comprises the steps of obtaining an unstructured industrial text, performing semantic analysis on the unstructured industrial text, and extracting a main body entity, an object entity, a method entity and attribute information thereof according to a predefined hierarchical tuple structure; calling an industrial field knowledge graph in real time to perform semantic enhancement on an extraction result; performing similarity calculation and evaluation on an enhanced result, and performing information fusion and duplicate removal based on a similarity evaluation result to obtain an entity-attribute set; generating a tree-shaped hierarchical structure framework by using a large language model, and calling a knowledge graph to perform structural integrity verification; semantic association between the entity-attribute set and the tree structure framework is analyzed, attributes and methods are mapped to corresponding structure nodes, and a mapping reasoning process is output in a natural language form; converting the mapped model structure into a format of XML (Extensive Makeup Language) grammar, and executing automatic compliance check and repair; by designing a progressive task chain of information extraction, semantic fusion, model construction, interpretable mapping and standardized output, a set of complete understanding-enhancement-construction-verification automation system is constructed.

The invention relates to the technical field of intelligent questioning and answering, in particular to a personalized intelligent questioning and answering system based on artificial intelligence. The method comprises the following steps: activating a face photo acquisition system based on question voice, acquiring a face photo of a current user, extracting a target feature representing a face from the face photo, carrying out similarity calculation on the target feature and a high-dimensional feature stored in a question and answer client, and judging the legitimacy of the user based on the similarity; substituting the question voice of the legal user into the voice text recognition model, and outputting a voice content text; and calculating module similarity between a correlation vector matrix between the entity references and a relation matrix corresponding to the central entity reference of the answer library, and outputting a preset answer text corresponding to the central entity reference with the maximum module similarity, thereby improving personalization, accuracy and intelligence levels of questions and answers, improving information acquisition efficiency, and improving user experience. And the expandability and the flexibility of the system are enhanced.

The invention discloses a semantic conflict recognition method and device based on event element extraction, and the method comprises the steps: carrying out the event detection of an input text, recognizing whether an event exists in the text, and determining the type of the event; determining an event element mode according to the event type; performing event element extraction on the event element mode corresponding to the text by adopting a pre-training model to obtain a labeling sequence of event elements; aligning events from different sources, and mapping the events to the same event framework after successful matching; performing semantic similarity calculation on event elements in the successfully matched multi-source event pairs; and judging whether the event elements have semantic conflicts or not according to a preset similarity threshold. According to the method, the accuracy and reliability of semantic conflict recognition can be improved, and the requirements for high-quality information processing in the fields of multi-source information fusion, public opinion monitoring, news reports and the like are met.

The invention relates to the technical field of scientific and technological achievement information management, and particularly discloses an enterprise scientific and technological achievement management system which specifically comprises the steps that a browsed text set and a non-browsed text set currently retrieved by a user are obtained in a calibration period based on a current retrieval scene, the text browsed by the user is divided into the target text and the non-target text through the behavior data, the target text and the non-browsed text are subjected to cosine similarity calculation, the feature value corresponding to the non-browsed text by the user is determined, the feature value is ranked and recommended to the user according to the interest feature value, and the reading efficiency of the user is improved by screening the text which the user is interested in. Therefore, the scientific and technological achievement management and conversion efficiency is improved.

The invention discloses a DMP data management method based on artificial intelligence, and relates to the technical field of data management.The method comprises the steps that multi-device user behavior data are collected through a cross-platform SDK, a browser plug-in and an application log, and data standardization is conducted on a central server or an edge computing node; in the user identity matching process, in combination with a device fingerprint and deep learning method, a user cross-device behavior mode is analyzed through a Transform model, static device features and dynamic behavior features are extracted, a self-attention mechanism is adopted to learn cross-device associated features, then the user matching degree is calculated through cosine similarity, and finally the user identity is judged. The stability and accuracy of identity matching are improved; and based on a user identity matching result, carrying out time sequence analysis by utilizing an LSTM (Long Short Term Memory) model, capturing evolution of user interests along with time, dynamically adjusting user portrait weights through reinforcement learning, and optimizing feature weight updating by using a Q learning mechanism, so that the portraits can be adaptively changed.

The invention discloses a semantic prompt automatic labeling method in a building construction scene, and relates to the technical field of intelligent construction, and the method comprises the steps: collecting a construction plan, on-site surveying and mapping data and real-time environment information, dividing a construction site into construction operation units, and extracting task features to generate a task priority coefficient Kt and a scene state vector Ei; calculating a semantic prompt priority index SPI based on Kt and Ei, sorting prompt words, and mapping the prompt words into feature vectors through BERT; extracting image global features by using a visual model, performing weighted fusion on the image global features and semantic prompt features, and outputting a detection region set; through comparative analysis of a semantic consistency coefficient Ssem and a threshold value Qsc, optimal prompt phrases are optimized and combined to generate a multi-granularity labeling structure; and calculating a semantic anomaly coefficient SAI by adopting cross-modal cosine similarity, comparing the semantic anomaly coefficient SAI with a threshold value Qsai to judge whether the annotation is abnormal or not, and realizing anomaly correction, history comparison and responsibility judgment by recording and tracking the whole process of the annotation link.

The invention provides an intelligent conference multi-modal interaction optimization method and system based on a large model, and the method comprises the steps: obtaining conference multi-modal data, and constructing a locked image frame set based on an image content snapshot locking mechanism; inputting each piece of modal data of the multi-modal data stream into a pre-trained large model to generate embedded vectors corresponding to the modal data, performing anchor point extraction by using similarity calculation between each pair of embedded vectors, and screening through the large model to obtain a semantic anchor point set; when the conference is carried out, each new speech is converted into an embedded vector, and then the similarity between the embedded vector and each semantic anchor point is calculated to obtain an anchor point reference frequency set; and according to the anchor point reference frequency set and the anchor point state set, generating a real-time interaction suggestion, and based on the interaction suggestion, feeding back an updated anchor point state and an updated anchor point priority, and updating the interaction suggestion. According to the invention, automatic identification, structured representation and process state perception of the conference content are realized.

The invention discloses a federated semi-supervised medical image segmentation method and system based on time sequence fluctuation aggregation and cross-client pseudo label mining, and the method comprises the steps: initializing global segmentation model parameters, and distributing the global segmentation model parameters to a plurality of federated clients; after each client receives the global segmentation model parameters, model updating is carried out based on the local annotated data and the unannotated data, and performance indexes are calculated on the verification set; the server collects the performance indexes uploaded by the clients and executes time fluctuation weight updating; each client generates a cross-client pseudo tag through feature clustering and semantic similarity calculation; constructing a neighborhood structure entropy loss function; and the server performs weighted aggregation on the model parameters uploaded by the clients according to the dynamically distributed aggregation weights to obtain updated global segmentation model parameters, distributes the updated global segmentation model parameters to the clients for iterative training until the global segmentation model converges, and performs image segmentation by using the global segmentation model.

The invention relates to the technical field of multimedia information processing, in particular to a recommended video feature storage and quick retrieval method based on picture similarity, which comprises the following steps: S1, establishing an alternative video service database; s2, original video feature extraction: after an original video is obtained, extracting picture feature value, character feature value and face feature value information of key frames of the original video by adopting a video key frame processing mode which is the same as that in the step S1; s3, retrieving characteristic values of alternative video picture frames; and S4, calculating the similarity of the alternative videos. According to the method, the accuracy and retrieval speed of video recommendation can be effectively improved, and more convenient and personalized video browsing experience is brought to users.

The present invention provides a ship design knowledge retrieval method and system based on a multimodal knowledge graph, the method comprising: collecting and constructing a ship design set from the field of ship design; forming a hierarchical multimodal knowledge graph that describes concepts, physical properties, and design specifications in the field of ship design; enhancing the association between different modal data by vector similarity calculation; integrating the hierarchical multimodal knowledge graph into a ChatGLM model; locating knowledge nodes related to user needs in the hierarchical multimodal knowledge graph; matching and locating ship design knowledge related to user query content, and feeding back the ship design knowledge obtained by intelligent semantic retrieval to the user in a hierarchical form. The present invention realizes ship design knowledge retrieval based on multimodal knowledge graph-enhanced ChatGLM, which improves the information acquisition efficiency of engineering designers and the practical application value of the system.