30 results about "Similarity relation" patented technology

The embodiment of the application belongs to the field of artificial intelligence, and relates to a data analysis method based on artificial intelligence, comprising the following steps: acquiring voice dialogue data of a user and an agent in a business conversation process; converting the voice dialogue data into text data, and obtaining dialogue text by converting the text data; performing classification processing on the dialogue text according to a preset rule to obtain a theme label corresponding to the dialogue text; constructing a corresponding directed graph based on the dialogue text and the theme label; and performing similarity analysis processing on the directed graph based on a preset similarity analysis model to generate a similarity result between each directed graph. The application also provides a data analysis device based on artificial intelligence, a computer device and a storage medium. In addition, the application also relates to blockchain technology, and the similarity result can be stored in the blockchain. The application analyzes the similarity relationship of the theme path of the dialogue technique based on the use of the similarity analysis model to quickly and accurately obtain the corresponding similarity result.

The application discloses a video search method and device and a computer readable storage medium. The method comprises the following steps: obtaining a video search text, and performing text feature extraction on the video search text to obtain first text features; mapping the first text features to a video feature space corresponding to video information to obtain second text features; performing video feature extraction on each candidate video in a candidate video library to obtain first video features of each candidate video; mapping the first video features to a text feature space corresponding to text information to obtain second video features of each candidate video; and searching for a target candidate video corresponding to the video search text in the candidate video library based on a first similarity relationship between the first text features and the second video features and a second similarity relationship between the first video features and the second text features. The method can greatly improve the accuracy of searching for videos based on text.

This application provides a method, apparatus, and computer device for constructing training samples for a recall model. The method is applicable to shopping platforms and includes: obtaining at least one second product displayed after a user triggers a preset behavior for a first product on a preset page of the shopping platform; the first product and any second product have a similarity relationship; obtaining at least one third product in the shopping platform that matches the attribute information of the first product; constructing any product pair based on at least one second product and at least one third product, inputting the product pair and a prompt word into a large language model to obtain the similarity corresponding to the product pair; the prompt word is used to instruct the large language model to determine the similarity corresponding to the product pair from the product name and the ingredient list of the product pair; removing product pairs with similarity less than a second threshold to obtain similar product samples, the similar product samples are used to train the model to obtain a first recall model, and the first recall model is used to recall products with substitution relationships.

The invention provides a knowledge graph construction method and device, equipment, a medium and a product, and relates to the field of data processing, and the method comprises the steps: constructing a financial knowledge base through a financial knowledge platform frame, and enabling the financial knowledge base to provide a multi-modal feature vector and structured relation data corresponding to multi-modal financial data; performing knowledge annotation on the multi-modal feature vector and the structured relation data to obtain annotated data; inputting the annotation data into the domain model, and obtaining summary information of text blocks output by the domain model and semantic similarity relationships of different text blocks; the domain model is related to the financial domain; the text blocks are obtained by slicing the annotation data, and the semantic similarity relationship is determined according to the summary information; performing entity relationship analysis on the multi-granularity hierarchical structure and the semantic similarity relationship of the annotated data to generate maps of different levels; and performing optimization processing on entities corresponding to the maps of different levels to construct the financial knowledge map. And the dependence on manpower and the cost are reduced.

The embodiment of the application discloses a data deduplication method and device, a computing device cluster and related products, and belongs to the technical field of big data. In the embodiment of the application, a similarity relationship between a plurality of data in a first data set is generated according to the similarity between the plurality of data; a first deduplicated data is determined according to the similarity relationship, the first deduplicated data being the data with the largest number of similarity relationships with other data in the first data set; and the first deduplicated data is deleted from the first data set. On the one hand, data deduplication is achieved, and on the other hand, compared with a deduplication scheme that only retains the data with the largest number of similarity relationships, the data deduplication scheme provided by the embodiment of the application can also retain a larger number of representative data. Therefore, compared with the deduplication scheme that only retains the data with the largest number of similarity relationships, the method provided by the embodiment of the application not only achieves data deduplication but also achieves data expansion, thereby improving the quality of the deduplicated data set.

The invention belongs to the technical field of radar radiation source identification. The invention provides a radar radiation source identification method based on relation perception and prototype optimization. According to the embodiment of the invention, through channel gating and a soft attention mechanism, invalid frequency band features are automatically inhibited, and the weight of a bad sample which deviates from the center is reduced. And a GCN is introduced to construct a sparse prototype graph, so that the model can perceive a similarity relationship between categories, and local consistency is enhanced through a graph propagation mechanism. And in cooperation with prototype comparison loss, centers of similar categories are forcibly pushed away, so that a clearer isolation strip is formed at a decision boundary, and the false identification rate is remarkably reduced. A residual mean fusion structure is also designed. When an attention mechanism may fail due to extremely small sample size, a mean value prototype on a residual path can play a role in guaranteeing the bottom, so that the stability of model training is ensured; and when the sample size is slightly large, the attention mechanism can provide finer feature expression, so that the universality of the model under different small sample settings is realized.

The application discloses a reciprocal interlayer-time discriminant target model for robust visual tracking, and belongs to the technical field of target tracking, and is used for solving the problem that traditional twin tracking algorithms rarely consider information interaction of templates and search regions, and that cumulative error of similar target interference influences tracking results. The application firstly constructs an interlayer target perception enhancement model, realizes interlayer feature information interaction by establishing pixel-by-pixel correlation between templates and search regions in a feature extraction process, reduces cumulative error caused by invisible target to search regions, and enhances target perception; meanwhile, in order to weaken the influence of interference, a time interference evaluation strategy is designed, similarity between multiple candidate positions in adjacent frames is established by using an interframe candidate propagation module, similar interference is eliminated according to the determined similarity score, a more reliable target position is obtained, and robust tracking is realized.

The invention discloses a common preference enhanced collaborative recommendation method and device, and the method comprises the steps: obtaining user implicit interaction data in a target domain and a source domain, and constructing corresponding user features and commodity feature representations; carrying out fusion coding on the target domain user features, the source domain user features and the target domain commodity features, and generating cross-domain migration features facing target domain recommendation tasks; performing fusion iteration de-noising processing on the cross-domain migration features, gradually inhibiting source domain noise interference and strengthening migratable collaborative preference information through multiple rounds of feature fusion and weight screening operation, and generating de-noised cross-domain user feature representation; after denoising is completed, user structure constraint is introduced, and consistency constraint is carried out on the similarity relation between the target domain users before and after feature enhancement, so that the overall distribution stability of the target domain user interest structure is kept; and finally, performing interactive modeling on the final user representation and the target domain commodity features to obtain a matching relationship between the user and the commodity. The device comprises a processor and a memory. Through a collaborative modeling mode of fusing iterative denoising and structural constraint, the problems of noise interference in cross-domain recommendation and user similarity structure change before and after feature migration can be relieved, and the stability and accuracy of a recommendation result are improved.

This invention discloses a refined model test method, system, and equipment for the soil displacement effect of open-pit pipe piles, relating to the field of open-pit pipe pile driving technology. The method includes: establishing multi-physical quantity similarity criteria and determining similarity constants; based on the similarity constants, preparing grouting mixture and constructing a micro-reinforcing cage; using the micro-reinforcing cage and grouting mixture to construct an open-pit pipe pile model; establishing a model box, preparing soil samples with target density and homogeneity using a layered filling method within the box, and setting up a pile driving test platform; constructing a comparative test scheme, and simultaneously monitoring and recording pile driving data using the pile driving test platform; processing and comparing the pile driving data to obtain data criteria, providing a basis for optimizing open-pit pipe pile driving parameters and construction control. This invention solves the problems of incomplete model similarity relationships, single test monitoring dimensions, difficulty in quantifying and comparing results, and the inability to provide an operational basis for optimizing engineering construction parameters.

This invention discloses a method for determining the non-perfect similarity ratio based on the classification similarity law. The method includes the following steps: S1, extracting a set of parameters describing the physical phenomenon based on experience, previous experimental results, literature, and data; S2, dividing the parameter set into key parameters, basic parameters, and derived parameters; S3, obtaining the similarity ratio of the basic parameters using measurement, material property testing, and other means; S4, deriving the similarity ratio of the derived parameters based on the similarity ratio of the basic parameters; S5, introducing constraint equations based on the physical relationship between the key parameters and the basic and derived parameters to construct a similarity ratio constraint equation; S6, substituting the similarity ratios of the basic and derived parameters into the similarity ratio constraint equation to solve for the similarity ratio of the key parameters. This method, by introducing constraint equations to construct similarity relationships, can effectively reduce scaling errors under non-perfect similarity conditions and is applicable to various scaling test scenarios such as open-loop and closed-loop tests.

The invention relates to the technical field of information retrieval, in particular to an image retrieval model training method, an image retrieval method, equipment and a medium. According to the method, multiple encoders focus on different aspects of images, diversified feature representations can be generated, dominant features of each image are recognized in combination with reconstruction loss (a bottleneck mechanism), the similarity relation between the images is enhanced, and effective capture of inter-class differences and intra-class changes is achieved; according to the method, interaction between individual features is promoted by means of meta-relationships, semantic aggregation of the features is achieved, and the generated binary codes can retain complex semantic relationships through relationship consistency loss optimization.

The application discloses a super-relation extraction method based on bidirectional time LSTM and multi-relation contrast learning, and belongs to the technical field of information extraction. The BiTime-LSTM is introduced, time-dependent relations are effectively captured through a bidirectional gate learning mechanism, main gates divide memory cells into regions with different time sensitivities, and slave gates perform fine-grained updates in different time regions, so that the performance of the model in a time prediction task is significantly improved. Meanwhile, a multi-relation contrast learning method is introduced, the predicted triad relation and the qualifier relation are compared with real labels, the similarities and differences between various relations are identified, discriminative feature representations can be learned, and the confusion problem between similar relations is effectively alleviated.

The invention belongs to the technical field of e-commerce user behavior analysis, and particularly relates to an e-commerce user re-purchase behavior prediction and accurate reaching method and system fused with a time sequence attention mechanism. The method comprises the following steps: acquiring a historical behavior sequence of a target user in a preset statistical period, extracting a behavior type, behavior occurrence time and commodity information, and constructing a time sequence enhanced behavior representation sequence containing a relative time difference of adjacent behaviors and a time interval from the latest purchase behavior; performing multi-head weighting processing based on the behavior time proximity relationship, the behavior type conversion relationship and the commodity semantic similarity relationship to obtain a user purchase intention vector; and further performing cross-cycle offset analysis in combination with the historical purchase intention vector of the previous statistical cycle, generating a re-purchase score and an offset risk mark according to an analysis result, and outputting a re-purchase prediction result. According to the method, the current behavior characteristics and the cross-cycle interest migration state of the user can be represented at the same time, and the dynamism and discrimination of re-purchase identification are improved.

According to the distortion compensation method and system of the coal and gas outburst physical simulation similarity criterion, a gas-solid coupling control equation of coal and gas outburst physical simulation is selected, and a similarity ratio relational expression of key parameters is established; processing the similarity ratio relational expression to obtain a first similarity criterion; based on an effective stress theory and a thermodynamic theory, constructing a mechanical strength attenuation function of the porous medium material under the action of gas pressure; carrying out a mechanical test under different gas pressure conditions to measure material strength parameters of the prototype and the model; according to the material strength parameters, strength attenuation factors under different gas pressure conditions are calculated, and corresponding strength attenuation functions are obtained through fitting; combining the intensity attenuation function with a first similarity criterion, deducing to obtain a distortion compensation equation of the similarity criterion, and constructing a distortion compensation similarity criterion; and finally determining a distortion compensation value of each physical property parameter. According to the method, the correction and compensation of the physical simulation similarity relation under the action of the gas pressure can be realized, and the accuracy of the physical simulation result is improved.

The invention provides a local consistency guided sparse label enhancement method, which is suitable for detecting a drivable area on a road and belongs to the technical field of images. The method aims at solving the problem that an existing deep learning model depends on a large amount of pixel-level annotation data, firstly, sparse annotation is conducted on an input image, context enhancement features are constructed according to local and global image representation, and the similarity relation between super-pixel nodes is established; and then constructing a label propagation model based on the graph convolutional network, and propagating the sparse labels to the unlabeled areas to generate pseudo labels. According to the method, a weak supervision training strategy guided by local consistency is adopted, a joint loss function is designed, and collaborative supervision is carried out on a labeled region and an unlabeled region, so that the reliability of a pseudo label and the overall segmentation precision are improved. Experimental results show that the method can be suitable for various road drivable area detection tasks, and the obtained high-quality pixel-level pseudo label can be used for subsequent fully supervised model training.

The application relates to a data processing method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: obtaining three-dimensional structure data and resource exchange information of a historical object; performing feature extraction on the three-dimensional structure data to obtain feature data of the historical object; determining a feature vector of the historical object based on the feature data; storing the feature vector and the identifier of the historical object in a target database; obtaining a feature vector of a query object, matching a target historical object matching the feature vector of the query object from the target database according to the similarity relationship between the feature vector of the query object and the feature vector of each historical object; and generating reference resource exchange information of the query object according to the indexed resource exchange information of the historical object. The method can improve the accuracy and efficiency of resource exchange information generation.

This invention discloses a method for constructing a digital twin model of a core engine rotor based on dynamic similarity, relating to the field of aero-engine condition monitoring technology. The method includes: establishing the dynamic equations of the prototype rotor based on the structural parameters and dynamic characteristics of the prototype core engine rotor system; determining the similarity relationship between the prototype rotor and a similar rotor based on dynamic similarity theory; determining the design parameters of the similar rotor based on the similarity relationship; and constructing the similar rotor based on the design parameters. This invention obtains time-domain characteristic response vectors by collecting and processing the dynamic response data of the similar rotor, capturing its dynamic characteristics; constructing a dynamic similarity mapping operator to realize the feature transformation between the similar rotor and the prototype rotor, alleviating the constraint of limited accuracy in finite element modeling; and inputting the prototype rotor features into the digital twin model to drive its update and reconstruction, achieving high-fidelity deduction of the internal state of the core engine rotor and improving the accuracy and reliability of its dynamic characteristic analysis.

This invention relates to the fields of natural language processing and information extraction technology, specifically to a few-sample relation extraction method based on memory enhancement and hierarchical decision-making. The method integrates the original text, structured hints, and learnable vectors into a unified input text sequence. It then determines whether a relationship exists between entities. If a relationship exists, it proceeds to the type enhancement stage, generating a relation type representation. A two-layer memory bank is then constructed to generate context-aware soft-hint vectors. During the training phase, uncertainty is assessed based on the information entropy of the sample prediction probability distribution, and learning weights are dynamically adjusted to optimize the model's ability to distinguish similar relations. During the inference phase, when the decision weights exceed a threshold, a large language model is introduced for joint decision-making, ultimately outputting the relation category. This method effectively solves the problems of low accuracy in identifying unrelated samples and difficulty in distinguishing semantically similar relations in few-sample relation extraction, significantly improving the model's generalization ability and robustness.

The application relates to the technical field of data processing and information management, and discloses a customer archive digitization collection and semantic database construction method and system. The application obtains customer original data through a multi-source collection terminal, carries out multi-modal fusion processing and standardization conversion on the data, generates a data set with a unified structure, further carries out feature extraction and vector expression on the standardized data, constructs a customer feature vector set, calculates the semantic similarity relationship between customers based on the feature vectors, generates a semantic correlation matrix and constructs a semantic graph structure, maps the semantic graph to a customer archive semantic database on this basis, and constructs an index structure through a dynamic updating mechanism, so that continuous updating and efficient retrieval of customer data are realized. The application can effectively solve the problems that multi-source heterogeneous customer data is difficult to be uniformly processed and semantic relationship is difficult to be modeled, and improves the intelligent level and retrieval efficiency of customer archive management.

The invention discloses a dynamic prototype coding and prompt self-construction method for zero sample relation extraction. The method comprises the following steps: S1, defining a task; s2, adaptive prompt words are generated through a dynamic prompt generator, and a dynamic prompt template is constructed; s3, dynamically generating and aggregating context-related prototype representations through a dynamic prototype aggregator; and S4, training and testing. While the sentences and prototype semantics are accurately aligned, the distinguishing ability of the model to complex relationships and similar relationships is improved.

This invention discloses a deep knowledge tracking method and a question recommendation method that integrates multiple features. The deep knowledge tracking method includes the following steps: S01. Pre-training an initial question representation vector matrix by inputting the absolute difficulty matrix of the questions and the similarity relationship between questions; S02. Using the student's answer interaction sequence set and the pre-trained initial representation vector matrix as input, constructing a student answer interaction sequence matrix, a historical relevance matrix, and a multi-knowledge point answer accuracy matrix, and concatenating them to form a comprehensive information matrix, which is then passed through a gating mechanism GLU to obtain the G matrix output; S03. Extracting the student's learning state matrix from the G matrix using multiple one-dimensional convolutional layers; S04. Obtaining the prediction result of the relative difficulty based on the question representation vector to be predicted and the learning state matrix. This invention has the advantages of simple implementation, high prediction accuracy and efficiency, and the ability to effectively handle questions of multiple knowledge points simultaneously.

The application discloses a diagnostic method based on intra-layer similarity relation, comprising the following steps: 1. According to the history answer record of students, the correlation between exercises and knowledge points, the similarity among students, exercises and knowledge points is calculated to form intra-layer similarity relation; 2. The embedding representation of nodes is learned by using the intra-layer similarity relation graph convolution network, and the nodes with less exercise records or knowledge point correlation information are called student tail nodes or exercise tail nodes. The head node information is transmitted to the tail nodes to improve the sparsity of the inter-layer relation of the tail nodes; 3. The cognitive state of students is obtained by fusing the student vector and the knowledge point vector, and then the cognitive state of students and the knowledge point mastery required by exercises are obtained. The application realizes the information transmission of tail nodes by increasing the intra-layer similarity relation among students, exercises and knowledge points, so that the cognitive diagnosis precision of students with less exercises can be improved.

The embodiment of the invention provides a counterfeiting risk identification method and device for a voucher image. Nodes in the voucher feature graph represent voucher images, features of the nodes comprise visual layout features and text semantic features of the voucher images, and edges represent visual layout feature similarity relationships, text semantic feature similarity relationships and visual semantic confrontation relationships among the voucher images. And determining a plurality of matching nodes matched with the first voucher image feature from each node included in the voucher feature graph, and determining a first association relationship between the first voucher image and the plurality of matching nodes according to the image feature. Then, a local sub-graph is determined from the voucher feature graph, the local sub-graph comprises a first voucher image node, a plurality of matching nodes and neighbor nodes associated with the first voucher image node, and edges between the first voucher image node and the plurality of matching nodes are constructed based on a first association relationship; and inputting the local subgraph into a risk identification model to obtain a forgery risk probability value of the first voucher image. The voucher image contains privacy data.

The application discloses an article recommendation method, device, medium and equipment, comprising: obtaining a first association result according to an external information resource; obtaining a second association result according to historical operation data of a user, and obtaining a third association result according to article index data; obtaining a meta-path sequence set according to the first association result, the second association result and the third association result; training a machine learning model according to the meta-path sequence set to obtain a vector embedding representation of the article; calculating a fourth association result between articles according to the vector embedding representation of the article; and determining a target article in the article according to the historical operation data of a target user and the fourth association result to recommend the target article to the target user. The application relates to the fields of natural language processing and content recommendation, and similarity between labels is constructed by introducing external information to calculate similarity between articles, so that the diversification and accuracy of article recommendation are improved.

The invention discloses a model training method and device, a classification method and device, computer equipment and a storage medium, and the method comprises the steps: extracting a first feature vector from a plurality of entity words in a first training sample through a first model; obtaining a first classification probability based on the first feature vector; determining a sample relationship pair formed by each first training sample and a plurality of second relationships; obtaining the similarity corresponding to the sample relation pair based on the first training sample in the sample relation pair and the second relation; determining a first weight of the sample relation pair based on the similarity and a first classification probability and a first relation corresponding to the sample relation; and based on the first weight of the sample relation pair and the corresponding first classification probability, determining the first loss of the first model to adjust the parameter of the first model, thereby endowing the first weight to the sample relation pair to calculate the loss so as to adjust the model parameter, enabling the model to pay more attention to similar relations, and improving the accuracy of the model. And the model can distinguish the similar relationship.

The invention discloses a territorial space region similarity relation mining method. The method comprises the following steps: acquiring spatial grid data to be mined and spatial data of a target area; constructing a sliding window; calculating the spatial grid data to be mined by using a sliding window to obtain a plurality of windows and index features corresponding to the plurality of windows; calculating the total similarity between each window and the target area according to the spatial data of the target area and the index characteristics corresponding to each window; and obtaining similar windows of the plurality of target areas according to the similarity between each window and the target area. The method has the characteristics of accurate similarity judgment and simple calculation.

The invention discloses a petition data set construction method based on a large language model, and relates to the technical field of natural language processing, and the method comprises the steps: collecting petition case information, and obtaining a standard petition text object through preprocessing; based on a large language model, performing semantic structuring processing on the standard petition text object to generate structured case features; according to the structured case features, a case similarity relation graph is constructed through similarity correlation analysis, comprehensive evaluation is carried out, and difficulty hierarchical case objects are obtained; selecting an alternative standard path through a multi-target path search algorithm based on the difficulty hierarchical case object, and generating an anti-fact path candidate set; and according to the anti-fact path candidate set and the case similarity relation graph, performing hierarchical sample arrangement on the difficulty hierarchical case objects, and constructing a petition data set. According to the method, the alternative standard path is selected by executing the multi-target path search algorithm, so that the decision reference capability and generalization robustness of a large language model in a complex petition scene are improved.

The application relates to a bridge model design method and system based on static and dynamic information fusion, and belongs to the technical field of bridge design. The bridge model design method based on static and dynamic information fusion comprises the following steps: obtaining physical quantities of a prototype and a scale model of a bridge; determining static dimensionless parameters based on the physical quantities and a dimensional analysis method; establishing a multi-degree-of-freedom motion equation; determining dynamic dimensionless parameters based on the multi-degree-of-freedom motion equation and the physical quantities; determining a similarity relationship between the prototype and the scale model based on the static dimensionless parameters and the dynamic dimensionless parameters; performing distortion analysis on the scale model based on the similarity relationship, determining an actual similarity relationship between the prototype and the scale model, and completing scale model design based on the actual similarity relationship.

The invention discloses a multi-view multi-label classification method based on hypergraph adaptive high-order semantic fusion, and the method comprises the steps: carrying out the adaptive modeling of a group semantic similarity relation in a view and a group semantic alignment relation between views through employing a hypergraph structure, and introducing a label-driven comparison learning mechanism; structuring consensus representation and view specificity representation which are higher in discrimination capability are constructed, and finally classification is conducted by combining the two kinds of representations. The method comprises the following four parts: (1) constructing a multi-view hypergraph; (2) performing high-order semantic fusion; (3) label-driven contrast learning; and (4) multi-label classification. According to the method, high-order semantic interaction can be directly expressed, a label-driven contrast learning mechanism is introduced, and interference caused by nodes with similar feature spaces and irrelevant label spaces is effectively avoided by drawing close the consensus of similar samples and deducing different samples through adaptive hypergraph construction.

The invention discloses a radiology report generation method based on enhanced bimodal features, and belongs to the technical field of radiology report generation methods. According to the method, a Visual Mama visual encoder is used for enhancing visual features to obtain an enhanced visual feature map with transverse and longitudinal association relationships and two-dimensional space information; the report similarity relation matrix module dynamically learns report similarity relation information of tokens in the text features based on a similarity relation learning matrix so as to enhance the text features; when the feature enhancement condition decoder obtains an enhanced feature map based on the enhanced visual feature map and the text features output by the report similarity relation matrix module and decodes the enhanced feature map, a predicted radiology report which is better close to a real report can be obtained. According to the present invention, the test results show that the BLEU-1 index, the BLEU-2 index, the BLEU-3 index, the BLEU-4 index, the METEOR index and the ROUGE-L index obtained by using the method of the present invention are good.