153 results about "Interaction nets" patented technology

The invention relates to the field of data visualization, in particular to a visualization method and system based on a vehicle service supply chain. The method comprises the following steps: acquiring whole-process information of a vehicle service supply chain; carrying out supply chain main body identification on the whole flow information of the vehicle service supply chain, carrying out spatial topology distribution analysis, and constructing a supply chain main body topology distribution map; obtaining supply chain main body interaction logic data based on the plurality of supply chain main bodies; performing topology interaction logic connection on the supply chain main body topology distribution graph by utilizing supply chain main body interaction logic data, and constructing a supply chain topology interaction network; performing intra-period path transportation frequency calculation on the whole-process information of the vehicle service supply chain to generate logistics demand trend characteristics; and performing operation behavior analysis of each link on the whole-process information of the vehicle service supply chain to obtain the capacity characteristics of each link. According to the invention, through visualizing the vehicle service supply chain, the resource scheduling efficiency and flexibility of the supply chain are improved.

The invention provides a face target detection method based on improved YOLOv8, and the method comprises the steps: carrying out the preprocessing of a face data set, constructing and training an improved YOLOv8 network model, inputting a to-be-detected image into the model, and obtaining a prediction result; the improved YOLOv8 network model comprises a backbone network, a neck network and a head network; the backbone network comprises a convolution module CBS and a feature fusion module C2f of an original model, a deformable attention mechanism DAttention is introduced to improve the feature fusion module C2f in the backbone network, the C2f of the last layer is replaced, an attention-based intra-scale feature interaction network is used to replace a spatial pyramid pooling module, and the feature fusion module C2f of the original model is obtained. And optimizing the bounding box loss function through the angle constraint loss function. According to the invention, the precision of a face target detection task is effectively improved, and especially for dense small face image detection, the problems of missing detection and wrong detection of targets are reduced.

The invention provides an unstructured data processing method and system, and relates to the technical field of natural language, and the method comprises the steps: carrying out the tensor decomposition of unstructured data, obtaining a standardized feature vector, constructing a self-organizing critical state feature learning network in an isolation calculation space, constructing a feature interaction network based on quantum entanglement perception, and carrying out the recognition of the feature interaction network. And establishing a hierarchical causal graph for feature screening, realizing semantic feature vector generation by adopting multi-scale feature fusion, and establishing an intelligent protection mechanism. The feature extraction accuracy can be effectively improved, the data processing reliability is enhanced, and the system operation efficiency is improved.

The invention provides a cognitive unmanned aerial vehicle network spectrum sharing method based on multi-agent reinforcement learning. The method comprises the implementation steps that a cognitive unmanned aerial vehicle network model and a cognitive unmanned aerial vehicle network and multi-agent reinforcement learning interaction network model are constructed; obtaining a training data set and defining a reward function; and carrying out iterative training on the interactive network model. In the process of training the interactive network model, the multiple agents interact with the cognitive unmanned aerial vehicle network in real time through reinforcement learning, spectrum resource allocation can be adaptively adjusted according to the communication requirement of each cognitive unmanned aerial vehicle and the dynamic change of the environment, and the spectrum resource allocation efficiency is improved. And the loss of the local Q network in each time step is calculated according to the target Q value corresponding to each experience calculated by calculating the reward of interaction between each Q agent and each environment state through the throughput of each cluster member on each channel, so that the resource waste is reduced, the spectrum conflict is avoided, and the maximization of the network throughput is realized. Therefore, the spectrum utilization rate is effectively improved.

The invention provides an industrial time sequence anomaly detection method and device and computer equipment, and belongs to the technical field of industrial time sequence anomaly detection. The method comprises the following steps: acquiring industrial time series data; the time sequence data are converted into graph structure data and then serve as model input data to be input into a reconstruction model, reconstruction data are obtained, and the reconstruction model comprises a digital composite graph attention network; and if the degree of difference between the reconstruction variable data of a certain timestamp in the reconstruction data and the original variable data corresponding to the timestamp in the model input data reaches a preset abnormal threshold, determining that the state of the original variable data corresponding to the timestamp is abnormal. The above reconstruction model extracts the superior representation of the variable interaction network based on knowledge according to the knowledge graph related to the industrial system, and fuses the superior representation with the time sequence dependency based on the sequence, thereby reducing the risk that the model learns pseudo-correlation, and improving the anomaly detection performance and anomaly interpretability of the industrial time sequence.

The invention discloses a multi-source monitoring data fusion method based on deep learning, and the method comprises the steps: carrying out the standardization and space-time dimension alignment processing of multi-source monitoring data, constructing a multi-scale data sub-sequence, inputting a multi-scale space-time feature interaction network based on a Mamba structure, extracting and interacting multi-scale space-time features, constructing a cross-modal data topological graph structure, and carrying out the fusion of the multi-source monitoring data. And dynamically calculating and adjusting attention weights of nodes and edges of the topological graph by using an adaptive cross-modal graph attention mechanism, generating dynamically optimized cross-modal fusion features, performing collaborative feature decoding, and outputting a fusion result. According to the method, the multi-source data fusion precision and generalization ability in a complex monitoring scene are improved, and the fusion feature expression ability and decision reliability are improved.

The invention discloses a deep forgery detection method based on a multi-granularity collaborative attention mechanism, and the method comprises the following steps: extracting multi-level features through a backbone network, achieving the self-adaptive region division and weight fusion through a dynamic space grouping attention mechanism, and solving the problem that a conventional method is insufficient in attention to a distributed forgery region; a double-path channel decoupling module is designed to separate high-frequency artifacts and low-frequency semantic features, and feature coupling interference is eliminated; a cross-granularity feature interaction network is constructed, local details and global semantic features are collaboratively optimized, and the fusion efficiency of multi-scale forged clues is improved; and finally, a detection result is output through the classifier. According to the method, the defects of weak cross-domain generalization ability, low multi-granularity information utilization efficiency and the like caused by incomplete space attention coverage and channel feature interference in the prior art are overcome, the detection precision and robustness are remarkably improved, the stable performance is kept in a complex degradation scene, and the method can be widely applied to scenes such as digital content security auditing and identity authentication.

The invention relates to the related technical field of rehabilitation information communication interaction, in particular to a real-time interaction method and system suitable for multi-modal rehabilitation medical information, and the method comprises the steps: determining a multi-modal data flow, generating an initial interaction network architecture, using a heterogeneous data feature encoder, adding an upper sampling feature pointer and a lower sampling feature pointer, and carrying out the real-time interaction of the multi-modal rehabilitation medical information. According to the method, dynamic allocation optimization is carried out by using dynamic adaptive elements, short-time communication adjustment is carried out by using different interactive requests according to the dynamic adaptive elements, configuration optimization is carried out according to real-time scheduling of multi-modal data streams, and a self-adaptive interactive network architecture is set, so that the problem that under different interactive requests and data streams, the real-time scheduling of the multi-modal data streams is not influenced is solved. According to the technical scheme, the technical problems that the multi-modal data feature extraction and fusion capability is insufficient, and interaction requirements cannot be met are solved, path selection optimization in a multi-path transmission topological structure is achieved, short-time communication adjustment is carried out according to different interaction requests and data flow conditions, and it is ensured that the interaction requests can be quickly responded and processed.

The invention provides a federal cross-modal retrieval method and system based on interactive prompts, and relates to the field of cross-modal information retrieve.The federal cross-modal retrieval method and system based on interactive prompts retrieve similar second modal data from a second modal data set by using first modal data based on the feature similarity of two modal data comprises the steps that initial features of the two modal data are extracted respectively; performing multi-layer bidirectional interaction between the initial features of the two modals by using a cross-modal interaction network obtained by federal learning and taking a prompt vector as an intermediary to obtain final features of the two modals after cross-modal interaction; calculating feature similarity based on the final features of the two modalities, and screening similar second modal data; according to the method, federal learning and prompt learning are combined, so that the effectiveness and universality of a cross-modal retrieval technology are improved, and the problems of privacy protection and performance optimization in cross-modal retrieval are solved.

The invention discloses a bidirectional GCN-BERT scientific data classification method based on rotary coding and dynamic gating, which comprises the following steps of: firstly, preprocessing original text data, acquiring a classification mark containing global information by utilizing a pre-training model, and adding the classification mark to the beginning of an input text; secondly, feature optimization is carried out on the classification marks through a CorNet neural network, and then position enhancement and global interaction are carried out through a rotation position enhancement multi-layer feature interaction network RP-MLFIN; and then the image is transmitted to a bidirectional image convolution neural network for feature interaction and image convolution processing, and image level representation is generated. Finally, the classification marks after feature optimization and the representation of the graph level are transmitted to a classifier, BERT prediction and GCN prediction are generated, prediction results are fused through an attention mechanism, and a final classification decision is generated. According to the method, the processing capability of text data with strong context dependence and the stability and robustness of a classification result are improved.

The invention relates to the technical field of computer mode recognition, and discloses a multi-source heterogeneous data fusion processing and key feature extraction method and system, and the method comprises the steps: achieving the adaptive caching and granularity normalization of streaming data through a dynamic buffering queue and a time alignment window; generating a structured vector of a unified space-time reference by using a structured analysis module; a high-dimensional fusion feature tensor is constructed through two-stage convolutional coding and a cross-source attention interaction network; and a key feature channel is screened based on gradient sensitivity through a differentiable channel pruning module. The system comprises a multi-source data access unit, a dynamic buffer management unit, a time alignment unit, a synchronous resampling unit, a structured analysis unit, a primary fusion coding unit, a cross-source attention interaction unit, a time sequence dependence modeling unit, a feature importance evaluation unit, a key feature screening unit and the like. According to the method, efficient, accurate and low-overhead multi-source heterogeneous data real-time fusion and task-oriented key feature extraction can be realized.

The invention provides a business data attribution analysis method and system based on artificial intelligence. The method comprises the following steps: determining a key attribution dimension related to a business result; modeling an interaction relationship between dimensions through a graph neural network; performing dynamic adjustment on an interaction weight between dimensions in the dimension interaction network by using a self-attention mechanism of Transform; evaluating the influence and propagation path of the dimension in the dimension interaction network through a PageRank algorithm; a GBDT model is used for evaluating the contribution degree of each dimension to a service result, and sorting is carried out; and calling a large language AI model to generate an attribution report causing the service result based on the service data, the dimension interaction network, the influence and the propagation path of the dimensions, the contribution degree of each dimension to the service result and the sequence of the contribution degree. According to the method, the interaction relationship between business data dimensions is analyzed, the dimension weight is dynamically adjusted, and the influence and contribution degree of the dimensions are evaluated, so that a comprehensive, deep and insight attribution report is generated, and a scientific basis is provided for business decision making.

The invention discloses a complex traffic flow prediction method based on dynamic graph convolution and interactive learning, and belongs to the technical field of intelligent traffic, and the method achieves the effective long-term traffic flow prediction through embedding a graph convolution network into an interactive learning structure, and capturing the time and space dependence at the same time. A new dynamic graph convolution method is used, a combined graph generated from an adaptive and learnable adjacent matrix in real time is used to capture spatial correlation changing in real time in a traffic network, then a space-time adaptive converter is integrated, global and local features can be extracted, and the spatial correlation is obtained. According to the method, the challenge of traffic flow prediction aiming at complex correlation and nonlinearity of urban traffic flow in time and space can be solved, the advantages of dynamic graph convolution and interactive learning are combined, a new efficient model of dynamic graph convolution and interactive network is provided, the dynamic spatial and temporal characteristics of the traffic flow are reflected, and the traffic flow prediction efficiency is improved. Therefore, traffic flow prediction can be realized more accurately.

The invention discloses a transport capacity resource integrated intelligent scheduling method and system oriented to multiple service scenes, and relates to the field of intelligent scheduling, and the method comprises the steps: constructing a hierarchical collaborative decision-making architecture comprising a macroscopic strategic layer agent and a microscopic tactical layer optimizer; and inputting the structured feature vector into a hierarchical collaborative decision-making architecture, dynamically distributing proper solution algorithms and parameters for a microscopic tactical layer optimizer according to a real-time scheduling situation by utilizing an online element learning optimizer, and outputting a pre-scheduling scheme. According to the invention, through integrated integration of multi-service scene data, comprehensive utilization of static basic information, real-time transport capacity data and prediction environment data is realized, and the data support capability of scheduling decision is improved. The feature interaction network of the multi-layer perceptron structure can accurately extract core features and provide effective input for scheduling decisions. The hierarchical collaborative decision-making architecture is combined with an online element learning optimizer, so that a solution algorithm and parameters can be dynamically matched, and a better pre-scheduling scheme can be output.

The invention provides a long-term action prediction method based on a multi-modal large language model and a related device, and the method comprises the steps: carrying out the frame extraction of a video according to a fixed sampling rate, and predicting text action labels corresponding to all video frames; inputting the text action tag into a multi-modal large language model to obtain a text feature, a visual feature and a preset action query feature; constructing a cross-modal interaction network architecture, and taking the text features, the visual features and preset action query features as input to obtain feature expressions after interaction of different modals; and obtaining an output result of action tuning based on the feature expression after interaction of different modalities and an action tuning module, then training a multi-modal large language model, and performing optimization solution on the network by using a loss function to obtain a prediction result.

The invention discloses a traffic incident detection method based on deep space-time memory and an interactive network. Firstly, a space-time information sensing network is designed, core semantics from a small space area to a large space area and short-term features in a time domain can be effectively captured from input video frames, the extraction capability of video frame features is improved, and traffic events are classified more accurately. Secondly, a time sequence feature learning network is provided, and the problem that long-term time information cannot be fully captured when a spatio-temporal information sensing network processes video frames is solved. According to the method, more complete context time information features can be obtained by strengthening learning of the time features, and accurate positioning of exception occurrence and end moments is improved.

The invention provides a blue-green algae image recognition method and system based on hierarchical self-adaption and domain driving, and the method comprises the steps: enhancing an image through employing an improved dark channel algorithm; constructing a blue-green algae biological attribute text database, and performing synonym replacement and sentence pattern recombination; multi-scale visual features are extracted through a hierarchical adaptive Swin Transform model, and key region characterization is enhanced in combination with dynamic spectrum attention; the text is input into a Bio-ALBERT model, and semantic embedding of field optimization is generated through term mask prediction and attribute relation pre-training; constructing a two-layer heterogeneous graph by using a graph attention interaction network GAIN, calculating a cross-modal association weight through a bidirectional graph attention mechanism, and outputting a cross-modal graph feature; multi-scale cross-modal association is modeled through a hierarchical graph attention fusion mechanism, and a comparison alignment loss optimization model is combined; and high-precision blue-green algae identification is realized. According to the method, through multi-scale perception, domain semantic adaptation and graph structure fusion, the accuracy of blue-green algae detection in a complex environment is improved.

The invention discloses a traffic scene training data generation method and device, electronic equipment and a medium, and relates to the technical field of intelligent traffic, and the method comprises the steps: obtaining multi-modal traffic data, building a dynamic semantic interaction network according to the multi-modal traffic data, and enabling the dynamic semantic interaction network to comprise a plurality of traffic entities, the dynamic attribute of each traffic entity and the space-time relationship between the traffic entities are determined; in the dynamic semantic interaction network, labeling the target event and a core node of the target event according to a preset event specific sub-graph; performing causal chain backtracking on the labeled target event and the core node of the target event to obtain structured causal chain data corresponding to the target event; and performing question and answer pair generation processing on the structured causal chain data corresponding to the target event to obtain an instruction fine tuning data set for training the traffic large model. Therefore, automatic and high-quality generation of the traffic scene training data is realized, and the logical reasoning ability and interpretability of the model are improved.

The invention relates to the technical field of streaming media detection, in particular to a complex element new communication streaming media detection method based on a multi-modal deep learning recognition technology, which comprises the following steps: S1, multi-modal time-space synchronization preprocessing: mapping video key frames, audio clips and bullet screen texts to a unified time axis through a combined time-space calibration technology, establishing spatial semantic association; s2, hierarchical multi-modal feature distillation is carried out, and discriminative multi-granularity features including local details, global semantics and cross-modal association modes are extracted from all modals; and S3, establishing a dynamic graph modal interaction network, constructing a learnable multi-modal relation graph, and dynamically modeling cross-modal semantic interaction. According to the complex element new communication streaming media detection method based on the multi-modal deep learning recognition technology, the problem that cross-modal complex semantic collaboration cannot be captured through single-modal analysis or shallow fusion, so that the detection missed judgment rate is high is solved.

The invention discloses a power transmission line real-time regulation and control method and system based on CNN and Transform cooperation, and relates to the technical field of power transmission line real-time regulation and control, and the method comprises the following steps: constructing a multi-modal input stream, respectively extracting a feature map and a time sequence feature according to the multi-modal input stream based on a preset dual-channel deep interaction network, and carrying out the calculation of the feature map and the time sequence feature; the dual-channel deep interactive network is constructed based on CNN and Transform fitting; establishing an incidence matrix based on the feature map and the time sequence features, wherein the incidence matrix comprises feature weights in the incidence matrix adjusted based on a preset cross-modal gating unit; performing feature complementary enhancement on the feature map and the time sequence features based on the incidence matrix to generate fusion features; and constructing a prediction model according to the fusion features, and outputting a power transmission line regulation and control strategy. According to the method, through weighted adjustment of different modes, mode conflicts and noise interference are effectively suppressed, the robustness of the model is improved, and the stability and accuracy in a complex environment are ensured.

The invention relates to a city scene safety field text question and answer method and system. The city scene security domain text question and answer method comprises the steps of collecting a city industry data source and preprocessing the city industry data source to obtain a first city security domain text; encoding the first city security domain text, importing the first city security domain text into a preset attention interaction network, and automatically extracting to obtain a first city security domain corpus; and performing structured processing on the first city safety corpus to obtain a second city safety field corpus, and constructing a city safety field knowledge base according to the second city safety field corpus and multivariate information data. According to the method, the key information and the resource position information in the text can be automatically identified, the thinking chain is constructed based on the key information and the resource position information, the answer to the question is gradually deduced, deep understanding and intelligent question answering of the security scene text are realized, and powerful guarantee is provided for city security.

The invention discloses a multi-modal interaction sentiment analysis method based on semi-supervised learning. According to the method, firstly, feature extraction is carried out on voice, text and visual modals, then feature fusion is carried out through an intra-modal interaction branch network and an inter-modal interaction branch network, finally, fusion information of the two interaction branches is extracted in a self-adaptive mode based on a gating mechanism, and the intra-modal interaction network and the inter-modal interaction network are based on a transformer encoder structure. The difference is that the provided attention mask mechanism is introduced into the multi-head self-attention layer; according to the training method, a pseudo label is generated in a semi-supervised learning mode to assist in stable training of the model, and dependence on label data is reduced. According to the method, the problems of high manual annotation cost and how to mine effective interaction information in the field of multi-modal sentiment analysis are solved, and the sentiment recognition precision is improved.

The invention relates to the technical field of social network application, in particular to a social media viewpoint evolution simulation method and device based on coupling dynamics, and the method comprises the steps: extracting event topic interaction network data based on social network public information, carrying out the statistics of user historical interaction behaviors, and determining an information transmission network and an internal association network according to the data, the two forms a multi-layer association network; determining a transmission state of forwarding and a text viewpoint baseline for calculating a transmission state and a viewpoint value of a user in the social network data set; and inputting the propagation state and the viewpoint value into a coupling dynamics simulation model, carrying out iterative calculation until the viewpoint value converges so as to obtain a final propagation state and a final viewpoint value of the user, and generating forwarding situation distribution and user viewpoint distribution. Therefore, the problems that errors are generated, public opinion monitoring and early warning and public opinion guide strategy construction are affected and the like due to the fact that a viewpoint evolution model adopts a single-layer propagation structure and coupling modeling is not carried out on propagation and an internal correlation structure in the related technology are solved.

The present application relates to the field of OTN resource matching, and more particularly to an OTN resource intelligent matching method based on artificial intelligence, comprising: inputting the preprocessed service demand, available wavelength resource, real-time bandwidth utilization and support OTU level into a feature extraction model based on a residual multi-layer perception architecture to generate a prediction base feature; inputting the prediction base feature into a feature interaction model based on a compressed interaction network architecture to generate a service comprehensive interaction feature; and inputting the service comprehensive interaction feature into an OTN resource matching model based on an expert gate network architecture to generate a recommended OTU level. Through the hierarchical intelligent model processing of the residual multi-layer perception architecture, the compressed interaction network architecture and the expert gate network architecture, the present application realizes the significant improvement in the accuracy, efficiency, adaptability and network operation value of OTN resource matching.

The invention relates to a sentiment classification method and device based on a perceptual fusion network, equipment and a medium. The method comprises the steps of obtaining a to-be-recognized image and a to-be-recognized text; extracting object features and scene features in the to-be-recognized image, and extracting text features of the to-be-recognized text; constructing an interactive network by any two of the object features, the scene features and the text features, and performing feature learning processing based on the interactive network to generate basic features of different modals; and carrying out feature fusion on the basic features of the different modes to obtain target fusion features, and carrying out emotion classification based on the target fusion features to obtain target emotion tags. According to the method, the object features, the scene features and the text features are extracted, details and background emotion clues in images and texts can be captured, meanwhile, an interaction network is constructed for feature learning and feature fusion is performed in different modes, emotion understanding between the different modes can be enhanced, and the accuracy of emotion classification can be improved.

The invention discloses an Ethereum user behavior analysis method based on a heterogeneous condition coding and decoding architecture, and belongs to the technical field of artificial intelligence and block chain behavior analysis. The method comprises the following steps: modeling an Ethereum interaction network into a directed heterogeneous graph for distinguishing a user type and an interaction type, and defining six meta-interaction types to carry out one-hot coding; aggregating neighbor source user features from a target user perspective to generate neighborhood features, and inputting the target features and meta interaction features into an encoder to output an intermediate hidden state; reconstructing and generating features through a feature decoder, and aggregating different interactive features by using attention to obtain enhanced features of a target user; splicing the original features and inputting the original features into a heterogeneous graph neural network to detect malicious behaviors; meanwhile, neighborhood features and hidden states are symmetrically generated from the perspective of a source user, and an adjacent matrix is reconstructed through a structure decoder to construct an auxiliary learning target. According to the method, refined interactive semantic modeling is realized, multi-type interactive modes are adaptively learned, topological relation characterization is enhanced, and the detection accuracy is improved.

The invention relates to the technical field of seabed engineering, and discloses a submarine cable construction simulation method under complex seabed geological conditions. The method comprises the following steps: slicing and recombining multi-source investigation time sequence data, and constructing a geologic feature space based on a multi-dimensional time sequence window and time-space synchronization. Through feature fusion and dimension mapping, three-dimensional geological semantic body units with consistent time and space are generated, and attributes of the three-dimensional geological semantic body units and construction process parameters are associated and coded. And a recursive segmentation algorithm is adopted, and layered and nested dynamic construction decision units are automatically divided according to an attribute parameter mutation threshold. On the basis, a simulation agent is initialized for each decision-making unit, and dynamic re-evaluation of attribute parameters is driven through an agent interaction network. According to the method, the fidelity of the simulation model to the space-time evolution characteristics of the complex geological conditions is improved, the construction decision can adaptively respond to the local mutation of the geological parameters, and the simulation accuracy and the engineering practicability are enhanced.

The invention discloses a method and a device for adjusting a service interaction behavior in a service interaction network, and electronic equipment. Relates to the field of artificial intelligence, and the method comprises the following steps: obtaining candidate interaction strategy data sets corresponding to a first account set and a second account set in a service interaction network, the first account set and the second account set having a service interaction behavior; simulating a service interaction behavior corresponding to the candidate interaction strategy data set by using the target model to obtain a simulation result; determining an interaction satisfaction degree corresponding to the simulation result, and determining target interaction strategy data from the candidate interaction strategy data set based on the interaction satisfaction degree; and according to the target interaction strategy data, adjusting a service interaction behavior in the service interaction network, so that the first account set and the second account set execute the adjusted service interaction behavior. Through the method and the device, the technical problem of low adjustment accuracy of the service interaction behavior in the related technology is solved.

The application discloses a multimodal interactive sentiment analysis method based on semi-supervised learning. The method first extracts features of voice, text and visual modalities, then fuses the features through two interactive branch networks of intra-modal and inter-modal, and finally adaptively extracts fusion information of the two interactive branches based on a gating mechanism. The intra-modal and inter-modal interactive networks are based on the encoder structure of the transformer, except that the multi-head self-attention layer introduces the proposed attention mask mechanism. In terms of training method, the method generates pseudo-labels to assist model training in a semi-supervised learning manner, thereby reducing the dependence on labeled data. The application solves the problems of high artificial labeling cost and how to mine effective interactive information in the field of multimodal sentiment analysis, and improves the sentiment recognition accuracy.

The invention relates to an intelligent logistics POI recommendation method based on semantic knowledge distillation and interpretability. The method comprises the steps of obtaining data for training; according to a feature engineering module, encoding each type of data, inputting obtained user identification features, POI attribute features and user behavior features into a deep interaction network of a teacher model, and calculating a preliminary interaction score between a user and a POI based on node features of user nodes and POI nodes after updating of a graph neural network; a lightweight model is used as a basic framework of a student model, semantic knowledge of a teacher model is inherited through adaptive semantic knowledge distillation, test data is input into the student model inheriting the semantic knowledge, final interaction scores of a user and POIs are output, a plurality of corresponding POIs with the highest final interaction scores are used as recommendation results, an SHAP value is calculated, and a recommendation result is obtained. And the recommendation result is explained. According to the method, efficient recommendation, semantic maintenance and decision transparency can be realized at the same time.