161 results about "Score vector" patented technology

The invention provides a geological data driving path optimization and settlement prediction method for pipe jacking construction, and relates to the technical field of artificial intelligence. The method comprises the following steps: firstly, acquiring sparse drilling point location geological data, constructing a geological data set including soil layer types, forming continuous geological feature tensors through multi-dimensional interpolation, and generating a path vector sequence by combining starting and ending points and building distribution to represent crossing tracks of different paths under geological conditions; and constructing a path graph topological structure through a graph coding neural network, performing feature propagation, obtaining a path comprehensive score vector, and screening an optimal path meeting structural integrity and settlement response constraints. Further identifying settlement trend sensitive points by utilizing the settlement risk prediction values and the accessibility scores of the nodes, constructing a settlement trend map and calculating settlement transaction coefficients. And the transaction coefficient is fed back to a path generation link and is used for iteratively optimizing path sampling density and node distribution so as to realize dynamic correction and stable convergence of a path scheme.

The invention relates to the technical field of personnel examination scoring, in particular to an end-to-end-based multi-agent personnel examination scoring method, which comprises the steps of generating fusion features based on image data of an examinee answer sheet, inputting the fusion features into a heterogeneous OCR engine for processing, and outputting an examinee answer text; establishing a knowledge base, analyzing a question requirement text, and screening out a reference answer text from the knowledge base; constructing a plurality of dimension scoring agents, determining an original state based on the question requirement text and the reference answer text, training each dimension scoring agent through the original state, inputting the examinee answer text into the dimension scoring agents, outputting dimension scoring scores, and forming dimension scoring vectors; carrying out self-consistency analysis by integrating the dimension score vector and the original state, and outputting a final score; the whole-process closed-loop processing from paper answer sheet scanning to high-precision, automatic and interpretable scoring is realized, and the marking efficiency, fairness and credibility are improved.

The invention relates to an engineering class case retrieval method and equipment based on a large language model and machine learning. Thinking chains and few sample cue words are introduced into class case retrieval, a cue word template is constructed in combination with legal professional knowledge, a class case retrieval task is refined, a large model is guided to construct a multi-dimensional vector, and output of the large model is controlled by adopting a self-adaptive mechanism. The large model score vector is used as the input of the machine learning sorting model, so that the semantic understanding capability of the large model is combined with the machine learning sorting model, a clear case similarity basis can be provided for a user, the interpretability of a retrieval result is enhanced, and high expandability is achieved.

The invention provides an intelligent training system and method for a portable control console, and relates to the technical field of man-machine interaction and intelligent training, and the method comprises the steps of original trajectory data collection, operation feature extraction, capability score vector construction, user capability vector generation, capability trend analysis and personalized training path construction. The method comprises the following steps: acquiring attitude angular velocity, displacement acceleration and key data of a user in a training process through a portable console, extracting multi-dimensional operation characteristics, performing alignment and evaluation in combination with a scoring rule and a dynamic time warping method, and generating ability scores and user ability vectors; and further analyzing a capability development trend, identifying a capability short board, and matching a training task to generate a personalized path. The technical problems that an existing training system does not have portability, data acquisition dimensions in the training process are insufficient, capability evaluation is rough, and trend analysis and personalized training path recommendation are lacked are solved.

The invention discloses an intelligent education robot question answering system based on voice recognition and a knowledge graph, and particularly relates to the technical field of artificial intelligence. The method comprises the following steps: performing feature extraction on a user voice signal to obtain a text sequence and a multi-modal context feature; recognizing subject domain judgment and question answering intentions based on supervised classification and keyword rule fusion, and outputting a subject domain prior probability and a question answering target vector; determining polysemy words in the text sequence by using a context window, generating a semantic item sequence and semantic item confidence, constructing a subject domain sub-graph based on a subject domain prior probability, and obtaining a candidate reasoning path and a path scoring vector; determining a target teaching concept and an optimal reasoning path by combining Bayesian inference and consistency verification; generating a personalized question answering result aiming at the question asked by the user through fact retrieval and knowledge derivation in combination with the question answering target vector; accurate and efficient intelligent teaching question answering can be realized, and question answering accuracy and intelligent interaction capability of the education robot are effectively improved.

The invention discloses a fault fusion diagnosis method of a distributed photovoltaic system and a related device, and belongs to the technical field of fault diagnosis, and the method comprises the steps: collecting sensor time sequence data and an infrared image of the photovoltaic system, and carrying out the preprocessing; extracting an original time sequence feature and a frequency domain feature of the sensor time sequence data, inputting the original time sequence feature and the frequency domain feature into a pre-trained SVM model to obtain a second probability vector, and inputting the infrared image into a pre-trained CNN model to obtain a first probability vector; performing weighted fusion on the first probability vector and the second probability vector to obtain a weighted score vector; and based on a preset threshold, performing confidence calculation on the weighted score vector, and after a confidence condition is satisfied, obtaining a fault category, and completing fault fusion diagnosis. According to the invention, the problem of misjudgment or missed judgment easily caused by direct output of a single diagnosis result during fault diagnosis in the prior art can be solved.

The invention relates to the technical field of airport operation management, in particular to a data-driven airport scene sliding path intelligent generation method, which comprises the following steps: acquiring a plurality of taxiway sequences to form a taxiway sequence set; establishing a neural network model and a target function, and training the neural network model to obtain a taxiway embedded vector generator; starting and ending point pair coding, environment feature coding and related flight coding are carried out based on the taxiway embedded vector generator, and heterogeneous fusion features are obtained; carrying out autoregression generation based on Transform to obtain a plurality of candidate taxiways and corresponding score vectors; the multiple candidate taxiways are expanded and screened according to the score vectors of all the candidate taxiways, multiple candidate paths are generated, and the high-quality airport scene sliding path can be generated.

The invention discloses a brain-controlled rehabilitation equipment-oriented control strategy generation method, device and equipment and a storage medium, and relates to the technical field of signal processing, and the method comprises the following steps: acquiring a multi-channel electroencephalogram signal, and preprocessing the multi-channel electroencephalogram signal to obtain a target electroencephalogram signal comprising a steady-state visual evoked potential signal and a motor imagery signal, the target electroencephalogram signal corresponds to a preset action category; decoding the steady-state visual evoked potential signal by adopting filter group task related component analysis to obtain a correlation score vector; decoding the motor imagery signal by adopting a Mangban dynamic routing space-time network model to obtain a classification score vector; performing posterior probability distribution conversion and weighted fusion on the correlation score vector and the classification score vector to obtain fusion probability distribution; and generating a target control strategy according to the action category corresponding to the highest probability value in the fusion probability distribution. The control strategy obtained by the invention can consider both intention recognition precision and rehabilitation nerve activation effect.

The invention discloses a mental disorder intelligent discrimination system based on a brain structure image similarity map, and belongs to the technical field of artificial intelligence medical image analysis. The system firstly collects brain structure magnetic resonance image data of a multi-center mental disorder patient and a healthy control and carries out standardization preprocessing; then constructing a brain structure standardized distribution model of the cross-age gender, and extracting individualized deviation degree features; establishing a typical brain structure characteristic spectrum database of various mental disorders through a neural network; after to-be-diagnosed individual features are vectorized, multi-dimensional parallel comparison calculation is carried out on the to-be-diagnosed individual features and the atlas database by using a special similarity quantization algorithm; and the system outputs a quantitative similarity score vector containing the matching degree with various mental disorders and health modes, and generates a structured differential diagnosis report. According to the invention, diagnosis normal form transformation from absolute classification to flexible matching is realized, the problem of identification of mental disorder heterogeneity and common diseases is effectively solved, and the interpretability and clinical credibility of an intelligent diagnosis system are remarkably improved.

The invention provides an automatic driving system reliability self-diagnosis method and platform based on multi-modal sensor fusion. The method comprises the following steps: acquiring image modal data, performing feature extraction by using a visual Transform model, generating an environment scene label, and representing the type of a current environment; carrying out statistics on key modal characteristics in actual automatic driving operation data to obtain a modal score vector corresponding to each environment scene label to reflect the prior credibility of each sensor modal in the environment, and calculating a modal trust weight through softmax to obtain a confidence weight vector; performing weighted fusion on the key modal features based on the confidence weight vector to obtain fusion features; generating a modal reconstruction feature according to the fusion feature, and generating a modal consistency error vector based on the modal reconstruction feature; and in combination with the confidence weight vector and the modal consistency error vector, determining an early warning signal of the current system for judging whether the system has a fusion degradation or modal failure risk.

The invention relates to the technical field of natural language processing, in particular to a retrieval system and method based on retrieval enhancement generation, and the method comprises the steps: obtaining a user query text through a user interaction interface, and carrying out the embedding processing of the user query text to obtain a query vector; inputting the query vector into a vector knowledge base, performing approximate nearest neighbor search on the query vector and a storage vector in the vector knowledge base, and inputting the storage vector meeting a preset semantic distance into a candidate text block sequence; analyzing the semantic association strength between each text block in the candidate text block sequence and the query vector; performing descending sorting on each text block in the candidate text block sequence according to the similarity score vector; and inputting the enhanced prompt text into the large language model for reasoning analysis to obtain a retrieval result. According to the method, the text semantic similarity and the context correlation can be considered at the same time in the retrieval process, and the accuracy of the retrieval result is improved through a dynamic weighting mechanism.

The invention provides a collaborative emergency processing method, system and equipment based on intelligent wearable equipment. The method comprises the steps that a user binds the wearable equipment through a VDTS platform; performing feature vectorization processing on a historical emergency event sample, constructing an event association map, and embedding the event association map into an application end of the wearable device; connecting a real-time detection data set input by the wearable device, judging whether the real-time detection data set triggers an emergency event conflict response rule or not, and if the emergency event conflict response rule is triggered, identifying a plurality of conflict emergency events in the real-time detection data set; performing main emergency event probability prediction on the plurality of conflict emergency events according to the event association map, and outputting a plurality of probability score vectors corresponding to the plurality of conflict emergency events; and outputting an emergency processing strategy of the first emergency event according to the plurality of probability score vectors corresponding to the plurality of conflict emergency events. The technical problem that in the prior art, an intelligent wearable device cannot timely and accurately perform emergency processing in a multi-event concurrent scene is solved.

The embodiment of the invention provides a model optimization method, electronic equipment and a storage medium. The method comprises the following steps: acquiring an initial quantization bit width and an initial pruning rate of each convolutional layer in a BEV perception model; on the basis of the initialized quantization bit width of each convolution layer, quantizing the weight of the convolution layer, and according to the initial pruning rate of each convolution layer and the importance score vectors of the multiple output channels, obtaining pruning masks of the output channels; based on the loss function, the quantized weight and the pruning mask training model, obtaining an initial deployment model; and carrying out online quantitative sensitivity evaluation training on the initial deployment model, and when an evaluation condition is met, taking the initial deployment model as a final deployment model. Therefore, the problems that the domain adaptation and generalization ability is limited, the detection precision is reduced, the domain adaptation and generalization ability is difficult to be efficiently utilized by NPU of edge chips such as horizon lines, model evolution in the training process cannot be responded, and error accumulation is serious are solved, and structured sparsity, dynamic adaptation and end-to-end collaborative optimization are achieved.

The invention discloses an urban road tunnel reconstruction and extension scheme comparison and selection method, which belongs to the field of tunnel engineering, and comprises the following steps: establishing a universal standardized comparison and selection condition parameter library, and quantizing and encoding parameters in the comparison and selection condition parameter library to construct an urban road tunnel reconstruction and extension scheme database matrix; constructing an expert authority coefficient matrix based on the expert basic information; training a random forest model by using the urban road tunnel reconstruction and extension scheme database matrix, and generating a preliminary scheme suggestion based on a proposed project demand; independently scoring the preliminary scheme suggestion and giving an independent total score to form an expert scoring tensor and an independent total score vector; weighting the expert score tensor based on the expert authority coefficient matrix, training a random forest model by using the weighted data and the independent total score vector, obtaining the comprehensive score of each scheme, and determining the optimal scheme; and adjusting the expert authority coefficient matrix according to the scoring result of the optimal scheme, and bringing the optimal scheme data into the scheme database matrix for continuous learning and optimization.

The invention relates to the technical field of computer vision and performance evaluation, and discloses a license plate recognition system performance evaluation method and device. The method comprises the steps that a scene image of a target scene is collected and preprocessed; inputting the preprocessed scene image into a scene element extraction model, and outputting an initial scene element score vector; performing anomaly detection and correction on the initial scene element score vector to obtain a standard scene element score vector; generating a disturbance score vector by applying predetermined disturbance to each dimension score of the standard scene element score vector; inputting the standard scene element score vector and the disturbance score vector into a regression prediction model to obtain a plurality of identification rate prediction values; and determining a predicted recognition rate of the external license plate recognition system in the target scene and a confidence interval of the predicted recognition rate according to the statistical distribution of the plurality of recognition rate predicted values. According to the method, reliable mapping from scene visual elements to an LPR system recognition rate is established, and uncertainty evaluation is carried out on a prediction result.

The invention relates to a model data processing method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: converting text sequence feature data into a query vector, a key vector and a value vector through a decoder-only converter model; then multi-head attention vectors are determined; performing non-parameter shielding processing on the multi-head attention vector to obtain an attention score vector shielded by a pseudo attention parameter; performing normalization processing on the attention score vector through a normalization exponential function to obtain a normalized attention score vector; performing shielding score replacement processing on the normalized attention score vector to obtain a target attention score vector; and determining model output data of the text sequence features according to the product of the target attention score vector and the value vector. According to the method, the problem of unproportionate attention distribution can be solved in a non-parameter shielding manner, and meanwhile, absolute position information is coded in a self-attention processing process.

The invention discloses a WMS adaptive putaway decision-making method and system, and a computer readable storage medium. The method comprises the following steps: receiving a racking request, and collecting dynamic data including an operation state and business pressure in real time; inputting the data and the basic strategy weight into a trained deep learning model, and outputting a normalized dynamic weight vector; calculating score vectors of each candidate storage location under different strategy targets, and carrying out weighted summation on the score vectors and the dynamic weights to obtain a comprehensive score; recommending the storage location with the highest score for racking; and continuously optimizing the model by using a deep reinforcement learning algorithm based on operation effect data generated after actual shelving. According to the method, a'perception-decision-optimization 'closed loop is constructed, so that the problems of static stiffness, incapability of responding to real-time change and difficulty in multi-target tradeoff of a traditional racking strategy are solved, dynamic self-adaption and autonomous persistent evolution of a racking decision are realized, and the intelligent level and the overall efficiency of warehousing operation are remarkably improved.

The invention relates to the technical field of employee evaluation, in particular to a high-risk industry multi-dimensional dynamic weight employee evaluation method, which comprises the following steps of: obtaining quantitative scores of three-level indexes of employees and normalizing the quantitative scores by constructing a multi-level evaluation index system; inputting the standardized score vector into a Transform-based weight learning model, dynamically learning an index weight through a self-attention mechanism, carrying out step-by-step aggregation through a multi-layer perceptron, and outputting a comprehensive score of the employee on each first-level index; inputting the comprehensive score and the historical background information into a pre-established large language model, and performing fusion analysis according to a structured template to generate a personalized evaluation text; and finally, based on the score, the text and the employee information, automatically synthesizing a visual comprehensive evaluation report containing the radar map, the information bar and the text. According to the invention, objectiveness, individuation and operability of employee evaluation are realized.

The invention provides a safety monitoring method and system for an underground pipeline of a municipal road, and the method comprises the steps: arranging multiple types of sensors along the underground pipeline, and collecting the original monitoring data, such as stress, vibration, gas concentration, soil humidity and displacement; the original data is sent to a preprocessing module, time synchronization, abnormity elimination, deletion completion and normalization are completed, and a standardized feature matrix is generated; indexes such as structure variability, trend change rate and sensitivity responsivity are calculated based on the feature matrix, and a health score vector is generated through a feature fusion mechanism; performing density clustering and boundary segmentation in combination with the spatio-temporal index of the score vector, and extracting a high-risk section which changes violently or has an abnormal trend; the section features are combined with historical data, risk levels are calculated through a rule judgment mechanism, and two-dimensional layers of corresponding positions are generated; and accessing the layer to an operation and maintenance platform, completing early warning linkage, instruction generation and log archiving, and synchronously writing into a security evaluation database.

The invention discloses an action sports scoring method and device based on big data analysis, and relates to the technical field of computer vision, and the method comprises the following steps: S1, constructing a human body posture tensor manifold; s2, generating aligned action track characteristics; s3, generating symmetrical positive definite manifold features; s4, generating deep manifold distribution parameters; s5, quantifying global distribution deviation characteristics; s6, calculating a residual vector based on the deep manifold distribution parameters and the standard action distribution parameters, inputting the residual vector into the improved AGCN model for processing, constructing an adaptive topology based on the residual vector, and performing aggregation analysis to obtain a joint-level physical angle error after multi-scale space-time convolution and manifold enhancement; and S7, outputting a comprehensive score vector. According to the method, the limitation that a traditional method only depends on local geometric features and ignores global topological structure constraints and statistical distribution priori is overcome, and an efficient solution is provided for intelligent scoring of sports actions.

The invention provides a dialogue method and system based on context awareness, electronic equipment and a storage medium, and the method comprises the steps: carrying out the vectorization of a current dialogue instruction, obtaining a current dialogue instruction vector, calculating the correlation between the current dialogue instruction and N historical dialogue instructions, and obtaining a plurality of first correlation; determining whether to call the context based on the plurality of first correlations; calculating the correlation between the current dialogue instruction and the N historical dialogue instructions to obtain a multi-dimensional correlation score vector, and determining a plurality of second correlations based on the multi-dimensional correlation score vector; for each second correlation, performing multi-dimensional feature extraction on the historical dialogue instruction based on the second correlation to obtain a multi-dimensional feature vector corresponding to the historical dialogue instruction; obtaining a corresponding target input vector based on the target input sub-vector corresponding to each dimension; and determining a target answer based on the target input vector and the current dialogue instruction vector. The accuracy of reply in the dialogue process can be improved.

The invention relates to the field of voice recognition, in particular to an industrial field high-frequency voice recognition method and a storage medium. The method comprises the following steps: performing short-time Fourier transform on an industrial field sound signal by using a double-branch window to obtain a double-branch spectrogram; performing channel stacking on the double-branch spectrogram to obtain a three-dimensional tensor; after the extracted features are classified and calculated, a classification head outputs score vectors of probability scores of three dimensions of target sound, environment sound and strong noise; softening the probability score by using a temperature coefficient, and inputting the softened probability score into a classification function to obtain probability distribution; and calculating an energy score, when the energy score is lower than a preset energy threshold value, calculating an attenuation coefficient to carry out scaling suppression on the probability distribution to obtain final probability distribution, judging whether the highest value in the final probability distribution is lower than a rejection threshold value or not, and obtaining an identification result. According to the method, the fault sound detection rate is greatly improved in industrial actual measurement, and the false alarm rate is remarkably reduced.

The invention belongs to the technical field of image recognition, and discloses a zero-sample industrial anomaly classification and segmentation system and method based on multi-source expert scoring, and the system comprises a multi-source feature extractor which is used for carrying out the feature extraction of a to-be-detected image through more than two pre-trained feature extraction models, obtaining more than two image source features; the expert scoring module is used for obtaining an abnormal scoring vector in each image source feature according to an expert delegate; and the cascading fusion module is used for carrying out cascading fusion processing on the abnormal score vector of each image source feature to obtain a fused abnormal score vector, and generating a final abnormal score graph according to the fused abnormal score vector, including abnormal classification scoring and abnormal segmentation output. According to the method, the difference of different domains is effectively relieved, the accuracy of system anomaly recognition is improved, the reasoning speed is ensured, the overall performance in anomaly detection is improved, and the method is expected to be widely used in industrial image detection.

The invention relates to a multi-causal model integrated abnormal root cause tracing method and device and computer equipment. The method is applied to industrial equipment, and comprises the following steps: acquiring multivariable time sequence data of the industrial equipment in an abnormal state; based on the multivariable time series data, training at least three causal inference models adopting different prediction functions; determining a unified root cause score vector based on the root cause score vector of each causal inference model after training; and determining a root dependent variable based on the unified root cause score vector. By adopting the method, the root cause identification efficiency and accuracy can be improved.

The invention discloses a time series data anomaly detection method and device, equipment and a storage medium. The method comprises the following steps: acquiring to-be-detected time series data, and inputting the to-be-detected time series data into a target anomaly detection model; performing frequency domain decomposition on the to-be-detected time sequence data through a frequency domain decomposition network in the target anomaly detection model to obtain a detail component and an approximate component; calculating a first attention score vector and a reconstruction time sequence feature vector based on a multi-modal attention mechanism through a feature coding network in the target anomaly detection model; and through a comparative learning network in the target anomaly detection model, calculating an anomaly score of the to-be-detected time series data at each time point, and determining an anomaly detection result of the to-be-detected time series data according to the anomaly score. By adopting a mode of combining frequency domain decomposition, a multi-modal attention mechanism and comparative learning, the robustness and precision of an anomaly detection model are enhanced from multiple angles, and the detection precision of different types of anomalies is improved.

The invention discloses a network security level protection management system and method based on risk guidance, and relates to the technical field of network security level protection management, and the system comprises a module 100 which carries out the multi-channel semantic coding of a level protection requirement and equipment evidence, and generates a vector set. The module 200 constructs a semantic interaction relationship, calculates a matching score and outputs a score vector in a standardized manner. And the module 300 fuses the score vector, the risk weight and the prediction grade label, and triggers a correction mechanism when the score is abnormal. According to the network security level protection management system based on risk guidance provided by the invention, parallel modeling is carried out on network security level protection requirements and evaluation evidence information by adopting a multi-channel semantic coding structure, and deep semantic features, statistical keyword features and structural grammar features are integrated; and the adaptability and distinction degree of semantic representation to different types of input texts are improved.

The invention provides a supplier credit assessment early warning method and system based on a time sequence neural network in the technical field of artificial intelligence and supplier management crossing. The method comprises the following steps: S1, acquiring supplier data through a crawler cluster; s2, inputting the data of each supplier into a dynamic alignment model constructed based on a knowledge graph, performing entity disambiguation and relationship alignment on the data of each supplier through the dynamic alignment model, and constructing a time sequence feature matrix; s3, reasoning the time sequence characteristic matrix based on a double-flow time sequence neural network, and generating a credit score vector; s4, reasoning the time sequence characteristic matrix and the credit score vector based on a risk conduction simulation model to obtain a supplier risk conduction path; and S5, generating an early warning notification based on the credit score vector and the supplier risk conduction path. The method has the advantages that the timeliness, the reliability and the comprehensiveness of supplier credit assessment early warning are greatly improved.

The invention discloses a supplier settlement risk management and control method based on AI big data, and the method comprises the steps: obtaining enterprise data, carrying out the preprocessing, obtaining a global feature matrix, constructing a global knowledge graph, and obtaining an optimized time sequence global graph; the method comprises the steps of aggregating node features and local model parameters based on local data and a time sequence global atlas, generating a multi-dimensional risk score vector and generating a personalized intervention strategy based on the multi-dimensional risk score vector, generating a supplier intervention strategy vector based on the intervention strategy, generating an ERP system instruction based on the supplier intervention strategy vector, and collecting instruction feedback data to update a risk score. And optimizing a global model after federal aggregation, and realizing risk visualization, automatic report generation and interactive query by combining a BERT model based on a multi-dimensional risk score vector, a supplier intervention strategy vector and instruction feedback data. According to the scheme, through dynamic modeling, accurate prediction, safe cooperation and efficient storage, an intelligent and reliable risk management tool is provided for enterprises, and the method has remarkable practicability and commercial value.

The invention provides a personnel portrait construction method and system based on a multi-dimensional talent evaluation model, and belongs to the technical field of human resource management and data processing, and the method comprises the steps: collecting the original data of talents, constructing a tree-level comprehensive quality index system, and outputting the index score vectors of the talents; performing dimension reduction processing on the original data to extract key features, calculating objective weights of the key features, and outputting weight vectors; inputting the key features and the weight vectors into a random forest machine learning model to calculate a preliminary evaluation score; on the basis of the key features, an analytic hierarchy process is adopted to calculate expert correction scores; integrating the two scores to calculate a comprehensive evaluation score, and constructing a personal portrait based on the index score vector, the comprehensive evaluation score and the key features; and updating the personal portrait, monitoring the index score vector or the comprehensive evaluation score, and generating growth early warning information for the talents meeting the early warning conditions. According to the method, multi-dimensional data are integrated, the weight is objectively calculated, and the portrait is dynamically updated.

An intent recognition method based on a large language model and a customer portrait classification model, the method comprising: obtaining a customer's current relevant order and background information according to customer ID information; a customer portrait classification model obtains a first score vector according to the relevant order and the background information; the first score vector is used to indicate a current customer portrait classification score; a fine-tuned large model performs semantic recognition on the context of an input customer conversation to obtain a second score vector; the second score vector is used to indicate a customer current emotional feedback score; the fine-tuned large model is obtained by fine-tuning a large language model; a comprehensive score vector is obtained by fusing the first score vector and the second score vector; the comprehensive score vector is used to indicate different tendency weights in different business scenarios; an inference large model performs inference according to the comprehensive score vector, and outputs a customer intent; a corresponding business logic is driven to be executed according to the customer intent, and the corresponding business logic comprises providing customized suggestions or triggering a specific service process.