46 results about "Explanation module" patented technology

The invention discloses a personal credit score real-time early warning system based on behavior chain mining, which relates to the technical field of financial data processing and comprises a data acquisition module, a behavior chain atlas module, a behavior mining module, a collaborative modeling module, a score fusion module, an early warning interpretation module and a safety optimization module. All the modules interact through an end-to-end data flow and a real-time message mechanism, and dynamic evaluation and intelligent early warning of personal credit risks are achieved. According to the method, diversified normal and high-risk virtual behavior chains are automatically simulated and generated through a generative AI technology, a federated learning mechanism is combined, multiple mechanisms are enabled to jointly confront novel fraud and complex risk behaviors, original data does not need to be transmitted during model training, privacy security and model generalization ability are greatly improved, unknown risk behaviors are virtualized through AI, and the method is high in practicability and easy to popularize. The capability of identifying unprecedented risks is improved, and the problems of data islands and privacy leakage are avoided through federal learning.

The invention relates to the technical field of intelligent futures decision, discloses an intelligent futures decision agent method based on NLP and multi-modal data fusion, and aims to solve the problems of representation decoupling, causal chain deficiency, decision response lag and the like caused by separation and shallow fusion of multi-modal data processing in the prior art. According to the method, a multi-modal data acquisition and preprocessing module, a dynamic heterogeneous causal atlas construction module, a causal conduction space-time diagram neural network reasoning module and a futures decision signal generation and interpretation module are adopted, so that a system for uniformly representing dynamic endogenesis of macroscopic events, industrial logic and microcosmic prices is constructed, and cross-modal is realized.

The invention discloses a dynamic trajectory prediction method and system based on a space-time causal atlas, and relates to the crossing field of artificial intelligence, geographic information systems and edge computing. According to the method, a space-time causal map including physical connection, function association and causal relationship is constructed, and road network topological data and a causal rule base are fused and dynamically updated; utilizing a multi-scale space-time sensing module to extract features from microcosmic, mesoscopic and macroscopic scales respectively; by means of a differentiable causal reasoning module, trajectory prediction is decomposed into a structural causal effect, an environment regulation effect and an individual heterogeneous effect, and robustness is enhanced in combination with anti-factual reasoning; and realizing cross-city knowledge migration and distributed training based on a federated migration learning framework. The system comprises a space-time causal map construction module, a multi-scale feature extraction module, a causal reasoning prediction module, a federal transfer learning module and a result interpretation module. According to the method, the problems of precision attenuation, insufficient interpretability and poor cross-domain adaptability of a traditional method are solved, the precision, interpretability and cross-domain adaptability of trajectory prediction are improved, and the method is suitable for scenes such as intelligent traffic and automatic driving.

The invention relates to an urban planning and design standard intelligent question-answering system based on a large language model and GraphRAG, and the system comprises a data collection and preprocessing module which is used for comprehensively collecting urban planning and design standard literatures, including national standard, industry standard, local standard and group standard multi-source documents, the standard knowledge extraction and structuring module is used for carrying out deep analysis on a planning standard text by utilizing a large language model; the graph construction and storage module constructs a city planning standard knowledge graph based on an extraction result, and the semantic retrieval and GraphRAG enhancement module constructs an enhanced question and answer mechanism based on GraphRAG in combination with the knowledge graph and text vector retrieval. The question and answer generation and result explanation module is used for inputting candidate sub-graph contents as context prompts into the large language model and generating accurate and structured answers, and the system interaction and visualization module is used for providing a visual man-machine interaction interface and supporting natural language questioning, result browsing and knowledge graph navigation.

The embodiments provide a system and method for interpreting predictions from machine learning systems using natural language. The system includes a prediction module and an explanation module. The prediction module includes a machine learning model to make predictions for quantities such as recommended coverage amounts for insurance policies. The explanation module includes an impact analyzer that calculates impact values, which represent the degree of influence that each attribute has on predicted values. The explanation module also includes a natural language processing system for transforming generating natural language explanations indicating how the user attributes have influenced the predicted value.

The invention discloses a personalized smart course recommendation system for oral medicine based on a knowledge graph, and belongs to the technical field of course recommendation, and the system comprises a collection module which collects and fuses dominant static features and recessive behavior features of a user to form preliminary capability evaluation data; the analysis module is used for matching the preliminary capability assessment data with a preset oral internal medicine knowledge graph to generate a knowledge state mapping graph; the generation module is used for generating an optimal learning path from the current knowledge state to the target knowledge state according to a preset teaching strategy and a learning target on the basis of the knowledge state mapping graph; and the recommendation explanation module is used for analyzing the optimal learning path, enabling a preset course unit to correspond to the knowledge node to be consolidated or the association relationship, generating natural language description and completing personalized course recommendation. According to the method, the dominant static characteristics and the recessive behavior characteristics of the user are collected and fused to form the preliminary capability evaluation data, so that the accuracy of the initial recommendation stage is remarkably improved.

The invention discloses a multi-omics data integration plant gene function inference system and method based on a large language model, and the system comprises a data obtaining and processing module which is used for collecting multi-omics data and converting the data into a unified data storage format; the storage retrieval module is constructed based on an enhanced retrieval generation framework and is used for organizing and querying the data through a unified index and presenting biological information of the data in an interpretable format; the analysis module is used for establishing a hierarchical evaluation framework and a double-layer verification framework so as to ensure the reliability of the data and solve the evidence conflict problem in multi-omics data integration; and the inference and explanation module is used for establishing a large language model guide framework and completing plant gene function inference according to a preset priority sequence. According to the method, multiple omics data can be comprehensively integrated for accurate gene function inference, and the reliability and stability of gene function inference are improved under the conditions of different species and limited data availability.

The invention provides a large-model false news detection method based on interpretable machine learning, and the core implementation process of the method is as follows: firstly, constructing an original data matrix and a tag vector for a to-be-detected data set, and generating a diversified enhanced sample associated with original data through a large language model; then extracting deep semantic feature representation of the false news to be recognized by using a large language model; in the feature fusion stage, surface features of an original text and deep semantic features extracted by a model form dual-channel input, and collaborative aggregation of the features and the original text is realized through supervised learning and fine-grained learning mechanisms of an ELVIN architecture; and finally, accurately positioning key text fragments and feature dimensions which influence a detection result by relying on an interpretable module built in the ELVIN, and generating an interpretation report with logic traceability. Compared with an existing medical data set feature dimension reduction method, the method has the advantages that classification accuracy and model transparency are remarkably improved through multi-level feature extraction of a large language model and an adversarial training mechanism in combination with an attribution analysis technology capable of explaining machine learning, and meanwhile, the method has robustness of resisting text stylized attacks.

The invention discloses a student academic early warning system and method based on multi-modal behavior analysis, and belongs to the technical field of artificial intelligence education, and the system comprises a non-intrusive classroom behavior collection terminal, a multi-modal student evaluation engine and a visual early warning application. The acquisition terminal locally processes acquired classroom behavior videos of students into time sequence behavior feature vectors, the evaluation engine receives the feature vectors and analyzes the feature vectors through a built-in academic risk prediction model, and the evaluation engine further comprises a large language model interpretation module. According to the system and the method, non-intrusive quantification is carried out on student behaviors through edge calculation of an acquisition terminal to protect privacy, deep analysis is carried out through a sequence model in an evaluation engine, operable teaching intervention suggestions are generated by utilizing a large language model interpretation module, and objectivity, real-time performance and operability of teaching evaluation are improved.

The invention discloses an electric charge bill intelligent interpretation and optimization system based on multi-source data fusion and rag, and belongs to the technical field of energy consumption optimization. Comprising a multi-source data access and analysis module, an electricity price policy and charging rule knowledge base construction module, a multi-source data fusion and feature modeling module, a rate optimization inference module, a bill attribution analysis module, an intelligent explanation module, a front-end interaction display module and the like. Structured bill data, load time sequence features and unstructured electricity price policy document information are uniformly included in the same feature space, an electricity price policy knowledge base is used as a constraint source, and an explicit mapping relation from bottom layer data to upper layer rules is established. The rate optimization result strictly follows the current charging regulations in form and can be semantically attributed to specific terms, and a finer and more personalized rate and power utilization strategy scheme can be provided on the premise of ensuring compliance.

The application discloses an intelligent abnormality detection system for evaluation site management, and relates to the technical field of evaluation management. The method comprises the following steps: a clock-in event generation module collects face feature information of an evaluation personnel and compares the face feature information with a pre-stored identity template to generate a clock-in event; a rhythm abnormality judgment module analyzes project rhythm parameters to generate an abnormality judgment threshold set; a behavior time sequence construction module aggregates multiple clock-in events of the same evaluation personnel under the same evaluation project to construct a behavior time sequence chain; an abnormality judgment module matches and analyzes the behavior time sequence chain with an abnormality detection rule to generate an abnormality judgment result; an abnormality cause explanation module extracts rule identification and time offset information triggering the abnormality judgment to generate abnormality cause explanation data; and an abnormality detection result module constructs an abnormality evidence chain to generate an abnormality detection result and stores the abnormality detection result for evaluation site management. The application significantly improves the intelligent level and data reliability of evaluation site management.

The invention relates to a micro-service resource prediction and regulation method and system based on time sequence diagram learning and reinforcement learning. The regulation and control method comprises the following steps: data acquisition and preprocessing; constructing a time sequence heterogeneous graph; learning and training the dynamic space-time diagram; thinking chain reasoning prediction; reinforcement learning regulation and control decision making: performing resource regulation and control decision making based on a prediction result; and executing and feeding back a result. The system comprises a micro-service time sequence heterogeneous graph construction module; a dynamic space-time diagram learning module; a thinking chain reasoning enhancement prediction module; a reinforcement learning regulation and control module; and an LLM enhanced decision interpretation module. Compared with the prior art, the method has the following beneficial effects that the prediction precision is remarkably improved, the complex nonlinear dependence between the micro-services is captured through dynamic space-time diagram learning, and compared with a traditional regression model, the prediction error is reduced by more than 40%; regulation and control perspectiveness is enhanced, a reinforcement learning strategy can trigger resource adjustment in advance when a flow rising trend is observed, and the system overload risk is reduced by 60%.

The invention relates to a method (100) for predicting at least one quality of service (QoS) metric, comprising the following steps, which are performed by an interpretation module (10): - Receiving (101) information from one or more recording devices (21), wherein the information received includes visual information and / or capture information relating to an environment (1), - Analyzing (102) the received information regarding the environment (1) based on an examination of the information to interpret the environment (1) and detected changes in the environment (1), - Determine (103) a prediction of at least one QoS metric based on the analysis (102), and - Initiate (104) a corrective action depending on the specific (103) prediction, if the specific (103) prediction includes a change to at least one QoS metric.

This invention provides an interpretability system for mortality prediction models, comprising a prediction model module, a global interpretation module, an individual interpretation module, and a visualization interface module. By introducing a dual interpretation mechanism of the global and individual interpretation modules, the SHAP analysis unit of the global interpretation module generates a global feature importance ranking and an individual feature contribution distribution visualization, demonstrating the importance ranking of multi-source features in predicting mortality risk, enabling users to establish a macro-level understanding of the key aspects of the mortality prediction model. The individual interpretation module generates personalized interpretation reports, indicating the quantitative contributions of key features and crucial characteristics that drive increases or decreases in mortality prediction values. Through this combination of macro-level understanding and micro-level attribution, the mortality prediction model's predictions are transformed from an incomprehensible "black box" into a verifiable chain of evidence, effectively solving the core pain point of traditional models being difficult to apply due to their lack of interpretability.

The invention discloses a multi-dimensional intention embedding interpretable judgment recommendation system based on a DIKWP model. The system is composed of a judgment inference engine, an intention embedding module, an interpretation generation module and an interactive display interface. The engine performs hierarchical reasoning according to data-information-knowledge-intelligence-intention, and generates referee suggestions in combination with fact elements, legal provisions and class case retrieval; the intention module acts value weights of fairness, efficiency, policies and the like on a target layer to regulate and control the cutting amount; and the interpretation module synchronously outputs a semantic reasoning path which comprises a fact basis, a legal basis, a class case reference and an intention description, and provides hierarchical presentation according to user types. The system improves judicial AI transparency and reviewable performance, and is suitable for court case handling assistance, judgment document interpretation, class case research and other scenes.

The application provides a driving risk prediction system and method based on a multi-modal large language model, and relates to the technical fields of intelligent transportation and artificial intelligence. The system comprises: a multi-source driving information acquisition module for acquiring high-quality following event samples and driving element information; a multi-source driving data set construction module for constructing a high-quality traffic accident and driving behavior data set suitable for fine-tuning training of a multi-modal large model; a large model fine-tuning training module for optimizing parameters of a multi-modal large language model; a driving feature semantic extraction module for extracting visual semantic features in multi-perspective driving videos; a time series risk prediction module for deeply fusing physical data of a vehicle-mounted sensor and visual semantic features to construct a heterogeneous vector, and realizing prediction of a future risk trend based on time series reasoning; and a risk grading and explanation module for establishing a risk quantification grading threshold, and realizing quantitative attribution results of the prediction results.

The application relates to the technical field of artificial intelligence, in particular to a quality-guided gradual fusion multi-modal rumor detection method, which comprises the following steps: firstly, extracting features of each mode; secondly, converting each mode into a scene graph based on an LLM and performing knowledge enhancement; thirdly, calculating the quality score of each mode according to the structure and entity semantics of the scene graph, and extracting scene graph features of each mode using an SGAT; on this basis, fusing the scene graph features and original features obtained in the first two steps, inputting the fused features and the quality score into a core mode-guided gradual fusion process, and calculating the dynamic quality score between each layer of the gradual fusion; secondly, fusing the final update result of the core mode and the final result of the secondary mode to obtain the final fusion feature of the event; finally, inputting the obtained final fusion feature into a classifier; meanwhile, using an explanation module to make certain explanation on the prediction made by the classifier in combination with part of the intermediate results in the quality score calculation.

The invention discloses a medical record file management and intelligent report interpretation system based on a multi-mode AI, and relates to the technical field of deep application and innovation of an artificial intelligence technology in the medical field, and the system comprises a medical record report uploading module, a medical record file management module, a report interpretation module, an auditing and modifying module and a digital duplicate video explanation module. The case file management module identifies and extracts contents in a medical record report based on optical character identification, large language model structuring and embedded model semantic alignment, and the identified and extracted contents are stored in a structured manner to obtain patient case file data; and the report interpretation module interprets the patient medical record file data to generate an interpretation report based on a private domain knowledge base and a generation framework generated by retrieval enhancement in combination with a pre-filtering and post-filtering dual guarantee mechanism. According to the method, the closed-loop workflow of first draft generation by the AI, examination and modification by the doctor and digital duplicate video explanation is adopted, so that the information reaching efficiency of special patient groups is greatly improved.

The method for providing an explanation for an artificial intelligence model based on the plug-and-play method according to the present invention includes a step in which a service provider server sets the type of artificial intelligence model; when an artificial intelligence model is input, a step in which a plug-and-play manager recognizes the artificial intelligence model and creates a list of usable explanation modules based on the recognition results; a step in which the service provider server provides key information for each explanation module to a user terminal; a step in which the user terminal selects one of the explanation modules and provides it to the service provider server; a step in which the plug-and-play manager derives the selected explanation module from each explanation module in the list; and a step in which the derived explanation module provides an explanation for the input artificial intelligence model to the user terminal.

The invention provides an interpretable knowledge mining system, method and device and a medium, and belongs to the technical field of interpretable artificial intelligence and intelligent data mining. The system comprises a data processing module, an algorithm and intelligent interpretation module, a visual analysis module and an operation management module. An explicit mathematical model which is simple in structure and easy to understand is generated through a multi-branch shallow symbol feature learning tree structure, a dynamic branch expansion strategy, LLM guide initialization and a staged grey wolf optimization selection strategy; and an LLM and retrieval enhancement generation (RAG) mechanism is introduced to perform automatic semantic analysis and intelligent interpretation on the generated model. According to the interpretable knowledge mining system, method and equipment and the medium, the technical threshold of interpretable modeling is effectively reduced, meanwhile, the prediction performance, the structure interpretability and the knowledge discovery efficiency of the model are improved, and the interpretable knowledge mining system, method and equipment and the medium are suitable for engineering optimization, scientific modeling and cross-domain data analysis.

The invention relates to the technical field of data analysis, in particular to a news propagation influence quantitative evaluation method and system based on data mining, and the system comprises a data collection and preprocessing module, a dynamic index management module, an intelligent modeling and analysis module, a causal inference and explanation module and a storage and retrieval module. In order to solve the problem that a traditional evaluation model is insufficient in scene adaptability due to the fact that a fixed weight and a linear attenuation hypothesis are adopted, a dynamic attenuation model and a reinforcement learning mechanism are introduced in the scheme, and chain reactions in the information spreading process are described through modeling of a self-excitation effect in a spreading event; on the basis, an evaluation index weight distribution mechanism is dynamically optimized in combination with an adaptive algorithm, so that the model can automatically adapt to rapid outbreak of sudden news and long-acting propagation characteristics of policy news, and a nonlinear rule presented in the propagation process is accurately captured; according to the method, the scene adaptability and timeliness of the evaluation result are effectively improved.

The application discloses a kind of data-driven model explanation and feature analysis system and method, including benchmark model module, calculation module, global explanation module, local explanation module, clustering analysis module, feature interaction effect analysis module and feature optimization module, method includes: step one, quantitative calculation;Step two, global explanation;Step three, local explanation;Step four, clustering analysis;Step five, feature interaction effect analysis;The application can clearly distinguish feature actual influence mode by calculating and analyzing SHAP value distribution, and eliminate the ambiguity of feature analysis;Individual level prediction decision traceability is realized relying on local explanation module, to make up for the limitation that prior art can only be globally explained;The positive and negative influence direction of feature and target variable is clear by SHAP value, and the real correlation law of feature and target variable is revealed;Based on SHAP interaction value, the interaction effect between features is quantified, and the coupling influence law of multiple features is revealed.

The embodiment of the application provides a picture dynamic display method and device, electronic equipment and storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: performing content identification on a picture to be displayed to obtain a plurality of picture display contents in the picture to be displayed; determining at least one explanation module from the picture to be displayed according to the plurality of picture display contents and preset picture explanation contents obtained; performing content splitting on the picture to be displayed according to the plurality of explanation modules to obtain a plurality of pictures to be explained; and sequentially displaying the pictures to be explained to form dynamic display of the picture to be displayed. The embodiment of the application can sequentially display the pictures to be explained to form dynamic display of the picture to be displayed, can reduce redundant visual interference of a speaker in an explanation process, can make a listener more focused on explanation contents that the speaker currently wants to emphasize, and is favorable for improving an explanation effect of the speaker.

The invention relates to the technical field of steam turbine generator unit fault diagnosis, and discloses a unit fault diagnosis model construction method and system, and the method comprises the steps: obtaining and structuring fault diagnosis data, and forming a basic database; the configuration is optimized to adapt to the requirements of different operation stages of the unit; constructing a multi-source feature input system, and obtaining multi-dimensional feature data; constructing a deep learning network model based on the data, and introducing constraint optimization training; designing double-path diagnosis trigger logic, and triggering model diagnosis according to feature matching; adjusting the weight by adopting a dynamic weight fusion mechanism in combination with historical data; and constructing an interpretation module to output a diagnosis result and a basis. The technical problems that in the prior art, knowledge boundaries are limited due to the fact that expert experience is relied on, non-structuralization is difficult to inherit, pure machine learning lacks guidance, data are insufficient or misjudgment is prone to occurring when distribution is uneven, and results do not have physical significance support are solved.

The invention provides a method and a system for building a scenic spot tourist flow prediction model based on artificial intelligence. The method mainly comprises the steps of data collection and processing, feature engineering and initial model training, model optimization and key factor screening, data feedback and continuous model improvement, prediction result display and explanation and the like. The system integrates structured data and unstructured data, and performs data training and model optimization by using a deep learning algorithm. In addition, the data display and interpretation module enhances the interpretability of the model, displays the data processing flow and the prediction result through a visual tool, and helps the user to understand the model logic. Compared with a traditional method, the dynamic change of the popularity of the scenic spot can be more accurately captured, the prediction precision and efficiency are improved, and the method is suitable for a rapidly changing tourism market.

The invention belongs to the field of path planning, and discloses an intelligent guide explanation system and method for a museum. Comprising a device issuing and position monitoring module, a tourist flow sensing module, a dynamic route optimization module and a voice interaction and explanation module, and is used for issuing to tourist receiving devices, monitoring the position of each receiving device in real time and generating position data; obtaining the number of tourists in each area, and generating a flow thermodynamic diagram in combination with the position data; generating an optimal touring route based on the flow thermodynamic diagram and tourist preferences, and dynamically adjusting the optimal touring route; cultural relic explanation is provided through the receiving device, voice conversation is supported, tourist demands are recognized, services are provided, and intelligent guide explanation for the museum is achieved.

The invention discloses an anti-interference remote team travel intelligent voice explanation system and a broadcast control method. The system comprises a PC terminal and an intelligent APP terminal. The PC end comprises a user management module and a marshalling management module; the intelligent APP terminal comprises a travel management module and a voice explanation module; the user management module is used for overall management of user information and adding travel planning users into the module, the marshalling management module is used for creating travel groups, and the travel management module is used for counting and displaying information of personnel joining the travel corresponding to the vehicle. The voice explanation module is used for automatically creating an exclusive voice room at the beginning of a travel, the system can automatically create the exclusive voice room according to travel participants, and the participants can join the room for real-time listening only by clicking the travel participants anytime and anywhere through the mobile terminal; the problems of unclear explanation sound, crosstalk and distance limitation when the number of teams is large and members stand dispersedly are effectively solved.

The invention provides an intelligent reservoir water flooded layer recovery system based on a spiking neural network, which belongs to the technical field of oil exploration and development and comprises a data acquisition module, a data preprocessing module, a model training module, a resistivity recovery module and a water flooded layer interpretation module. The data acquisition module is used for acquiring multi-modal data such as conventional and electrical logging curves; the data preprocessing module is used for carrying out range normalization and Z-Score standardization on the data; the model training module constructs a resistivity recovery model by using a pulse neural network, and the resistivity recovery model is optimized and verified in various modes; the resistivity recovery module recovers the original resistivity accordingly; and the water-flooded layer interpretation module determines and interprets the water-flooded layer in a grading manner based on the calculation related parameters. According to the intelligent recovery system for the oil reservoir water flooded layer based on the spiking neural network, the resistivity of the water flooded layer is recovered with high precision, the water flooding influence is accurately quantified, and a scientific and reliable basis is provided for oil reservoir development and adjustment.

The application discloses an artificial intelligence silent evolution control system based on a fusible, contractible, auditable and interpretable edge protocol and belongs to the technical field of artificial intelligence system safety and reliability management. The system comprises an edge protocol bus, an interpretable module, a fusible module, a contractible module, an auditable module and a silent evolution management layer. The interpretable module calculates the confidence and the feature attribution path of the model output. The fusible module triggers the fuse according to the preset confidence condition and publishes an event. The contractible module executes dynamic resource contraction in response to the fuse event. The auditable module generates a three-tuple association log of the confidence, the fuse action and the contraction action. The edge protocol bus provides standardized event communication, module registration and version negotiation. The application solves the technical problems that the existing AI management module is isolated and cannot actively defend and heat evolve based on the output quality, and is suitable for high-risk AI application scenarios such as finance, medical treatment and automatic driving.