209results about How to "Interpretable" patented technology

The invention discloses an anti-fraud modeling modeling method and an anti-fraud monitoring method based on machine learning. The anti-fraud model modeling method based on machine learning comprises the following steps: extracting sample data required for modeling from a database, and carrying out labeling processing on each sample data; matching the association information of each sample data from the database, using the results of labeling processing to establish the multi-dimensional credit data based on the user, and processing and dividing the credit data into training set data and test set data; training and adjusting the parameters of the anti-fraud model by using the training set data; using the test set data to test the anti-fraud model, obtaining the fraud probability value thatthe test set data is fraudulent users. The obtained fraud probability value is compared with the corresponding actual sample situation, and the stability of the anti-fraud model is judged according tothe comparison result, and the anti-fraud statistical threshold value is established. The method can effectively reduce the risk of fraud through label processing and supervised machine learning.

The invention relates to a user position predicating system based on a wireless network. The user position predicating system comprises a mobile data obtaining module, a mobile data cleaning module, a user history mobile record module and a mixed position predicating module, wherein the mixed position predicating module is provided with a stop assumption predicating sub module, a history predicating sub module, a jump predicating sub module, a collaborative filtering predicating sub module and a predicating result sub module, the history predicating sub module is based on time period division, and the predicating result sub module completes the user position predication according to predicating results of the stop assumption predicating sub module, the history predicating sub module based on time period division, the jump predicating sub module and the collaborative filtering predicating sub module. The invention also provides a user position predicating method based on the wireless network. The system and the method provided by the invention have the advantages that the real-time position predication in 24 hours is provided, and in addition, the predication result accuracy is high.

The invention provides a pulmonary nodule benign and malignant prediction method and device, and the method comprises the steps: obtaining a chest flat-scanning thin-layer CT image, carrying out region-of-interest delineation of pulmonary nodules in the CT image layer by layer to acquire the clinical information and pathological information of a patient; extracting an image omics feature of the pulmonary nodule in the region-of-interest based on a PyRadio toolkit; screening the image omics features by using a plurality of feature selection algorithms; training a deep convolutional neural network model by using the CT image to acquire deep learning features, forming a multi-dimensional clinical feature vector in combination with clinical information of the patient, and splicing the deep learning features, the clinical features and the imaging omics features to obtain a multi-modal feature vector; and establishing a pulmonary nodule benign and malignant prediction model by using variousclassifier algorithms based on the multi-modal feature vector, and analyzing a prediction result by using the pathological information of the patient to obtain an optimal pulmonary nodule benign and malignant prediction model to perform benign and malignant prediction on the pulmonary nodule.

The invention discloses a student modeling and personalized course recommendation method in an online learning system, and belongs to the education data mining field. According to the method, cognitive level modeling and personalized course recommendation of students are mainly studied, firstly, knowledge mastering states of the students are judged based on a cognitive diagnosis model, learning behaviors of the students are analyzed through data on a system platform, and then the cognitive abilities of the students are modeled by integrating course mastering conditions; secondly, an online course is modeled ; and finally, the features of the online course are fused according to the cognitive level of the student to perform personalized recommendation. According to the invention, personalized recommendation is carried out based on the cognitive level of students and in combination with the feature indexes of the online courses, so that the user can be helped to carry out more accurate personalized course recommendation, and the online course recommendation is more interpretable and acceptable.

The invention relates to the technical field of passenger flow prediction, and discloses a rail transit space-time short-time passenger flow prediction method, device and equipment and a storage medium. The method comprises the steps of acquiring pull-in data and train timetable data of a historical time period, constructing an adjacency matrix according to the train timetable data; standardizingthe pull-in data and the adjacency matrix; adopting a graph convolutional neural network to extract spatial feature matrixes of the standardized pull-in data and the adjacency matrix; and extracting time features of the spatial feature matrix by adopting a sequence-to-sequence model based on a gating cycle unit and an attention mechanism so as to predict an outbound amount at the current moment. According to the method, the space-time relationship of large-scale passenger flow can be captured, high precision and high interpretability are achieved, the passenger flow distribution situation canbe mastered conveniently, and a basis is provided for passenger flow state analysis and early warning. Meanwhile, passenger flow organization is facilitated, transport capacity resources are reasonably allocated, congestion is relieved, and the service quality is improved.

The invention discloses a personal-credit evaluation method of intelligent combination and a system. The method is suitable for use in the system. The system includes a sample data acquisition module,a sample data processing module, a data feature training module, a scoring model construction module and a scoring model test module. The method includes the steps of: S1, acquiring sample data of model training by the sample data acquisition module; S2, dividing samples into a training set I, a training set II and a test set; S3, carrying out feature grouping on the sample data of the training set I, the training set II and the test set; S4, carrying out training on each set of data features, and generating a corresponding trained sub-model by each set of data features; S5, carrying out prediction on the trained sub-models; and S6, testing a finally constructed credit scoring model on the test set by the scoring model test module to evaluate model test effect.

The invention provides a big data access authorization method and device and a big data platform. The method comprises the following steps: extracting behavior information of an authenticated user ofthe big data platform; evaluating the behavior information of the user according to the service scene, marking user tags of multiple dimensions a for the user to construct a user portrait, and acquiring the tag weight corresponding to each user tag by fusing the subjective importance degree and the objective importance degree of the tags; updating the user portrait at preset time intervals to realize dynamic updating of the user tag and the tag weight thereof; and performing access authorization on the user according to the updated user portrait. According to the invention, personalized authorization of different users can be realized, and the constantly changing personalized access requirements of different subjects are met. Dependence on professional knowledge of an administrator can bereduced, and the administrator can actively authorize a user in advance conveniently. The user rsuspicion and the service demand can be accurately predicted and the data security of the big data platform can be effectively improved.

Disclosed is a wireless-network-based user position predicting system. The wireless-network-based user position predicting system comprises a movement data obtaining module, a movement data cleaning module, a user history movement recording module and a mixed position predicting module. The mixed position predicting module is provided with a staying assumption predicting submodule, a history predicting submodule based on time division, a skip predicting submodule, a collaborative filtering predicting submodule and a prediction result submodule wherein the prediction result submodule completes user position prediction according to the prediction results of the staying assumption predicting submodule, the history predicting submodule based on time division, the skip predicting submodule and the collaborative filtering predicting submodule. The invention also discloses a wireless-network-based user position predicting method. The wireless-network-based user position predicting system and the wireless-network-based user position predicting system have the advantages of providing 24-hour real-time position prediction and being highly accurate in predicting results.

The invention provides a dialogue generation model determination method and device, a storage medium and an electronic device, and relates to the technical field of artificial intelligence. The dialogue generation model determination method comprises the following steps: obtaining a plurality of groups of training samples, wherein each group of training samples comprises first input information and target reply information; based on the first input information and the target reply information, determining a first target word according to a word library, and determining a first hidden variable according to the first target word; training the dialogue generation model based on the first hidden variable; the lexicon is a word set of the dialogue text. According to the technical scheme provided by the invention, the diversity of replies generated by the dialogue generation model is improved, so that questions proposed by a user can be accurately and effectively answered, and the dialogue experience of the user is improved.

The invention discloses an interpretability recommendation method based on a knowledge graph path. The method comprises the steps of obtaining the interaction history of a user, taking the interactionhistory as a seed set of a knowledge graph, and obtaining a user-from the seed set; on the premise that a seed set is obtained, constructing a triad query corresponding to a knowledge graph on the seed set, extracting triads, generating a path representation through semantics of combined entities and relations in triad information, and conducting the reasoning process according to paths to deduceuser preferences; after the determination of a triple path, querying other paths from the head entity to the tail entity of the path on the premise of limiting the length of the path to be 4, and representing the paths by using a plurality of triples; after the finding of multiple paths, performing pool operation on each path to distinguish contributions of different paths to prediction recommendation; and selecting the path with the maximum contribution score to carry out interpretive recommendation on the user. The method is high in recommendation precision, and the problem of opaqueness ofrecommendation is solved.

The invention relates to a recommendation prediction method based on attribute information reference self-leaning. Attribution information is utilized for slowing down cold start, the advantages of user and commodity collaborative filtering and model decomposition based on a matrix are combined, the training speed is high, and the interpretability is achieved; meanwhile, on the condition that rating data is sparse, the prediction precision is superior to that of user and commodity collaborative filtering and the matrix-based decomposition method.

The invention provides a ship target detection method based on joint training of deep learning features and visual features, which includes the following steps: sample data collection, CNN feature extraction, traditional moment invariant feature and LOMO feature extraction, feature dimension reduction, feature fusion network FCNN construction, and training the network with sample data and testingthe model with test data. Compared with the prior art, the visual feature extraction process of the invention comprehensively considers the characteristics of the ship shape, color and texture, so that the detection process is interpretable, and other features other than the traditional features can be learned in the normalized CNN back propagation process. This method is fast and efficient, highaccuracy, for complex scenes such as clouds and fog, cloudy days, rain and other circumstances still have good detection results, high robustness. Features complementary to the traditional features can be extracted, and the speed is very fast, can achieve the effect of real-time monitoring.

The invention provides a method and a system for forecasting the top oil temperature of a transformer. The method comprises the following steps: based on historical meteorological data, historical oiltemperature data and historical load data, a linear regression analysis model of the top oil temperature of the transformer is established; based on historical load data, a load forecasting model with meteorology and date attributes is established; based on the load forecasting model and meteorological forecasting information, the load forecasting value of the transformer in future time length isdetermined; and based on the linear regression analysis model and the load forecasting value of the transformer in future time length, the maximum oil temperature of the transformer in future time length is forecasted. The top oil temperature forecasting model of the transformer constructed by the invention realizes short-term oil temperature early warning analysis and provides a reference for transformer condition maintenance and accident prevention.

The invention relates to a spectral unmixing method based on core prototype sample analysis. The method includes the steps of collecting hyperspectral data to be processed, determining the parameters of the whole process, preprocessing input image data and achieving spectral unmixing on the preprocessed data through a core prototype sample analysis method. The spectral unmixing method is easy to implement, the spectral unmixing process does not need to be independently factorized into an end member extraction process and an unmixing process, the unmixing problem of the inexistence of pure end members can be solved, and the optimal end member selecting and unmixing problem of data at different mixing degrees can be solved as well. In addition, the physical meanings of an ultimate extraction result are definite, and data deciphering capacity is higher. Meanwhile, the result obtained in the method is more stable compared with a non-negative matrix factorization spectral unmixing result, and the precision is better.

The invention provides a risk conduction evaluation system and method for public credit customer risk early warning. The risk conduction assessment system comprises the following system and steps of (1) risk conduction unit establishment, (2) label establishment, (3) feature establishment, (4) model training and application, (5) model training, (6) model application and risk conduction range estimation and (7) overall modeling process. The risk evaluation system and method has the advantages that the influence range of risk enterprises can be quantitatively assessed, and due to the fact that the risk assessment method is essentially a supervised machine learning process, compared with an existing risk conduction assessment scheme, the risk conduction assessment system and method has the characteristics that self-learning and adjustment can be conducted on historical data, and subjectivity of expert experience is avoided; B, the conduction probability of risk conduction can be evaluatedquantitatively based on the historical data; and c, the model is objective and has certain interpretability.

The invention discloses a method for positioning performance problems of a large-scale parallel program, belonging to the technical field of parallel program performance analysis. The method comprises the following steps of: grouping functions in a parallel program; establishing a cause-effect relationship diagram; performing clustering computation on function performance data of profile logs and track logs in the parallel program; selecting function groups with large overhead according to the clustering computation results and inputting the function groups into a sorter; and performing performance problem positioning according to the mapping from the function groups to performance problems. With the method disclosed in the invention, the performance problems of the parallel program can be accurately positioned and results can be interpretable.

The invention discloses a behavior simulation training method for an air intelligent game. The method comprises the following steps: S1, constructing an intelligent agent game decision model; S2, determining an environment state and an action space, and shaping a continuous non-sparse reward function of each action; S3, carrying out an air game in the model, and executing the following steps: S31, generating a next environment state according to an executed action, obtaining a reward, and carrying out loop iteration in sequence to realize maximum accumulated reward; S32, realizing reverse reinforcement learning based on expert behaviors, and obtaining a target reward function; S33, calculating the similarity between each agent behavior and the expert behavior; S34, obtaining a comprehensive reward; and S4, training the agent game decision model. According to the method, a traditional low-efficiency reward function design process and a model training random exploration process are improved, so that the reward function has interpretability and human intervention ability, the agent decision level and convergence speed are improved, and the cold start problem of model training is solved.

The invention provides an automatic question and answer method and device; the method has the following steps: firstly, obtaining user question statements and a candidate question set; training an attention neural network through a question statement training sample to obtain a question model; determining question statements matched with the user question statements from the candidate question set, and finally, adopting the answer information corresponding to the target question statement to respond to the user question statement, and utilizing the characteristics of the attention neural network, so that the automatic question-answering process and result realized by utilizing the question model are more interpretable, and the automatic question-answering effect is improved.

The invention discloses an abnormal brain connection prediction system, method and device and a readable storage medium, and the method comprises the steps: automatically extracting high-order correlation features in different modes and high-order complementary features between different modes through a deep learning method; and realizing the analysis of abnormal connection of the multi-modal brain network and prediction of different cognitive diseases through an adversarial training method. The method solves the problem that an existing method cannot accurately evaluate the change rule of brain structural morphology and functional connection. According to the method, a prior knowledge guide model is used for learning interpretable characterization, the consistency of different modal characterization distribution is restrained through a paired collaborative discriminator, and then brain graph data is reconstructed for feature codes through a reverse generator and a decoder; and finally, inter-modal and intra-modal high-order correlation features are extracted through a hypergraph perception fusion module, and an adversarial loss function, a reconstruction loss function and a classification loss function are set to guide model learning so as to achieve the purpose of mining the abnormal brain connectivity of the Alzheimer's disease.

The invention discloses a context awareness-based fine-grained emotion classification method for a hybrid neural network, and the method comprises the steps of introducing a context vector into a word-level attention mechanism through employing a bidirectional Bi-LSTM during the coding of a word in each clause; on the clause level, using a convolution layer to extract local features from clauses,gathering all the local features through maximum pool operation, and obtaining a sentence vector with a fixed size; and inputting the sentence vector into a softmax classifier for classification to obtain a label with the highest probability to represent the predicted aspect sentiment polarity. According to the fine-grained emotion classification method of the hybrid neural network based on context awareness, context vectors are introduced into a word-level attention mechanism, so that the representation of each obtained clause vector fully considers context information. According to the invention, the convolutional neural network is used at the clause level to achieve the same function, but the calculation cost is greatly reduced.