1013 results about "Interpretability" patented technology

The invention relates to an industry comment data fine grain sentiment analysis method. The industry comment data fine grain sentiment analysis method is applied to Internet data analysis and comprises obtaining comment data of e-commerce industry goods and preprocessing the comment data; establishing initial industry sentiment word libraries and computing distribution of words under different sentiment polarities through 1-gram and 2-gram; performing Chinese word segmentation on the comment data; based on the sentiment word libraries established through the 1-gram and the 2-gram, utilizing combined sentiment models to perform word modeling to obtain the probability distribution of the words which belong to different topics under different sentiment distributions; utilizing context information to re-determine the sentiment alignment of sentiment words in sentences; performing named entity identification and extracting comment characteristics through conditional random fields to compute the sentiment alignment of comment words of the comment characteristics. The industry comment data fine grain sentiment analysis method computes the sentiment of the comment words through the two dimensions of topic and sentiment to achieve fine grain sentiment analysis on the industry comment data, thereby achieving high precision and interpretability of analysis results.

The invention discloses a sparse dimension reduction-based spectral hash indexing method, which comprises the following steps: 1) extracting image low-level features of an original image by using an SIFT method; 2) clustering the image low-level features by using a K-means method, and using each cluster center as a sight word; 3) reducing the dimensions of the vectors the sight words by using a sparse component analysis method directly and making the vectors sparse; 4) resolving an Euclidean-to-Hamming space mapping function by using the characteristic equation and characteristic roots of a weighted Laplace-Beltrami operator so as to obtain a low-dimension Hamming space vector; and 5) for an image to be searched, the Hamming distance between the image to be searched and the original image in the low-dimensional Hamming space and using the Hamming distance as the image similarity computation result. In the invention, the sparse dimension reduction mode instead of a spectral has principle component analysis dimension reduction mode is adopted, so the interpretability of the result is improved; and the searching problem of the Euclidean space is mapped into the Hamming space, and the search efficiency is improved.

The present invention provides a method for analyzing emotional polarity of heterogeneous migration images based on multimodal depth potential correlation. The method comprises the following steps: 1)constructing an initial emotional image dataset, and taking the emotional polarity corresponding to the emotional words as the emotional polarity tag; 2) removing the noise data in the initial emotion image data set, and removing the noise by using the method of emotional consistency and the probabilistic sampling model based on the multimodal deep convolution neural network; 3) constructing theheterogeneous migration model based on the multimodal depth potential correlation, and then training the source domain text and the target domain image; 4) constructing the multimodal embedded space,embedding semantic information of the source domain text into the target domain image; and 5) training the image emotional polarity classifier for the image emotional polarity analysis. According to the method provided by the present invention, the obtained data is large in scale, the labor cost is low, the data noise is small, the prediction accuracy is high, the model is strong in interpretability and has strong classification capability, and a better image emotional polarity analysis effect can be reached.

The invention discloses an implementation method of a fusion network question and answer system based on a multi-attention mechanism, which comprises the following steps of constructing a question andanswer system network model, preprocessing an original data set to obtain a standby data set, and performing text length distribution analysis; subjecting text in standby data set to one-hot vector representation, using a CBOW model to train one-hot word vector and forming a word2vec word list; adjusting the sequence length of each sentence in the text, and adding a sentence end mark; training the word2vec vector by using an ELMO language model to obtain an ELMO word vector; encoding the ELMO vector to obtain a sentence vector; performing coarse-fine granularity attention on the sentence vectors respectively to obtain memory vectors and attention vectors based on each word; carrying out vector splicing to obtain expression vectors based on words and sentences; and decoding an answer representing the vector generation question sentence. According to the method, the representation ability of sentences is improved through an ELMO language model; and various attention mechanisms are fused, so that the decision making accuracy of the system is improved, and the interpretability of the system is enhanced.

The invention belongs to the field of computer vision and machine learning, and particularly relates to a cross-visual angle gait recognition method based on a multitask generation confrontation network. The objective of the invention is to solve a problem of reduced generalization performance of a model under big visual angle change of the gait recognition. The method comprises steps of firstly carrying out pretreatment on each frame of image for an original pedestrian video frame sequence, and extracting gait template features; carrying out gait hidden expression through neural network coding and carrying out angle transformation in a hidden space; generating a confrontation network through multiple tasks and constructing gait template features of other visual angles; and finally using the gait hidden expression to carry out recognition. Compared with methods based on classification or reconstruction, the method has quite strong interpretability and recognition performance can be improved.

The invention relates to the technical field of machine learning, and especially relate to a method and a system for integrated machine learning convenient for data analysis personnel to use. The method comprises the following steps: (1) data exploring; (2) data cleaning; (3) feature extraction; (4) feature selection; (5) sampling; (6) model training; (7) model optimization; (8) model combination;(9) model interpretability; (10) nature language processing. The system comprises a data processing module, a feature processing module, a model processing module, and a nature language processing module. The method and the system provide a unified algorithm modeling process for machine learning engineers, students, teachers, and machine learning fans, so that the machine learning engineers, students, teachers, and machine learning fans complete a modeling process by 20% efforts, and concentrate 80% efforts on understanding of business and model application, to deeply understand business andpreferably realize requirements of business personnel on models.

Data analysis methods are characterized in terms of tangible and non-tangible parameters such as user-friendliness, accuracy, simplicity, interpretability, speed, cost, comprehensibility, and transparency, and then matched to pre-specified requirements. A data analysis method is selected in accordance with a user preference, wherein the user preference relates to a feature of the data analysis method and is represented by a fuzzy set comprising a range of values.

The invention discloses a credit assessment model training method, and a credit assessment method and apparatus. The training method comprises the steps of obtaining original training behavior data of a training user in a business system; extracting original training features in the original training behavior data; performing feature combination on the original training features according to a GBDT model to generate corresponding training cross combination features; and performing training on a logistic regression model according to the training cross combination features to build a credit assessment model. According to the method, the original training features are trained through the nonlinear GBDT model to generate the corresponding training cross combination features, and the linear LR model is trained through the training cross combination features to build the credit assessment model, so that the credit assessment model has not only high performance of the nonlinear model but also interpretability of the linear model.

The invention discloses a multi-strategy fusion knowledge question and answer method and system, and the method comprises an offline part and an online part, the offline part is mainly used for data preparation and model training, and the online part is mainly used for system service, and the method comprises the steps: receiving a statement input by a user, and correcting a spelling error; Performing word segmentation and part-of-speech tagging on the statement input by the user; Extracting entity information in the user statement, and linking the entity to the knowledge graph node; Obtainingan executable query statement through a multi-strategy fusion semantic analysis step according to the result of the entity recognition and connection process; Executing a query on a knowledge graph by the executable query statement to obtain an answer, and then generating a corresponding natural language reply user according to the answer through a reply generation mode combined by multiple methods, so that the question and answer system is suitable for question query of general and domain knowledge graphs, the system robustness is improved, and meanwhile, the good interpretability and controllability are achieved.

The invention discloses an activity similarity and social trust based social networking website friend recommendation system and method and belongs to the field of information recommendation and data mining. According to the activity similarity and social trust based social networking website friend recommendation method, position-based friend recommendation in the social networking website is achieved mainly according to user social contact trust values and activity preference similarities; friends with preferences similar to users can be discovered through the activity preference between the users due to the fact that user interest preferences can be showed in the activities; friend recommendation according to trust relationships is more reasonable in interpretability. Experiments show that, a recommendation effect of the activity similarity and social trust based social networking website friend recommendation method is superior to the existing friend recommendation method in accuracy and reasonable interpretability and high in practical application values, the activity similarity and social trust based social networking website friend recommendation method has great importance in guidance and decision making for enterprises and public institutions to confirm client aims, improve the advertising service relevancy and accuracy and improve the advertising values.

The invention discloses an intelligent commodity recommending method based on word vector data driving. The method includes the following steps: data pre-processing, word vector generation, score prediction constructing, model training and score prediction. The method includes the following steps: using a word vector method to take user numbers, commodity numbers and commodity scores as semantic words, changing the semantic words to a sparse vector by conducting one-hot encoding, then multiplying the sparse vector and a weight matrix and mapping a high-dimension and sparse original vector to a dense, continuous and fixed and low-dimension feature space, then inputting the original vector to a deep model for carrying out training to obtain weight parameters of each layer of a model, predicting and scoring a new user's favor to the commodity by using a well-used model so as to complete intelligent recommendation of the commodity to the user. According to the invention, the method applies the word vector method which classifies texts to score prediction and commodity recommendation which is based on favor of the user of an e-commerce platform to the commodity. The method can ensures precision and provide better explanation.

A method is provided for data classification that achieves improved interpretability and accuracy while preserving the efficiency and scalability of univariate decision trees. To build a compact decision tree, the method searches for clusters in subspaces to enable multivariate splitting based on weighted distances to such a cluster. To classify an instance more accurately, the method performs a nearest neighbor (NN) search among the potential nearest leaf nodes of the instance. The similarity measure used in the NN search is based on Euclidean distances defined in different subspaces for different leaf nodes. Since instances are scored by their similarity to a certain class, this approach provides an effective means for target selection that is not supported well by conventional decision trees.

The invention discloses an item recommendation method for combining user comment content and grades. A model for combining the user comment content with user grades is provided for solving the cold start problem and the poor interpretability problem in a recommendation system. By means of the rich information contained in comments, the prediction accuracy is greatly improved; particularly, when data is quite sparse, the cold start problem and the poor interpretability problem can be well solved. The method mainly considers that descriptions of item characteristics are contained in user comment information, and potential characteristics in numerical value grades correspond to item characteristics in comment information through a mapping function. A model can be well built for user preferences, and therefore prediction and recommendation can be well carried out even if data is quite sparse.

A generalizable and interpretable deep learning model for predicting microsatellite instability from histopathology slide images is provided. Microsatellite instability (MSI) is an important genomic phenotype that can direct clinical treatment decisions, especially in the context of cancer immunotherapies. A deep learning framework is provided to predict MSI from histopathology images, to improve the generalizability of the predictive model using adversarial training to new domains, such as on new data sources or tumor types, and to provide techniques to visually interpret the topological and morphological features that influence the MSI predictions.

The invention provides a recommendation system and a recommendation method based on an attention mechanism, wherein the merchandise in the user history and the merchandise to be predicted are mapped to the feature vector of the merchandise through the feature embedding layer, according to the self-attention mechanism, the attention layer learns the user's expression to obtain the user's feature vector, a prediction value for the merchandise from a user is output according to the feature vector of the merchandise and the feature vector of the user through a fusion output layer. The invention provides a recommendation system based on attention mechanism, which adopts a neural attention mechanism to automatically distribute weights to each merchandise in a user's history record, so that the interpretability is enhanced. The invention outputs a prediction value of the user to the merchandise according to the characteristic vector of the merchandise and the characteristic vector of the userthrough a fusion output layer. The neural attention mechanism can be used to simulate user behavior more reasonably and get finer grained weights so as to obtain better recommendation effect.

The invention discloses a random-forest-model-based power transformer fault diagnosis method, which comprises the following steps of: acquiring transformer state overhauling data, training a random forest model by utilizing the transformer state overhauling state, checking the sensitivity of the random forest model, and diagnosing a fault of a transformer by using the trained and checked random forest model. The method provided by the invention is high in adaptability and interpretability; and criticality between normality and the fault is separated by utilizing a k-means clustering method, and a system is endowed with fault early warning capability.

The invention discloses a meta-learning algorithm based on stepwise gradient correction of a meta-learner, and the algorithm comprises the steps: firstly, obtaining training data with noise marks anda small amount of clean unbiased metadata sets; establishing a meta-learner, namely a teacher network, on the metadata set relative to a classifier, namely a student network established on the training data set; and carrying out united updating of student network parameters and teacher network parameters by using random gradient descent; obtaining a student network parameter gradient update function through a student network gradient descent format; feeding the network parameters back to the teacher network, and updating the teacher network parameters by using metadata to obtain a corrected student network parameter gradient format; and then updating the student network parameters by using the correction format. Accordingly, the student network parameters can achieve better learning in thecorrection direction, and the over-fitting problem of noise marks is weakened. The method has the characteristics of easiness in understanding, realization, interpretability and the like of a user, and can be robustly suitable for an actual data scene containing noise marks.

A new method and a computer system to implement the method for SKU sales prediction in the presence of promotions for existing and new SKUs are introduced. These predictions are necessary for the retailer's order quantity decisions. Considering that the retailers are faced with shorter product life cycles and increasingly dynamic categories, they need tools that can cope with new SKUs and other changes in the category. The interpretability of the models is an important requirement for user acceptance of automated systems. The Driver Moderator method models are parsimonious and consistent across sister categories, and provide generalized insights across SKUs and stores into store, brand, SKU dynamics for various decisions such as category assortment or promotion planning.

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 discloses an analog circuit fault diagnosis method and an implementation method of the analog circuit fault diagnosis method. The content includes the first part of analog circuit fault feature information extraction, the second part of fault classifier construction, and the third part of implementation of algorithm software. The analog circuit fault diagnosis method includes the following steps of constructing a fault feature information base, selecting an optimal mother wavelet through an information entropy maximizing principle, conducting wavelet decomposition on response nodes of a measured circuit, extracting the optimal feature of the measured circuit, conducting dimensionality reduction on the fault features through principal component analysis, conducting fault classification and intelligent diagnosis, constructing a fault diagnosis device according to the obtained fault feature information and through a multi-classifier cascade connection model and the classifier integration technology so as to recognize existing faults and causes of the faults, and conducting specific implementation on the algorithm through a C#.NET platform and through combination with the Weka software. The diagnosis method and the implementation method have the advantages of being high in fault diagnosis performance, wider in diagnosis range, higher in algorithm robustness and higher in interpretability.

The invention discloses a heterogeneous information network based content providing method. When users subscribe to a recommendation service, content is recommended to the users through an optimal prediction matrix, a similarity matrix of the users and a similarity matrix of projects are obtained by an element path based similarity calculation method according to a heterogeneous information network through the optimal prediction matrix, the fusion is performed on two or three of the user similarity matrix, a user and project evaluation matrix and the project similarity matrix to obtain the internal relation between the users and the projects, the prediction is performed by a collaborative filtering based matrix decomposition prediction method, and results are combined to obtain the optimal prediction matrix. According to the heterogeneous information network based content providing method, the recommendation accuracy is effectively improved, the results which are accord with the will of the users can be recommended to the users through limited time calculation on the basis of the existing data, the cold start problem is partially solved, and the interpretability of the recommended results is improved.

The invention discloses a short text topic modeling method based on word semantic similarity. The method comprises: according to word semantic similarity provided by external, establishing a similar word set of short text centralized words; determining the number of topics used in modeling; randomly distributing the topic of each short text; through a Gibbs sampling process, iteratively determining the topic of each short text and the distribution of the words in the topic; according to a final distribution result of the above variable, feeding back the word distribution under each topic and the topic associated to each short text. The method preferably solves problems of sparse information contents of short texts and unclear semantic expression. According to the model result, short texts can be preferably expressed as topic vectors, and the topic vectors are used as final feature vectors of a short essay. The topic vector-based expression has good semantic interpretability, and can be used as algorithm basis of various applications. The method can be widely applied in various short text data, and has wide actual meaning and commercial values.

The invention discloses a trajectory recovery method based on depth learning and Kalman filter correction, which comprises the following steps: S1, discretization of trajectory points; S2,loop NeuralNetwork and Trajectory Modeling; S3, trajectory recovery; S4, using spatio-temporal attention mechanism, obtains attention model based on sequence-to-sequence model; S5, combines Kalman filter and circulating neural network, introduces Kalman filter to optimize the mean square error, cooperatively trains the Kalman filter and the attention model, and obtains the final model. The invention providesa trajectory recovery method based on depth learning and Kalman filter correction, the transfer law between the trajectory points is modeled by the loop neural network. The attention mechanism in depth learning is used to help trajectory recovery. Finally, Kalman filter is introduced to model the movement of the object in time and space, which reduces the unexplainability and error of depth learning model, has stronger explainability, and reduces the error of trajectory recovery.

The invention discloses a method and device for identifying text emotion types based on a cognitive structure model. The method comprises the steps of automatically constructing an emotion dimensionality dictionary for input massive open-source texts by using a statistical approach based on a general semantic dictionary and a syntax dependence relationship; carrying out refinement on the constructed emotion dimensionality dictionary, wherein the refinement specifically comprises the steps of carrying out inconsistency processing of semantics and emotional tendency and filtering non-emotional words; obtaining corresponding emotion types by combining the corresponding relation of the emotion dimensionality values and the emotion types in the emotion cognitive structure model based on the high-quality emotion dimensionality dictionary obtained after the refinement. According to the technical scheme, the method for identifying the text emotion types based on the cognitive structure model is obviously superior to an existing method on the design concept, the interpretability, the using flexibility and the validity, and can be used for emotion analysis and identification for texts in the fields such as business intelligence, social public sentiment and decision-making evaluation.

The invention discloses a project-level and feature-level deep collaborative filtering recommendation algorithm based on an attention mechanism. The algorithm comprises the following steps of S1, counting historical project scores of a user; S2, calculating the feature level content representation of the user on the target project according to the historical project score of the user; and S3, calculating a project-level prediction score of the user for the target project according to the historical project score of the user and the technical result of the S2. According to the algorithm, the recommendation precision is improved to a certain extent by combining attention mechanisms on a project level and a feature level, and compared with the prior art, the algorithm has higher interpretability in analysis of historical preferences of users. The extended DACFs, such as recently proposed neural collaborative filtering and discrete collaborative filtering, will also be considered in othercollaborative filtering models, a higher-order characteristic level attention mechanism is explored for future research, and the theoretical basis of research of the recommendation system is further tamped.

The invention discloses an explainable cervical cell image rapid hierarchical recognition method and system. The method comprises steps of 1) pretreating a segmented cervical cell image; 2) judging whether the pretreated cell image is a single cell and, if so, moving to a step 3), otherwise, the image being an indivisible cell cluster, and moving to a step 7); 3) calculating a cell parameter characteristic; 4) establishing a cell knowledge map reasoning judgment model and inputting the cell parameter characteristic into the model to obtain a first hierarchical result of the single cell; 5) constructing a double-flow convolution neural network model with additional domain knowledge, and obtaining the second hierarchical result of the single cell based on the model; 6) obtaining a hierarchical result of the single cell in conjunction with the first hierarchical result and the second hierarchical result; and 7) constructing a double-flow convolution neural network model of cell clusters, and using the model to perform hierarchical identification of the cell clusters of indivisible cell clusters.

The invention relates to an information recommendation method based on a convolutional neural network and a joint attention mechanism, which is used for effectively utilizing potential semantic information of text and overcoming inherent defects of a feature extraction method of traditional machine learning. According to the method, feature vectors of the evaluation text processed by a CNN deep neural network is processed by a layer of attention mechanism, so that the attention weight of key points of interest in the evaluation text is increased. The vector sets of users and projects and thescore of the previous attention mechanism respectively use a layer of attention mechanism to acquire attention mechanism weight vectors of the users and the projects respectively. Point multiplication is carried out on the attention mechanism weight vectors and vector sets of the users and the projects respectively to obtain final representation, the users, the projects and the evaluation text are combined to obtain the final representation, and score prediction is made. Compared with traditional recommendation technology, the method has the advantages that recommendation can be performed more effectively, the recommendation quality is improved, and the interpretability of recommendation is enhanced.

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 provides a personalized recommendation method and system based on collaborative filtering and deep learning, and the method comprises the steps: obtaining historical behavior feature data of commodities purchased by a user, carrying out the preprocessing, and sorting the purchasing behaviors of the user according to the time, wherein the sorted data is called a behavior feature sequence of the user; performing the personalized recommendation system modeling, which comprises the steps of obtaining input vectors of a user and a commodity from an interaction matrix, then respectively generating embedded vectors of the user and the commodity, weighting the embedded vectors through an attention neural network, and performing linear and nonlinear interaction on the weighted embedded vectors, thereby obtaining the explicit and implicit relationship between the user and the commodity; finally, estimating the click rate of the user to the commodity; and training and testing the model by using the user behavior characteristic sequence. According to the method, the collaborative signals of the users and the commodities are fully mined, a basis is provided for capturing personalized demands of the users, and the accuracy and interpretability of a recommendation system can be improved.