127 results about "Matrix decomposition algorithms" patented technology

The invention relates to a method and a device for performing personalized recommendation on users. The method mainly comprises: calculating the similarity of users through label data of the users, calculating the similarity of products through label data of the products, and obtaining a neighboring set of each user and each product according to the similarity information; calculating characteristic vectors of the users and the products through a nearest neighbor matrix decomposition algorithm on the basis of the neighbor set information of the users and the products; and forecasting grading data of products not judged by the users according to the characteristic vectors of the users and the products, and performing personalized product recommendation according to the grading data. Labeling information and the neighbor set information can be guided into a matrix decomposing model by means of an embodiment of the method and the device, so that similar users / products can have the similar characteristic vectors, and the device for performing personalized recommendation on users can have the advantages of the matrix decomposing model and can capture the strong ties among neighbors in an overall situation.

The invention provides a face recognition method capable of extracting multi-level image semantics based on the CNN (convolutional neural network). The method comprises the following steps: implementing further optimization on the basis of VGGNet, providing a new network structure, and implementing cross-level splicing on multi-level features to ensure that the finally-extracted image features have multi-level image semantics; and meanwhile, adding the extracted traditional features in the training of the CNN as additional features to ensure that the CNN feature information is more complete; then optimizing the structure of a shallow convolutional layer to ensure that the redundancy calculation is reduced and the calculation amount of models is greatly reduced; and finally, accelerating the convolutional layer by using an improved matrix decomposition algorithm to ensure that the network can be accelerated within 1 second and the accuracy rate of the models cannot be reduced when the speed-up ratio reaches four times. A face matching algorithm achieved by the invention has the characteristics of high accuracy and high real-time performance, and has higher accuracy and higher calculation efficiency compared with existing algorithms.

The invention discloses a project recommendation method based on attribute coupled matrix decomposition. The method comprises steps as follows: firstly, giving attribute information of projects and calculating the similarity between the projects with a coupled object similarity measuring index; then learning conceal eigenvectors of users and the projects with a matrix decomposition algorithm, during learning of the conceal eigenvectors of the projects, constructing a regularized term by means of the attribute information of the projects, and constraining the execution process of matrix decomposition, so that the projects with similar attribute information have similar conceal eigenvectors; finally, according to the learned conceal eigenvectors of the users and the projects, projecting scores of projects which are not scored by the users by use of inner products of the conceal eigenvectors of the users and the predicts, and providing personalized project recommendation for the users according to the predicted scores. The problems of similarity calculation of the projects, cold start of project terminals and recommendation accuracy in a recommendation system are solved with the method.

The invention discloses a personalized information recommendation method based on metapath with attributes, including modeling all the information as a heterogeneous information network, dividing themetapath with attributes into completely symmetrical and semi-symmetrical metapath with attributes according to whether the attribute values are the same, obtaining the correlation matrix of entitiesunder each metapath with attributes, and obtaining the correlation matrix of entities in the whole network by weighting; based on the matrix decomposition algorithm combined with the regular term which is composed of the correlation degree and the correlation degree weight vector, the implicit meaning matrix of the user and the object, constructing the objective function and updating the implicitmeaning matrix and the correlation degree weight vector iteratively, and calculating the prediction score of the user to the object from the obtained implicit meaning matrix, according to the predicted score, recommending the item to the user as the object to be recommended. The invention satisfies the personalized demand of the user for the information recommendation, can improve the recommendation accuracy and solve the cold start problem to a certain extent.

The invention provides a space interest point recommendation method of considering both diversity and personalization, and relates to the technical field of space interest point recommendation. The method comprises: constructing a geography-society relationship model; calculating relevance which is of place pairs in the model and on locations and society connection; constructing a relevance matrixW; dividing a user society relationship network graph G constructed in the model; calculating a loss function in dividing; selecting an eigenvector enabling the loss function to be smallest, and dividing vertices in the graph G to obtain k interest point sets with the diversity; and selecting an interest point, which best fits user preference, from each of the k interest point sets to form an interest point recommendation list fusing the diversity and the personalization. According to the space interest point recommendation method of considering both the diversity and the personalization provided by the invention, the geography-society relationship model of interest points, a spectral clustering algorithm and a matrix decomposition algorithm are fused, and the diversity is also consideredwhile interest points recommended for a user are enabled to have a higher accuracy rate.

The invention discloses a method for quick electromagnetic simulation of a complex circuit, comprising the following steps of efficiently simulating the complex circuit based on MDA (Matrix Decomposition Algorithm)-combined novel SVD (Singular Value Decomposition); precisely simulating a model by utilizing a triangle; and carrying out low rank compression on a far field group by utilizing a favorable low rank characteristic principle of the far field group in a tree structure to obtain a relatively sparse matrix representation form. The invention provides an MDA-based novel singular value compression method which is independent of an expansion form of a green function so as to shorten internal memory and computation time. The method disclosed by the invention is a pure mathematical method, can reduce computation complexity to O (NlogN) and internal memory consumption to O (NlogN), is particularly suitable for electromagnetic simulation analysis of the large scale complex circuit and can also provide effective condition for simulation of more complex circuit structure of the green function.

The invention discloses an Alzheimer's disease preliminary screening method based on speech feature non-negative matrix decomposition. The Alzheimer's disease preliminary screening method includes thefollowing steps: extracting acoustic features including fundamental frequency, energy, harmonic-to-noise ratios, formants, glottal waves, linear prediction coefficients, and constant Q cepstrum coefficients, from speech samples of Alzheimer's patients and normal humans, and splicing the features into a feature matrix; using the non-negative matrix decomposition algorithm to decompose the featurematrix, and obtaining the dimensionality-reduced feature matrix; using the dimensionality-reduced feature matrix as an input, and training a support vector machine classifier; and inputting the dimensionality-reduced feature matrix of a test speech sample into the trained support vector machine classifier, and determining whether the test speech is speech of normal humans or speech of Alzheimer'spatients. The invention adopts non-negative matrix decomposition to perform dimensionality reduction transformation on high-dimensional input acoustic features, the dimensionality-reduced feature matrix has better discrimination, and the method can obtain more excellent effects in Alzheimer's disease preliminary screening.

The invention discloses a method capable of implementing quick electromagnetic simulation of complex layered medium structures, wherein the method is used for efficient simulation of the complex layered medium structures on basis of combination of matrix decomposition and novel singular value decomposition. The method includes: using triangular meshes to precisely simulate a model, and subjecting far-field groups to low-order compact according to the principle of fine low-order characteristics of far-field groups in tree structures to obtain a sparse matrix representation form. The method which is a pure mathematic analytical method is independent of an expansion form of the green function, internal memory and computing time are reduced by the novel singular value compact method based on MDA (matrix decomposition algorithm), the computing complexity can be lowered to O (NlogN), and memory consumption can be lowered to O (NlogN). The method is especially suitable for electromagnetic simulation analysis of complex layered medium structures and can be also used for providing effective means for layered medium structure simulation with the complex green function.

The present invention discloses a decomposition method for mixed pixels of a hyperspectral image. According to the method, in a process of using an NMF algorithm to decompose the mixed pixels, an SC-SVM algorithm is adopted to solve the quantity of end members and spectral vectors in the mixed pixels, the quantity of the end members can be actively recognized and the spectral vectors of the end members can be marked by the SC-SVM algorithm, thereby solving a basis matrix from the NMF algorithm; and a coefficient matrix (abundance matrix) is calculated by a non-negative matrix decomposition algorithm, gradual iteration to convergence is carried out by minimizing an objective function to calculate a wanted result, and finally achieve a purpose of decomposition of the mixed pixels.

The invention discloses a route recommendation method and system, electronic equipment and a computer storage medium. The method comprises the steps of obtaining historical data of a user in an OTA website; inputting the historical data into a matrix decomposition algorithm to obtain theme preferences of the user and theme preference weights corresponding to the theme preferences; acquiring all scenic spot information of a destination, and generating a scenic spot set; generating a plurality of initial recommended routes and route preference scores; calculating a recommendation score of each initial recommendation route according to the theme preference weight and the route preference score; sorting the plurality of initial recommendation routes according to the recommendation scores, andrecommending the routes to the user based on sorting. According to the invention, a plurality of initial travel recommendation routes can be recommended to the user; in addition, personalized routes can be recommended to the user, the route obtaining efficiency of the user is also improved, the complexity of making a strategy before the user visits a strange city is reduced before travel, and theconvenience of obtaining traffic modes and real-time information by the user based on the routes in travel is improved.

The invention provides a key-element-based matrix decomposition and fine tuning method. The method comprises: scoring information of a commodity by a user is collected and a sparse scoring matrix R is formed based on the scoring information; according to the sparse scoring matrix R, two low-dimensional matrixes P and Q are initialized, multiplying of the low-dimensional matrixes P and Q is carried out to obtain a predicted matrix R^, and an objective function based on minimization of a difference value between the sparse scoring matrix R and the predicted matrix R^; and the objective function is optimized by an iterative process to obtain the low-dimensional matrixes P and Q and the predicted matrix R^, and fine tuning and updating are carried out on the predicted matrix R^, thereby obtaining a final predicted matrix. According to the invention, in order to overcome defects in the existing matrix decomposition algorithm, a quick matrix decomposition and fine tuning algorithm is put forward based on improvement of the algorithm. With the method, the decomposition efficiency and the decomposition speed of the matrix are increased and the recommendation accuracy is enhanced.

The invention discloses a distributed nonnegative matrix decomposition method. An existing distributed matrix decomposition algorithm DSGD (Distributed Stochastic Gradient Descent) is improved, and data skew in a program distributed execution process is reduced to effectively lower the time consumption of locking wait in the DSGD so as to improve algorithm execution efficiency. On the basis, nonnegative control and the dynamic step size therefore are added to select and design the distributed nonnegative matrix decomposition algorithm. Meanwhile, the algorithm is realized on a Spark platform. The distributed nonnegative matrix decomposition method is high in execution efficiency is high and good in astringency, and the distributed nonnegative matrix decomposition method can be favorably applied to each application field including collaborative filtering and the like.

The invention provides a recommendation method for aggregating a knowledge graph neural network and adaptive attention, and the method comprises the following steps: S1, taking a knowledge graph triple of a user, a relationship and an entity as an input, and distributing initial embedding representations, namely a user embedding representation, a relationship embedding representation and an entity embedding representation, for the input; S2, using an inner product to represent the importance degree of the relationship to the user; converting the heterogeneous knowledge graph into a weighted graph, then selecting neighbor target nodes, and training domain embedding representation of the neighbor target nodes; feeding the initial entity embedding representation into a graph neural network for training and generating a new entity embedding representation; performing polymerization to obtain a final article embedding expression; and S3, taking an inner product of the user embedded representation and the final article embedded representation as a final prediction score, and recommending the article corresponding to the highest score to the user. According to the method, the limitation problem that a matrix decomposition algorithm only utilizes interaction between a user and an article is effectively solved, and the neighbor nodes are considered when the vector representation of the neighborhood of the target node is aggregated.

The invention discloses an article recommendation method based on hierarchical multi-granularity matrix decomposition. In a recommendation system, a matrix decomposition algorithm is a recommendationalgorithm for decomposing a scoring matrix into two low-dimensional matrixes, and user preferences and article features can be learned. However, an existing matrix decomposition algorithm and an improved algorithm of the matrix decomposition algorithm only utilize a single feature vector to represent a user and an object, and therefore the problem of low prediction precision exists. In order to solve the technical problem, the invention provides a hierarchical multi-granularity matrix decomposition recommendation method based on deep learning, which can be used for recommending purchased articles with user scores. According to the method, the advantages of feature extraction by deep learning are combined, and the same user or article is represented by utilizing a plurality of different feature vectors, so that the preference representation of the user is more accurate. In addition, the technical problem that an existing recommendation algorithm based on deep learning only uses the lastlayer for prediction, but neglects information loss caused by feature transformation of each layer of the neural network is also solved.

The invention discloses a matrix decomposition recommendation algorithm combining a Word2Vec word vector model and LSH local sensitive hash based on cosine similarity. The method comprises the following steps: firstly, converting word similarity into vector similarity by utilizing a Word2Vec model, then performing LSH locality sensitive hash high-speed calculation based on cosine similarity to obtain a project similarity matrix, combining the obtained project similarity matrix with an original score, outputting a pre-score of an unscored project, and filling the pre-score into a training set;and finally, taking the training set as the input of an ALS matrix decomposition algorithm to obtain a recommendation result. Compared with a traditional collaborative filtering recommendation algorithm, the improved algorithm is lower in MAE value and better in performance, and meanwhile the problem that recommendation is not timely in a large amount of data can be effectively solved.

The invention discloses a mixed pixel decomposition method of a high-spectral remote sensing image. A multi-scale non-negative-matrix decomposition algorithm of approximately sparsity constraint is applied to demixing of the high-spectral remote sensing image, and a new approximate sparse model replaces an L0 model for solution aimed at the problem that the L0 model in the sparsity de-mixing technology of the high-spectral remote sensing image is hard to solve, and on such basis, the complex multi-scale space geometric structure of the high-spectral remote sensing image is considered, a totalvariation space constraint is added, and the decomposition precision and performance of mixed pixels are improved obviously.

The invention discloses a rapid Web server QoS (Quality of Service) prediction method based on user metadata. According to the method, the field of a request user is accurately selected through use of the metadata (geographic position information and history QoS record) of a user, and meanwhile the accuracy of prediction is effectively increased through a matrix decomposition algorithm for fusing user extension intrinsic attributes. The solving efficiency is increased greatly through a matrix decomposition acceleration technology, so that real-time response to the individual QoS inquiry requests of a plurality of users is realized.

The invention discloses a matrix decomposition recommendation method based on joint clustering. The matrix decomposition recommendation method includes the steps that a user-item rating matrix is constructed; the user-item rating matrix is divided into multiple categories through joint clustering; for the categories obtained after clustering, a probability matrix decomposition method is used for parallely predicting unknown scores of all the categories, and recommendation is carried out according to the predicting scores. Close correlation in the clustering and high precision of the probability matrix decomposition algorithm can be used fully, and for the big data processing problem in the information overload times, recommendation can be conducted at high speed while high precision is ensured.

The invention provides a location-based service travel recommendation method comprising the steps that step one, entity track-event information is acquired, and user-location-activity associated modeling is performed on the entity information through a three-dimensional tensor; step two, user track-event information is recorded; and step three, similarity matching is performed through user analogy, and the user track-event information with high similarity is actively pushed, wherein the user track-event information (from location information of mobile phones, sports bracelets and sports watches) is integrated and then matched with the related user track-event information, and pushing is performed. According to the method, mass user data are integrated and similar users and activities are found in different locations by using a collaborative filtering algorithm, modeling of the mutual relation of user-location-activity is performed through the tensor and a regularized tensor / matrix decomposition algorithm is put forward. The problem of sparsity of single user movement location data can be effectively alleviated by the algorithm and thus mining of the movement mode of the users can be performed.

The invention discloses a knowledge system-oriented personalized recommendation method and a computer storage medium, and the method comprises the steps: firstly building an initial scoring matrix, extracting scoring data information in a data source, building a two-dimensional matrix, and filling a missing element value in the two-dimensional matrix with 0; then, training the initial scoring matrix based on a project collaborative filtering algorithm, and redefining 0 elements in the initial matrix so as to form a new matrix A; training the initial scoring matrix by using a matrix decomposition algorithm, and redefining an element 0 in the initial matrix by using a training result so as to form a new matrix B; and finally, comparing corresponding element values in the matrixes A and B, resetting elements with different numerical values to obtain a matrix C, repeatedly training the matrix C, and determining a final prediction value by adopting a weighted sum mode. The method is appliedto a knowledge system, a good recommendation effect can be achieved, and the problem of knowledge overload is well relieved.

Embodiments of the invention disclose a recommendation information determination method and apparatus. A recommendation information model building process comprises the steps of obtaining identifier data of N users and at least two types of feedback data of each user, wherein N is greater than 1; by utilizing a preset partial order sample generation algorithm, performing operation on a first type of feedback data of each user to obtain a partial order sample corresponding to each user; and in combination with a preset modeling framework and a preset matrix decomposition algorithm, performing operation on the partial order sample corresponding to each user and a second type of feedback data of each user to obtain a recommendation information model. According to the method and the apparatus, the recommendation information model is built by utilizing the at least two types of feedback data of the user, and the multiple types of obtained feedback data can be effectively utilized; and compared with the model built only for scoring data, namely, one type of feedback data in the prior art, the accuracy of determining the recommendation information is improved.

The invention provides a customized product similarity judgment method based on the product structure and relates to the field of mass customized production. According to the method, at first, an adjacent matrix is used for expressing the subordination relation between parts in a product structure; secondly, column vector recombination is carried out on a part matrix representing the single product structure to form an adjacent vector of the single product structure, and on this basis, adjacent vectors of target products and inquiry products are combined to form a bank matrix; a target product and inquiry product bank matrix is decomposed based on a non-negative matrix decomposition algorithm, and dimensionality reduction of the product structure is achieved; finally, the product similarity is judged by calculating the distance between the vectors of the product structure after dimensionality is reduced. The customized product similarity judgment method based on the product structure can reduce the operation steps, improve accuracy and enhance similarity significance.

The invention relates to a non-negative matrix factorization and convolutional neural network-based color image processing method. The method comprises a step of obtaining a color image by acquisitionequipment and preprocessing the color image, wherein the preprocessing comprises the following steps of: extracting image features by adoption of a non-smooth non-negative matrix factorization algorithm; and combining a convolutional neural network to optimize a color image classification and reconstruction process: firstly reading an image data matrix to obtain a row number and a column number,initializing corresponding parameters, and combining the convolutional neural network to complete image classification and reconstruction. In color image processing, the image processing method is more accurate in feature extraction, few in reconstruction errors and high in classification efficiency.

The embodiment of the invention provides a flow matrix estimation method and device. The problems that the accuracy rate is low, the calculation complexity is high, and efficiency is low when a flow matrix is solved can be solved. According to the specific scheme, network flow in a preset range is obtained, and a constraint matrix B is obtained according to the obtained flow value; according to the preset algorithm and the constraint condition, a target function min(rank(LR<T>)) is solved so as to obtain L and R, wherein the flow matrix X=LRT, L is the left feature vector obtained by decomposing the flow matrix X through the non-negative matrix decomposition algorithm, R is the right feature vector obtained by decomposing the flow matrix X through the non-negative matrix decomposition algorithm, the constraint condition is lambda (LR<T>)=B, lambda ( ) is the linear operator, rank (LR<T>) is the rank of the flow matrix X, and the min function is the minimization function.

The invention provides an item pushing method and device based on clustering and matrix decomposition. The method comprises the following steps: constructing a user label matrix, then clustering usersto obtain a plurality of user group clusters, then generating a user-to-item scoring matrix for each user group cluster, and then carrying out matrix operation through an ALS matrix decomposition algorithm, thereby carrying out item recommendation. Because label clustering is firstly carried out on the user and then matrix operation is carried out, the calculation amount of the matrix operation can be effectively reduced, and therefore the system resource consumption is reduced. According to the method disclosed in the scheme, the technical problem of low pushing efficiency of existing item pushing is solved, the technical effect of efficient and accurate pushing of government affair items is achieved, and item pushing recall rates are increased.

The invention discloses a scoring prediction method and device thereof. The method comprises the following steps of: a generating step: generating a representation of both a user and an item by constructing a neural network model for a user review of an item; a prediction step: optimizing a matrix decomposition model by utilizing the representation of a user and an item, and training the matrix decomposition model to predict target score, wherein the neural network model is a long and short time memory network model; extracting features of the user and the item by using an attention mechanismin the long and short time memory network model, and generating representation according to features; and adding the representation of the user and the item to the matrix decomposition model by nonlinear transformation. The scoring prediction method and device thereof provided by the invention is low in cost, and migration can be well performed among different data fields; the unknown score can bepredicted more comprehensively and correctly by combining the user scoring numerical information and the user comment text information; the prediction results can be obtained in a faster time with anoptimized matrix decomposition algorithm.

The invention discloses a sparse limited non-negative matrix decomposition algorithm based ultrafiltration membrane water treatment prediction method. Firstly, a sparse limited non-negative matrix decomposition algorithm is utilized to construct a prediction model for a membrane distillation wastewater treatment process; secondly, a GA algorithm is utilized to optimize parameters of the prediction model in a pseudo-steady state and a non-steady state respectively; thirdly, the optimized prediction model is utilized to predict the change trend of the membrane flux and the membrane pollution resistance and to analyze the influence of basic operation parameters of membrane distillation on the membrane flux and the membrane pollution resistance in the pseudo-steady state and the non-steady state respectively; and finally, a prediction result is subjected to sensitivity analysis calculation to determine leading factors which influence the membrane flux and the membrane pollution resistance. According to the method, the sparse limited non-negative matrix decomposition algorithm is utilized to predict the change situations of the membrane flux and the membrane pollution resistance in real time, and the influence of the basic operation parameters of membrane distillation on the membrane flux and the membrane pollution resistance is clarified and quantified.

The invention provides a face recognition construction method and system based on non-linear non-negative matrix factorization and a storage medium. The present invention has the following advantages:1. by using norm substitution F to measure the loss degree of matrix factorization, the problem that the existing kernel non-negative matrix factorization algorithm is sensitive to the singular value is solved, and the stability of the kernel non-negative matrix factorization algorithm proposed by the invention is enhanced. 2. The sparsity of the feature representation of the algorithm of the present invention is enhanced by adding a regularization restriction on the coefficient matrix H into the objective function. 3. By integrating the kernel non-negative matrix factorization with new objective function and fractional inner product kernel function, a sparse fractional inner product nonlinear non-negative matrix factorization algorithm with high performance is obtained, which solves theproblem that the super parameters of polynomial kernel function can only be integers, and improves the discriminant ability of the algorithm.

The invention discloses a matrix decomposition-based lensless holographic microscopic speckle noise removing method and a matrix decomposition-based lensless holographic microscopic speckle noise removing device. The matrix decomposition-based lensless holographic microscopic speckle noise removing method comprises the following steps: S1, turning off a light source, and acquiring a dark field image; S2, turning on the light source, and acquiring a bright field image under uniform irradiation of the light source; S3, placing a particle-containing solution sample above a sensor, guaranteeing that the distance from the sample to the sensor is much smaller than the distance from the sample to the light source, turning on the light source, and acquiring a holographic image sequence of the sample; S4, performing flat field correction on any hologram image required to be calculated; S5, performing noise separation on the corrected holographic image by a matrix decomposition algorithm, and decomposing the corrected holographic image into two parts, namely a particle hologram part and a background noise part; S6, further performing image analysis and processing on the calculated holographic image. Through the matrix decomposition-based lensless holographic microscopic speckle noise removing method and the matrix decomposition-based lensless holographic microscopic speckle noise removing device, the speckle noise as well as interference fringe noise generated by multiple reflections of the sample can be removed, so that high-precision dynamic 3D imaging is achieved.

The invention discloses a smart power grid data decomposition method based on non-negative matrix factorization. On the basis of a non-negative matrix decomposition algorithm, the similarity featuresamong users are introduced, the performance of the algorithm is improved, so that the decomposition result is closer to actual electricity information of the users, and more precise feedback is provided for the users. As is proved by the experiment result, compared with an existing method, by means of the smart power grid data decomposition method, the performance is obviously improved.