56 results about "Dirichlet distribution" patented technology

The invention discloses a video semantic representation method and system based on a multi-mode fusion mechanism and a medium. Feature extraction: extracting visual features, voice features, motion features, text features and domain features of a video; Feature fusion: performing feature fusion on the extracted visual, voice, motion and text features and domain features through the constructed multi-level hidden Dirichlet distribution topic model; And feature mapping: mapping the fused features to a high-level semantic space to obtain a fused feature representation sequence. The model utilizesthe unique advantages of the theme model in the semantic analysis field, and the video representation mode obtained through model training on the basis of the model has ideal distinction in the semantic space.

The invention discloses a parallel data processing method based on the hidden Dirichlet distribution model, which belongs to the data mining field. The method includes three solutions, including the multi-process parallel processing, the multi-thread parallel processing and the composite multi-process multi-thread processing; the data DM for being processed is divided into data segments in equal or unequal length in the three solutions; each data segment is provided with an index; each computer process / thread processes the corresponding data segment through applying the index, so as to obtain the subject information of each data item and generate the local sufficient statistic; when the whole DM is processed, the global sufficient statistic is obtained through the merge of the local sufficient statistics so that the current Mi model is obtained through the estimation until the model becomes convergence. The parallel data processing method can utilize the multi-kernel parallel frame of a single computer and the cluster large-scale parallel capability of multi-computer to realize the high-speed processing of the large-scale text sets and effectively reduce the memory usage during the parallel processing process.

Provided is a tax document hierarchical classification method based on multi-tag classification. Firstly, generated subject distribution is extracted from a latent Dirichlet allocation model, and a latent Dirichlet allocation topic character of a tax file is built; then, tf idf feature vectors corresponding to training data are built, the tf idf feature vectors including the training data and files to be classified are calculated, and similarity is calculated to obtain candidate category tags; finally, source data of candidate category tag nodes are supplemented with auxiliary data, a multi-tag classification model based on transfer learning is built through a transfer learning algorithm TrAdaBoost, and the files to be classified are classified. According to the method, a hierarchical classification problem is converted into a searching stage and a classification stage, calculated amount is greatly reduced by means of incremental candidate category searching, computation complexity is lowered, the tax files are mapped to tax category hierarchical categories by means of the multi-tag classification model based on transfer learning, the auxiliary data are effectively used, and classification performance is improved.

The invention belongs to the technical field of information processing, and provides a label extraction method and device. The label extraction method comprises the following steps: obtaining a plurality of pieces of evaluation information of a commodity; extracting candidate labels in each piece of evaluation information according to a preset label syntax rule; carrying out subject analysis on each candidate label through a potential Dirichlet distribution model LDA to obtain the subject probability distribution corresponding to each candidate label, wherein the subject probability distribution comprises the probability of the candidate label belonging to each appointed subject; and determining a candidate label set corresponding to each appointed subject according to the subject probability distribution, and determining a representative label corresponding to the appointed subject according to the weighted value of each candidate label in the candidate label set. According to the method and device, the problem that the existing label extraction algorithm cannot sufficiently solve the short text sparsity problem is solved, and the correctness of the commodity evaluation similarity and the commodity evaluation mining degree are improved.

The invention belongs to the technical field of the wireless communication, specifically a channel of a multiuser large-scale MIMO system under a FDD mode, and relates to a multiuser channel estimation method under large-scale MIMO system and DP priority. The method comprises the following steps: realizing the self-adaptive clustering of the multiuser channel by using the Dirichlet distribution feature, and deriving a method capable of intelligent reconstructing the channel H under the variational Bayesian principle. Compared with the ordinary channel estimation method, the training expenditure of achieving the ideal estimation performance is greatly reduced, and the error between the reconstructed channel and the real channel is small. Compared with the traditional method, the method has the advantages that the computational burden is simplified, the operation speed and the operation precision are improved, and the accuracy of the channel estimation is improved.

The invention discloses a video resource popularity prediction method. The video resource popularity prediction method comprises the steps of counting ratings data of group users in a certain region, obtaining ratings type data and interactive behavior data of the group users and calculating the resource popularity of the data by utilizing the ratings type data; traversing the ratings type data and the interactive behavior data by utilizing an LDA model of a coupling user behavior so as to generate corresponding dirichlet distribution respectively, deducing a full probability of each behavior mode through a chain rule and getting an expected value of each dirichlet distribution so as to obtain a behavior mode matrix; regarding the resource popularity of the counted data and the behavior mode matrix as an input of a neural network by combining with a neural network model, training to generate a prediction model and further predicting the video resource popularity in future. The method comprehensively considers influence from the ratings interactive content data and interactive behavior data of users to the prediction of the resource popularity, researches the relationship between the two types of data and the popularity, and improves the prediction accuracy of the resource popularity.

The invention discloses a natural image classification method based on potential Dirichlet distribution. The natural image classification method mainly solves the problems that an existing full supervision natural image classification method is long in classification time and reduces the classification accuracy on the premise that the classification time is shortened. The natural image classification method includes the implementation steps of obtaining hue, saturation, luminance and distinguishing characteristic images of each natural image, respectively conducting gridding dense sampling on the characteristic images to obtain gridding sampling points of the characteristic images, extracting SIFT characteristics in the peripheral region of each gridding sampling point, conducting K clustering on the SIFT characteristics of the characteristic images in the same kind to generate a vision dictionary, using the vision dictionary to quantize all the characteristic images into vision documents, sequentially connecting the vision documents, inputting the sequentially-connected vision documents into an LDA model to obtain potential semantic theme distribution, and inputting the potential semantic theme distribution of all the natural images into an SVM classifier to carry out classification so as to obtain classification results. Compared with a classic classification method, the natural image classification method shortens the average classification time, meanwhile, improves the classification accuracy and can be used for object identification.

The invention discloses a user attribute scoring guide-based personalized recommendation system and a recommendation method thereof. The system comprises a data preprocessing module used for performing crawling preprocessing on network text contents and performing keyword tag marking on a preprocessed text, a user interaction module used for collecting historical browsing information and scoring information of a user to generate a user attribute document and a scoring matrix, a similar neighbor user search module used for building a latent Dirichlet distribution model according to the attribute document corresponding to historical data of the user, and an interest distribution mining module used for mining interest distribution of the user in combination with Pearson similarity and latent Dirichlet distribution according to neighbor users. By utilizing a scoring guide-based latent Dirichlet distribution method, the scoring accuracy is improved; and the text can not only assist the user and a content platform to manage massive text contents, but also improve the accuracy of an existing system in numerous application scenes of new media, e-commerce and the like.

The invention relates to the field of data mining, and particularly discloses a telecom user similarity finding method based on a latent Dirichlet allocation (LDA) subject model. The telecom user similarity finding method is a method organically linking multi-dimensional characteristics of telecom users with a subject finding algorithm based on a probability model, and considering from four different aspects: basic attributes, call records and short message records of the users, and position information, and connection start time and end time of all base stations to which the users are connected in a day. The method provided by the invention focuses on modeling an information corpus of the base stations to which the users are connected in one day by using the LDA subject model, digging potential subject information hidden in texts by using statistical properties of the texts to acquire subject distribution of the texts, computing similarity of the texts accordingly, and thus providingstrong guarantee for deeply digging similar characteristics of the telecom users.

According to the Bayesian network-based dynamic approximate weight wind turbine generator running state comprehensive evaluation method, The method comprises steps of utilizing scada data to determinea wind turbine generator state evaluation parameter vector, and classifying wind turbine generator fault modes; And constructing a three-layer Bayesian network to describe the operation parameter vector of the wind turbine generator and the fault causal relationship, determining the prior distribution of the Bayesian network parameter vector, and determining the hyper-parameter of the product Dirichlet distribution through the experience knowledge of the wind turbine generator. And determining the posterior probability distribution of the Bayesian network parameters. And calculating conditional probability distribution of each node in the Bayesian network under different states of father nodes of the node. And comprehensively evaluating the operation state of the wind turbine generator according to the dynamic approximate weight of the comprehensive evaluation of the operation state of the wind turbine generator. According to the method, the operation state of the wind turbine generator is evaluated rapidly and effectively, the abnormity and degradation trend of equipment are found in advance, predictive maintenance is achieved, faults are effectively avoided, economic losses arereduced, and the economy and safety of the wind power plant are improved.

The invention discloses a topic clustering-based power public opinion abstract extraction optimization method and system. The method comprises the steps of obtaining a power industry news text of a to-be-extracted abstract; clustering the power industry news texts of which the abstracts are to be extracted by taking sentences as units; performing subject term extraction on the clustering result byusing implicit Dirichlet distribution LDA to obtain subject terms of the power text; performing statistics on words with the same or similar semanteme as the text subject term and word frequencies thereof in the power industry news text to be abstracted, and combining the words with the text subject term to obtain a high-frequency word set under a topic corresponding to the power industry news text; constructing a text network diagram for the power industry news text of which the abstract is to be extracted; performing abstract extraction processing based on the text network diagram and the high-frequency word set to obtain a candidate abstract sentence group; redundancy elimination is carried out on the candidate abstract sentence group to obtain a primary abstract; and optimizing the primary abstract to obtain a final abstract, and outputting the final abstract.

The invention discloses an online social network (short for OSN) community discovery method based on a Latent Dirichlet Allocation (short for LDA) theme model. The method comprises the following steps first pre-processing data, building an LDA theme model (including an LDA-F model and an LDA-T model) based on a relationship between a user in the online social network and other friends and word information expressed by the user to solve a model probability distribution, then estimating parameters via a Gibbs sampling algorithm, and at last discovering an OSN community according to the estimated parameters. By the use of the OSN community discovery method based on the LDA Theme model, a corresponding probability model can be achieved based on user blog semantic information discovery without the use of information connection via the network topology; blog content semantic similarities are introduced to effectively describe user interest and hobby probability distribution conditions; and with the introduction of community internal topological connection tightness, communities with close internal topological connections can be discovered.

The invention discloses a code element rewriting information hiding detection method based on Bayesian inference. The method comprises the steps that 1) steganographic sensitive code elements are selected according to the entropy of compressed speech code element value distribution in a training sample, and a strong code element association network is constructed; a code element Bayesian network classifier is constructed based on the strong code element association network, and Dirichlet distribution is used as parameters for priori distribution learning of the code element Bayesian network classifier; 2) according to the code element Bayesian network classifier and the training sample, a steganographic index threshold Jthr is calculated; and 3) for an unknown type of compressed speech, the steganographic index J0 is calculated; if J0 is greater than or equal to Jthr, the speech segment is an unimplerfied speech segment; if J0 is less than Jthr, the speech segment is a steganographic speech segment. According to the invention, the method can acquire a more accurate steganographic detection result; the method uses the code elements in a codestream as the analysis object; decoding isnot needed; and real-time steganographic detection is realized.

The invention discloses a retrieval method of remote sensing images. The method includes acquiring and preprocessing the remote sensing images; then detecting and describing bottom visual features, and clustering to generate visual vocabularies; performing remote sensing information retrieval on the basis of a hidden dirichlet distribution model; finally implementing highly-accurate ground feature retrieval. The retrieval method of the remote sensing images has the advantages that fine dimension reduction effect is provided, retrieval accuracy is higher, and a retrieval map can be marked accurately.

The invention discloses an online theme modeling method on the basis of theme heredity. The online theme modeling method includes the steps of: capturing text data of a current time slice, performing theme modeling according to an LDA (latent dirichlet allocation) model, computing theme strength, ranking a theme, computing a gene of the theme, capturing text data of the next time slice, converting distribution vectors of theme-vocabulary, computing prior parameters of Dirichlet distribution of the next time slice, adopting a Gibbs sampling method and the like. The online theme modeling method has the advantages that 1, an online theme model is suitable for processing of time-sequential text streams and can be applied to a public opinion monitoring system greatly; 2, alignment features of the theme in an OLDA (online latent dirichlet allocation) model are reserved, different genes are set for the themes according to the theme strength, and the defects that the themes are mixed and new themes are not detected timely are overcome; 3, scores of broad themes can be effectively lowered by the aid of a theme strength computing method.

The present invention provides an LDA-based user consumption forecasting method for electronic commerce. The method comprises: mining a correlation between a current consumption behavior and a historical consumption behavior, a track of browsing commodities of a user, consumption information of the user, review information of a merchant for the user, personal information of the user and such more dual information of the user and commodity; modeling a consumption behavior and a product by using a Dirichlet distribution relation and an LDA topic model; constructing a probability model of a commodity, a user and between two; analyzing a new consumption behavior according to an obtained probability distribution model, so as to realize consumption forecasting of an e-commerce platform.

A method for an application provides weights for a neural network configured to dynamically generate a training for the neural network to detect uncertainty with regards to data input to the neural network. A training loss is determined for the neural network to minimize an expected Lp norm of a prediction error, wherein prediction probabilities follow a Dirichlet distribution. A closed-form approximation to the training loss is derived. The neural network is trained to infer parameters of the Dirichlet distribution, wherein the neural network learns distributions over class probability vectors. The Dirichlet distribution is regularized via an information divergence. A maximum entropy penalty is applied to an adversarial example to maximize uncertainty near an edge of the Dirichlet distribution.

The invention provides a multi-content implicit Dirichlet distribution model, a construction method of the multi-content implicit Dirichlet distribution model and an implicit pathogenesis mining method of a traditional Chinese medicine case by utilizing the model. According to the invention, the method comprises the steps: taking each traditional Chinese medicine case as a word containing a groupof symptoms and a group of corresponding medicine words, giving a solving method of model parameters based on the multi-content implicit Dirichlet distribution model, and finally obtaining a pluralityof symptoms and medicines belonging to the same pathogenesis. According to the method, the traditional Chinese medicine diagnosis and treatment process is digitized, implicit pathogenesis can be mined based on implicit meaning analysis on symptoms in the medical case and corresponding Chinese herbal medicine prescriptions, and the relation between the implicit pathogenesis and the symptoms and the relation between the implicit pathogenesis and medicine are found.

The invention discloses multi-attribute decision-making software for sustainable traffic network design by using R language.. Firstly, an upper-layer policy maker generates a plurality of feasible design schemes by using Dirichlet distribution. Then, a lower traveler makes a series of behavioral responses to each feasible scheme, which is a four-stage sequence model with a feedback mechanism. Thefeedback process of a lower layer can converge to traffic system equalization, and the equalization result is further fed back to an upper layer so as to measure a plurality of attributes of economy,environment, society, safety and the like of the traffic network. Finally, after a plurality of attribute values are calculated for each scheme. An optimal feasible scheme is determined by adopting amulti-attribute decision-making method. Due to the fact that the open-source free R language is adopted, the software is easy to operate and spread. The method is very useful for designing a sustainable traffic network under the condition of giving a traffic infrastructure investment budget.

The invention discloses potential sustainable traffic network design integrated with societal, economic, environmental and safe dimensions. Therefore, a multi-target double-layer model is developed, and the leader-follower essential relationship between a planner and a traveler is clearly expressed. On an upper layer, a planner designs a sustainable traffic network by optimizing investment budgetallocation. At the lower layer, the traveler makes a series of behavior responses to the traffic network design at the upper layer through a four-stage model with feedback. The feedback process can converge to traffic system equalization, and an equalization result is further fed back to the upper layer to measure the sustainability of the traffic network. In order to solve the proposed multi-target double-layer model, the invention designs a multi-attribute decision-making method on the basis of Dirichlet distribution, a continuous average method, a Frank-Wolfe algorithm and a Dijkstra algorithm. An example of an Nguyen-Dupuis network is used for verifying the effectiveness of the method disclosed by the invention. Research finds that a designed method is effective for searching satisfactory solutions.

The invention discloses a topic modeling method based on data enhancement. The method is characterized by comprising the following steps that step one, a document collection is acquired and represented; step two, a topic of the document collection D is extracted by using a potential dirichlet distribution model, and K topic-word distributions and |D| document topic distributions are obtained; stepthree, topic influence assignment is carried out on the words; step four, the data enhancement is carried out on each document; step five, the topic model of the data enhancement is constructed, andthe final topic-word distribution is obtained. According to the modeling method, the document information can be fully utilized to carry out the data enhancement under the circumstance of data sparseness, so that the topic quality is enhanced.

The present invention provides an object partitioning method and device based on the object behavior and subject preference. The method comprises the following steps of: acquiring an initial documentset of a first number of target objects; obtaining a topic preference vector of each target object by using a potential Dirichlet distribution model; normalizing a behavior vector of each of the target objects; updating the initial document set with the subject preference vector and the standard behavior vector; calculating a category group to which each target object belongs by using a potentialDirichlet mixing model based on an updated document set of each target object; obtaining an average value of subject preference vectors and an average value of standard behavior vectors of all targetobjects in each category group respectively to obtain a feature vector of each category group. The embodiment of the invention can realize the modeling of a plurality of different characteristics, isconducive to determining the classification of each target object, and improves the object division result.

The invention discloses a method for estimating a master user duty ratio through variational inference. The method comprises the following steps: sampling signals in a plurality of continuous time slots; calculating the average power of each time slot according to the samples collected in each time slot; introducing a mixed Gaussian model with a plurality of Gaussian distributions; calculating a Dirichlet distribution parameter corresponding to the mixing coefficient of each Gaussian distribution, two hyper-parameters of a mean value and two hyper-parameters of precision by using a variationalinference method; calculating a variational lower bound according to the five parameters; determining a decision formula according to the values before and after the variation lower bound change, anddetermining whether to update the probability that the average power of each time slot obeys each Gaussian distribution or not according to the decision formula; for the average power of each time slot, classifying the average power according to a plurality of probabilities corresponding to the average power; obtaining an estimated value of the duty ratio of the main user according to the averagepower number and the total number of time slots in the category with the minimum average value; the method has the advantages that the duty ratio of a main user can be accurately estimated, noise power does not need to be known, and a threshold value does not need to be set.