895results about How to "Easy to dig" patented technology

The invention relates to a database searching method based on knowledge graph, and belongs to the field of structural data mining and searching. The method provided by the invention comprises: firstly analyzing factors such as a type of a database table and inter-table constraint, then generating a corresponding concept, an entity, and an inter-entity relation by using the table and the inter-table constraint, and establishing a knowledge graph service. After a natural language query input by a user is obtained, each factor queried by the user is detected to obtain a factor mode and a factor value of the query, then the factor mode is matched in a template base to obtain a corresponding query mode, then the factor value of the query is substituted into to the query model to obtain a knowledge graph query statement, and finally the query statement is executed in the knowledge graph service, to obtain corresponding knowledge queried by the user and return the knowledge to the user. According to the method provided by the invention, data and an internal relation in a database can be effectively organized and shown, and the natural language query by the user is supported, thereby improving user experience of database searching.

The invention discloses a question and answering (QA) system realization method based on deep learning and a topic model. The method comprises the steps of: S1, inputting a question sentence to the Twitter LDA topic model to obtain a topic type of the question sentence, extracting a corresponding topic word, and indicating the input question sentence and the topic word as word vectors; S2, inputting word vectors of the input question sentence to a recurrent neural network (RNN) for encoding to obtain an encoded hidden-layer state vector of the question sentence; S3, using a joint attention mechanism and combining local and global hybrid semantic vectors of the question sentence by a decoding recurrent neural network for decoding to generate words; S4, using a large-scale conversation corpus to train a deep-learning topic question and answering model based on an encoding-decoding framework; and S5, using the trained question and answering model to predict an answer to the input questionsentence, and generating answers related to a question sentence topic. The method makes up for the lack of exogenous knowledge of question and answering models, and increases richness and diversity of answers.

The invention is suitable for technical field of videos, and provides a method and a system for editing and playing an interactive video, and an electronic learning device. The method provided by the invention comprises the following steps: pre-leading an original video resource or an existing interactive video script; making area marks on frames of the original video resource or frames of the existed interactive video script corresponding to the original video and adding interactive video objects related to the area marks so as to generate an interactive video script; displaying the area marks when playing the interactive video script to the frames of the area marks; and displaying the interactive video objects when receiving activating orders. The method, the system and the electronic learning device provided by the invention have the function of interacting with a user, so that the user can conveniently search and mark the useful information pictures of the original video resource to obtain the area marks. In addition, the method, the system and the electronic learning device can take interactive operation on the area marks displayed in the play process, so that the user can understand and learn more information relative to video content when watching the video.

The invention provides an industrial control protocol fuzzing test method based on protocol state, comprising the steps of extracting a protocol state machine, building a message sequence library, guiding the protocol state, sending and storing test cases, carrying out abnormality monitoring based on heart rate, and locating a test message causing abnormality. In view of the problem that industrial control protocol fuzzing test is of high blindness and low efficiency, test cases belonging to the protocol state of an industrial control component are sent to the industrial control component based on the protocol state, and therefore, the coverage of fuzzing test is extended effectively, and the test cases are more targeted. The abnormality monitoring based on heart rate is of extensive applicability. In addition, through the method for locating a test message causing abnormality, a single message or a message sequence causing abnormality of the industrial control protocol can be located efficiently and accurately, and excavation and analysis of security holes are facilitated.

The invention discloses a Chinese word segmentation method based on two-way LSTM, CNN and CRF which improves and optimizes traditional Chinese word segmentation base on deep learning algorithm. The method comprises following specific steps: preprocessing the initial corpus, extracting corpus character feature information and pinyin feature information corresponding to characters; using the convolutional neural network to obtain pinyin feature information vector of the characters; using the word2vec model to obtain the character feature information vector of text; splicing pinyin feature vectors and character feature vectors to obtain context information vectors and put the context information vectors to a bidirectional LSTM neural network; decoding the output of the bidirectional LSTM using the linear chain condition random field to obtain the word segmentation sequence; decoding the word segmentation label sequence to obtain word segmentation results. The invention utilizes the deep neural network to extract text character features and pinyin features and combines the conditional random field decoding, can effectively extract Chinese text features and achieve good effect on Chinese word segmentation tasks.

The invention relates to a real-time multi-target human body 2D attitude detection system and method. The system comprises an image acquisition module used for acquiring image data, a real-time processing module used for inputting the image data to the neural network for learning and prediction and generating the human body attitude information according to a hot spot map of the acquired joint point position and a hot spot map of the direction vector field among joint points, and a visual display module used for presenting the predicted human body attitude information to users in a line connection mode. The system is advantaged in that the depth learning method is utilized to encode the joint position and the position and the direction of bones formed by joints through interconnection, accurate human body 2D attitude estimation of a single image is realized, for complex people gathering conditions, multiple human body attitudes of the scene can be accurately estimated, the users are facilitated to carry out further analysis processing and mining of the human body attitudes, and next behaviors are predicted.

The invention discloses a drone's low-altitude remote-sensing image high-resolution landform classifying method based on characteristic fusion. The method comprises the following steps: selecting common and representative landforms from to-be-processed remote sensing images and using them as the training samples of the landforms; extracting the color characteristics and the texture characteristics from the training samples of each landform; fusing the color characteristics and the texture characteristics; using a classifying method to classify and learn the fused characteristics to obtain the classifying model for each landform; extracting and fusing the color characteristics and the texture characteristics of the low-altitude remote sensing images of the to-be-classified drones; and finally, based on the fused characteristics of the classifying objects and in combination with the classifying model of each obtained landform, using the classifiers to divide the classifying objects into a certain landform. Therefore, the classification of the drone's low-altitude remote sensing images is achieved. According to the method of the invention, it is possible to more effectively and more quickly to extract the verification characteristics so that the classification result becomes more accurate.

The invention discloses a decision making method and system based on big data. The method integrates a series of decision making influence factors commonly used for all industries in advance and has the setting function of setting all the influence factors so that users in all walks of life can customize acquisition rules, decision making rules and the like required by decision making through simple setting, and for instance, the users can set industry types and data keywords to formulate the acquisition rules. Meanwhile, a universal classification model is built in advance, data to be classified can be classified by the model on the basis of auxiliary information set by the users according to service requirements of the users, so that data required by the users can be further mined conveniently, and target data are obtained. For instance, acquired medical data can be classified in terms of diseases by the model according to the auxiliary information of the diseases set by the users, and finally decisions are made according to the decision making rules set by the users. It is clear that the decision making method and system based on big data overcome the defect that an existing decision making method is not universal.

The invention discloses a deep belief network-based short text feature optimization and sentiment analysis method. The method comprises the following steps of: 1, obtaining a microblog short text corpus set, a thesaurus, semantic progressive associated words, microblog expression dictionary and a word segmentation model; 2, expanding and reconstructing short texts; 3, carrying out word segmentation and preprocessing on the short texts; 4, constructing a word similarity calculation model; 5, expanding feature vectors of the short texts; 6, self-adaptively extracting an expanded candidate feature set on the basis of a feature depth of a deep belief network; 7, carrying out classified training on the feature set obtained by the deep belief network by utilizing a machine learning classification algorithm so as to obtain a classification and prediction model; and 8, carrying out sentiment annotation on a test data set by utilizing the classification and prediction model. The method is capable of finding potential feature semantic information more effectively and improving the quality of sentiment feature extraction so as to improve the correctness of sentiment classification.

The invention relates to a device and method for constructing road network topology based on bus GNSS space-time tracking data. The device comprises a mobile client and a server. The mobile client comprises a GNSS data acquisition module, a data preprocessing module, a location matching module and a client communication module which are connected in sequence. The server comprises a driving characteristic mining module and a server communication module. The driving characteristic mining module is provided with mining algorithm programs and a database for storing data. A road network map is constructed by analyzing the bus GNSS space-time tracking data, extra hardware equipment is not needed, and the device is not subjected to influence of environmental infrastructure. On-site data collection and survey by the consumption of plenty of manpower and material resources is avoided, the road network map updating cycle is shortened greatly, and further mining of traffic network information parameters and traffic situation prediction are facilitated.

The invention discloses a short-term photovoltaic power prediction method based on VMD-IPSO-GRU, and belongs to the technical field of photovoltaic power generation and grid connection. Firstly, a historical photovoltaic power time sequence is decomposed into sub-sequences with different frequencies through variational mode decomposition, geographic information and component parameters contained in photovoltaic sequence data are fully mined, and signals and noise of original data are separated; secondly, main meteorological factors influencing photovoltaic output are determined through Spearman and Pearson correlation coefficients; and finally, gating cycle unit network models are established for the sub-sequences decomposed by the VMD respectively, and the GRU nerve is optimized through an improved particle swarm algorithm and an adaptive moment estimation algorithm, thereby improving the network convergence rate and the data fitting effect, accurately and efficiently finishing short-term photovoltaic power prediction, and avoiding errors caused by manual parameter adjustment.

The invention discloses an e-commerce website cheat user identification method and system based on a random forest algorithm. The system comprises an e-commerce website user data processing module, an e-commerce website user data storage module, an e-commerce website user data analysis module and a result display module, wherein the e-commerce website user data processing module, the e-commerce website user data storage module, the e-commerce website user data analysis module and the result display module are connected in sequence. The method includes the steps of data collection, data pre-processing, data conversion, data analysis, data display and the like. On the basis of the random forest algorithm in a machine learning method, static data and dynamic data of a user are collected, data mining and data analysis are carried out fast, and not only are completeness, continuousness and effectiveness of information of the user guaranteed, but also a good classification result is obtained.

The invention relates to an output power prediction method based on similarity data selection for a photovoltaic plant, and belongs to the technical field of photovoltaic power generation. The method comprises the following steps: step 1, collecting irradiation intensity values, temperature values and actual photovoltaic output power values of historical days, as well as irradiation intensity values and temperature values of predicted days in weather forecast; step 2, determining weights w1 (i) corresponding to irradiation intensity of all whole points from 6 am to 18 pm every day, and determining weights w2 (i) corresponding to temperature of all whole points from 6 am to 18 pm every day; step 3, performing selection on similar days; step 4, determining weight of power in each similar day during prediction according to the degree of correlation between the similar days and the predicated days; step 5, obtaining a power predication value required in the process that the photovoltaic output is performed in the predicated days through calculation, and performing evaluation on a predicated result. The method can well excavate the correlation between the predicated days and history data, is easy to implement and improves predicated accuracy of the photovoltaic output power.

The invention discloses an exercise semantization method based on a problem solving process. The exercise semantization method mainly comprises the steps of establishing a knowledge network (the relation between knowledge points and the relation between theorems and derivation conditions and conclusions of the theorems), and performing semantization on exercise problems and the problem solving process. According to the exercise semantization method disclosed by the invention, by performing analysis processing on the problem solving process of an exercise, extracting the derivation conditions, causes and the conclusions of problem solving steps and performing the semantization, so that the problem solving process of the exercise has a detailed logical relation, and then, is identified by a computer, and therefore, shared utilization of resources is facilitated.

The invention provides a fault prediction method based on synthetic minority class oversampling and deep learning. The Means method is used for clustering a few types of samples in the sample set; deleting the noise class cluster after clustering; dividing the class cluster into noise class samples in each class cluster by using a KNN method; fault samples and risk samples, deleting the noise samples; and finally, inputting a random number into each class cluster, and selecting a certain sample as an output sample according to a proportional relation between the random number and the fault class sample and the risk class sample in the class cluster;realizing oversampling of the SMOTE method ; and then increasing the number of a few types of samples through multiplication operation, so thatthe types of the samples in the finally obtained fusion sample are more balanced, and the acquired feature data are balanced, thereby facilitating model training, maximally mining the law behind thedata, and realizing a better fault prediction effect.

The invention discloses a speech spectrum characteristic extracting method facing speech emotion identification. The method comprises the first step of framing a speech signal and conducting the fast Fourier transformation to obtain a corresponding speech spectrum, the second step of resolving the speech spectrum, the third step of conducting the central peripheral subtract calculation on the resolved image and normalizing to obtain a characteristic image of each resolving image, the fourth step of extracting the characteristic matrix of each characteristic image and a fifth step of conducting dimensionality reduction on the characteristic matrix and reconstituting. The method comprehensively uses the image processing methods from the perspective of analyzing the speech spectrum characteristic, excavates the emotion identification characteristic from a creative perspective, uses a multiscale and multichannel filter to resolve the speech spectrum, conducts the processing in different characteristic fields, and combines the PCA analysis to better excavate beneficial information to speech emotion.

The invention relates to an image visual characteristic extraction method based on sparse coding and belongs to the technical field of digital image processing of computers. The method comprises the following steps of: extracting low-level characteristics of a picture set; removing labels with extremely low frequency, and generating a label vector; generating a matrix W similar to the low-level characteristic to serve as a basis of manifold constraint, and essentially combining low-level visual characteristics and high-level textual characteristics; establishing a target function; and minimizing the target function, so as to obtain an optimal matrix consisting of sparse coding of the low-level characteristics of the picture set. According to the method, by the adoption of the sparse coding, hidden type information of the low-level visual characteristics and the high-level textual characteristics of an image is well mined, and a model has high robustness; according to the method, a maximization pool method is adopted, and a unique image visual characteristic vector of each picture is obtained; and moreover, the visual characteristics of a final image are simple and effective.

The invention discloses an improved data cleaning method for a stack noise reduction auto-encoder, which comprises the following steps of firstly, introducing an Adam and SGD hybrid algorithm to continuously adjust the network parameters of a stack noise reduction auto-encoder model; secondly, training the normal state data by using the model, obtaining the hidden characteristics of the data, andobtaining a reconstruction error in a normal state; thirdly, using the model for detecting the abnormal state data, analyzing the influence of various types of data on the model according to reconstruction errors of the abnormal state data, and rapidly classifying, cleaning and repairing the dirty data and the abnormal data reflecting the equipment faults. The AS-SDAE can directly and intelligently analyze the monitoring data, can better mine the high-order characteristics hidden in the data, guarantees the high efficiency of cleaning the dirty data, retains the useful data reflecting the abnormal condition of the equipment, and improves the data analysis efficiency.

The invention discloses an MGKFCM (multipath Gauss kernel fuzzy c-means clustering algorithm). The MGKFCM comprises the following steps: 1, performing optimized division on a sample set according to the principle of minimization of an MGKFCM objective function; 2, initializing clustering centers and calculating fuzzy membership and an objective function value 1 with a gradient iterative formula; 3, estimating the clustering centers and calculating the fuzzy membership and an objective function value 2 with PSO (particle swarm optimization); 4, selecting a cluster center with smaller objective function value as an iteration path on the basis of the objective function value 1 and the objective function value 2; 5, obtaining the MGKFCM objective function through calculation. Two Gaussian kernel clustering algorithm iteration paths including a gradient method and the PSO are integrated, the path with the smaller clustering objective function value is taken as the parameter iteration path, optimal performance of the two algorithms is used effectively, and the clustering performance of the clustering algorithm is improved.

The invention provides a ship trajectory prediction method and system based on an LSTM automatic encoder and a bidirectional LSTM, and the method comprises the steps: carrying out the preprocessing ofship AIS trajectory data, and carrying out the feature extraction of the trajectory data through an automatic encoder; then, combining the extracted features with trajectory longitude and latitude data to represent the current navigation state of the ship; and taking the extracted features with trajectory longitude and latitude data as model input, learning a ship motion law implied in the trajectory data through a bidirectional LSTM neural network model containing an attention mechanism, and predicting the position of the ship at the next moment by using the ship motion law learned by the model. According to the method, the ship track prediction is carried out by adopting a scheme of the LSTM automatic encoder, the attention mechanism and the bidirectional LSTM neural network, and the bidirectional LSTM model can better mine the space-time association relationship of the track data on the premise of reserving enough effective information of the ship track data. According to the method, the trajectory prediction precision can be effectively improved, the real-time prediction is realized, and the requirement of a scene with relatively high trajectory prediction timeliness and accuracy is met.