1766 results about "Similarity computation" patented technology

The invention discloses a Chinese question-answering system based on a neural network, which comprises a user interface module, a question word pre-segmentation module, a nerve cell pre-tagging module, a learning and training module, a nerve cell knowledge base module, a semantic block identification module, a question set index module and an answer reasoning module. The system comprises the steps of: firstly adopting an SIE encoding mode to encode the in-vocabulary words of the semantic block according to corresponding position, later converting an identification problem of the question semantic block into a tagging classification problem, and then adopting a classification model based on the neural network to determine the semantic structure of the question, and finally combing the semantic structure of the question to realize the question similarity computation based on the neural network and comparing the weight of various semantic features of the question by extracting the tagged semantic features of the question, thereby providing a basis for final answer reasoning. The Chinese question-answering system integrates the syntax, the semantics and the contextual knowledge of the question and can simulate the process that human beings process the sentence.

An instance-level retrieval method and system are provided. A representation of a query image is embedded in a multi-dimensional space using a learned projection. The projection is learned using category-labeled training data to optimize a classification rate on the training data. The joint learning of the projection and the classifiers improves the computation of similarity / distance between images by embedding them in a subspace where the similarity computation outputs more accurate results. An input query image can thus be used to retrieve similar instances in a database by computing the comparison measure in the embedding space.

A phishing webpage detection method based on Hungary matching algorithm is characterized by firstly extracting the text feature signatures, image feature signatures and general webpage feature signatures of the rendered webpages and more comprehensively depicting the features after access to webpages; and then computing the optimal matching of bipartite graphs by Hungary algorithm to search for the matched feature pairs among different webpage signatures and more objectively measuring the similarity among the webpages on the basis, thereby improving the phishing webpage detection efficiency. The method is also characterized by determining the inside weights of the text features, image features and global image features by utilizing the area under curve and determining the relative weightsamong the text similarity, image similarity and global image similarity during webpage similarity computation by utilizing logarithmic regression analysis. The precision and the recall rate are greatly improved in the method provided by the invention.

The invention provides a method for network equipment identity recognition. The method for the network equipment identity recognition includes the following steps: receiving visit requests sent by network equipment through a webpage browser; searching corresponding webpage codes and collection codes according to an address of a target webpage and returning the codes to the webpage browser; receiving an attribute vector set collected from the network equipment and sent by the collection codes through the request of hyper text transport protocol (HTTP); searching attribute groups matched with the attribute vector set from a preset attribute bank based on the attribute vector set; carrying out similarity computation on the attribute vector set and the searched attribute groups and selecting the attribute group which has high similarity with the attribute vector set; and comparing the highest similarity value with a preset new equipment threshold value and an old equipment threshold value and determining the type of the network equipment according to comparison results. The invention further provides a network equipment identity recognition system for realizing the method. The method and the system for network equipment identity recognition can reduce occupation on network equipment resources, and improve visit speed and recognition accuracy.

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 invention discloses a user credibility authentication system and method based on user behaviors. The system comprises a user identity and behavior model determination module, a user behavior collection module, a user behavior mining module, a user behavior sequence matching and credibility authentication module and a local security policy module. The method comprises the steps as follows: the user behavior mining module performs behavior mining on user behavior logs; a personal user behavior characteristic sequence is established; the user behavior sequence matching and credibility authentication module performs sequence similarity calculation on the personal user behavior characteristic sequence and a to-be-matched sequence, so that a user credibility grade is obtained, and a corresponding security policy is started using. According to the invention, a sequence pattern mining manner is adopted to collect real-time user behaviors so as to establish the to-be-matched sequence, the similarity matching of the behavior sequence is performed, accordingly, the real-time user behaviors are subjected to credibility authentication, the accurate rate of authentication is improved, and enterprise and personal property safety is guaranteed.

The present invention discloses an object-based image search system and method, whereby a user may visually specify query objects for retrieving relevant images without using any image-segmenting software tool. The method of the present invention comprises: specifying target feature points from sample images which are displayed in an image query interface and determining logic relationships of the target feature points; utilizing a feature-point checking program to correlate the target feature points with target objects; utilizing a feature-similarity calculating program to find out the images relevant to the target objects via the image feature parameters stored in a feature database; arranging and presenting the relevant images in sequence.

The invention discloses a phishing website detection method based on uniform resource locator (URL) classification. Firstly, a modeling is performed for URL characteristics, a method specific to a domain name imitation phenomenon of the characteristics is provided for calculating the similarity between a suspicious domain name and a protected domain name by a dynamic programming thought, in order to collect phishing URL high frequency suspicious character features, a suspicious character extraction algorithm based on a generalized suffix tree is provided, then on the basis of characteristic modeling, a support vector machine (SVM) algorithm is utilized to perform classified training for experimental training set, a SVM classification model is obtained after the training, the SVM classification model is used for classifying the URL to be detected, and a server for detecting a phishing website updates the current SVM classification model according to a specific online incremental learning strategy.

The invention discloses a method and equipment for constructing an intelligent question-answering system through a question generation data set. The method comprises the following steps: constructinga tourism domain knowledge graph; performing question analysis on natural language questions proposed by a user, performing word segmentation and word vector training on the questions, and in a word segmentation process, using a jaeba tool and adding a tourism domain dictionary set in advance; performing entity extraction on the natural language questions by using a Bert-BiLSTM+CRF model; matchingthe extracted entities with entities in a knowledge graph; if a matched entity exists in the knowledge graph, selecting the entity; if no matched entity exists in the knowledge graph, performing semantic similarity calculation, and selecting the closest entity; matching the selected entities and attributes with triples in a knowledge graph; and returning the corresponding attribute value as an answer to the question to be provided for the user. The invention also provides a device for realizing the method, a terminal and a readable storage medium. According to the invention, the information required by the user can be returned conveniently and accurately.

The invention discloses a collaborative filtering recommendation method for integrating time contextual information, which is used for integrating the time contextual information on the basis of an original item-based collaborative filtering recommendation algorithm and an original user-based collaborative filtering recommendation algorithm and combining the original item-based collaborative filtering recommendation algorithm and the original user-based collaborative filtering recommendation algorithm into a uniform algorithm. The collaborative filtering recommendation method comprises the steps of for the user-based collaborative filtering recommendation algorithm, firstly, integrating a time attenuation function in a user similarity calculation stage; then, clustering items, and training interest attenuation factors of a user on an article category; finally, integrating the time attenuation function in a rating prediction stage, wherein for the item-based collaborative filtering recommendation algorithm, the process is similar to the process of the user-based collaborative filtering recommendation algorithm, and the two algorithms can be finally combined into the uniform algorithm. According to the collaborative filtering recommendation method disclosed by the invention, the time attenuation function is introduced in both the similarity computation stage and the rating prediction stage, different time attenuation factors are used for different types of items by different users, and thus the prediction accuracy can be effectively increased.

The invention discloses an image matching algorithm of bonding point characteristic and line characteristic descriptors. The method comprises the following steps: (1) carrying out angle point extraction on a template drawing and a real-time drawing under multiscale; (2) acquiring an edge set surrounding angle points of the real-time drawing and the template drawing; (3) calculating a class ORB point characteristic descriptor of real-time drawing and template drawing angle points which are acquired from the step 1 and are selected finally; (4) using a minimum cut square Hausdorff distance to describe a matching similarity of the real-time drawing and template drawing edge set acquired from the step 2; (5) calculating a matching similarity of the class ORB point characteristic descriptor of the real-time drawing and template drawing angle points, wherein the characteristic descriptor is acquired from the step 3; (6) matching result integration. In the method of the invention, firstly, a stable point characteristic is used to carry out primary selection on the angle points so as to acquire a candidate point set and a correct position is included; then a global line characteristic is used to screen the candidate point set so that a repetition mode can be reduced and a correct rate is increased.

A similarity calculation unit calculates a similarity between a hand candidate area image and a template image. A consistency probability calculation unit and an inconsistency probability calculation unit use probability distributions of similarities of a case where hand shapes of the template image and the hand candidate area image are consistent with each other and a case where they are not consistent, and calculate a consistency probability and an inconsistency probability of hand shapes between each of the template images and the hand candidate area image. A hand shape determination unit determines a hand shape most similar to the hand candidate area image based on the consistency probability and the inconsistency probability calculated for each hand shape, and outputs it as a recognition result.

The invention discloses an urban traffic illegal behavior detection method based on a video monitoring system. The urban traffic illegal behavior detection method based on the video monitoring system includes the following steps of trajectory extraction, trajectory structuring, trajectory similarity calculation, trajectory clustering and modeling and abnormality detection, wherein in the trajectory extraction step, a video movement target is detected and tracked to extract a target trajectory; in the trajectory structuring step, a trajectory section is segmented and structured, and the trajectory section is represented through four structural characteristics; in the trajectory similarity calculation step, the characteristic distances corresponding to the four structural characteristics of the trajectory section are calculated respectively, and the similarity between trajectories is calculated through weighing and calculation of the relative similarity between the trajectories; in the trajectory clustering and modeling step, a similarity matrix is structured according to the similarity between the trajectories, the trajectories are clustered, the clustered trajectories are built into Gaussian model sets, and the trajectories belonging to the same class are built into one same set of Gaussian models; in the abnormality detection step, the probability of a trajectory belonging to each model is calculated, and abnormality is judged according to whether the largest probability is larger than a preset threshold or not. According to the method, traffic illegal behaviors are detected based on the video monitoring system, and the efficiency and the accuracy of detection and the illegal behavior class are improved.

The invention discloses an individualized recommendation method based on a knowledge map, and belongs to the technical field of knowledge maps and machine learning. The method comprises the followingsteps: 1, carrying out vectorization on goods in the knowledge map so as to acquire a vector set D and quantized values of each goods; 2, calculating goods semantic similarity between the objects according to the quantized values acquired based on the knowledge map; 3, calculating goods interaction similarity between the goods in user historical interaction data based on user behaviors; 4, calculating goods fusion similarity of the all goods according to the goods semantic similarity and the goods interaction similarity; and 5, scoring goods which are not evaluated according to the goods fusion similarity, and generating recommended lists for users according to the scores. According to the individualized recommendation method based on the knowledge map disclosed by the invention, through combination of the goods semantic similarity based on the knowledge map and the goods fusion similarity based on the user behaviors, the recommendation effect of a recommendation system is improved.

An image management device acquires an image group with an image acquisition unit, extracts objects and feature amounts from each image in the image group with an object detection unit, and sorts the objects into relevant clusters with an object sorting unit. Next, a similarity calculation unit calculates a similarity between the feature amounts of each object and each relevant cluster, a co-occurrence information generation unit finds co-occurrence information for each cluster, and then an accuracy calculation unit and an evaluation value calculation unit find an evaluation value for each object with respect to each cluster from the similarity and co-occurrence information. An object priority evaluation unit evaluates the object priority of each object with the evaluation value, and an image priority evaluation unit evaluates the priority of each image from the object priority.

The invention discloses a friend recommendation system based on user sign-in similarity. The friend recommendation system comprises an interest recommendation module, a similarity computation module and a theme extraction module. Firstly, sign-in records of previous users are obtained from an LBSN (Location Based Social Network) database through the theme extraction module, and a potential theme in the extracted sign-in records of the users is obtained through a theme extraction algorithm; secondly, by using the potential theme extracted through the theme extraction module, similarity of each user in a candidate user set and a request user under each theme is respectively calculated through calling a computing method of the similarity computation module; thirdly, summing the similarity of each user under each theme so as to obtain a final similarity; lastly, determining a final recommendation friend by the interest recommendation module through request parameter setting of the request user, and returning to the request user.

The invention discloses a semantic retrieval method oriented to field data. Semantic information of a certain field is described by using a body, so that a computer understands content inquired by a user more accurately, and retrieval on a semantic level is realized. A semantic retrieval model oriented to field data is provided in combination with the structural characteristics of a semantic network body on the basis of analyzing the prior art. The semantic relativity between a keyword and different examples is deduced by using the model according to the position of the keyword in the body, and a relativity weight is written into an example inverted list which corresponds to the keyword, so that the similarity computation complexity during semantic retrieval is lowered, and the retrieval efficiency is increased. According to the method disclosed by the invention, retrieval is raised from a grammatical level of simple keyword matching to a semantic level understandable for the computer, and the computer can understand the input keyword, so that retrieval intelligence is realized.

The invention discloses a web service clustering method based on labels. The web service clustering method comprises the following steps: 1) collecting WSDL (Web Services Description Language) files and label information of web services on the internet; 2) extracting characteristic values of the web services from the WSDL files, wherein the characteristic values comprise contents, types, messages, ports and service names; 3) carrying out similarity computation on the characteristic values and the label information of the web services, and computing the comprehensive similarity according to the characteristic values and the label information; and 4) clustering the web services by using an WTCluster algorithm according to the comprehensive similarity, wherein more accurate clustering results can be provided by using the WSDL files and the label information in combination with the WTClusterweb service clustering method in the prior art. The optimal mixture ratio can be adjusted for data types with different characteristics by adjusting system parameters lambda, omega 1, omega 2, omega 3, omega 4 and omega 5, and two label recommending methods are proposed for solving the problem of excessively few service labels on the internet, so that the clustering effect of the WTCluster algorithm can be improved by using the labels.

The invention discloses an ES-based electronic medical record retrieval method, which relates to the technical field of medical data retrieval. This method introduces semantic analysis model into electronic medical record analysis, including the extraction of subject words and the calculation of semantic similarity, taking advantage of their advantages in text semantic mining, This paper providesthe algorithm support for the latent semantic mining of text information in electronic medical record retrieval by establishing the general medical semantic database (negative word, synonym, ambiguousword), It realizes the high accuracy and recall rate of information retrieval, better adapts to the medical terminology compared with the common natural language is often more complex and constantlychanging, and medical abbreviations, synonyms and polysemous words more characteristics. It meets the scientific research needs of multi-dimensional combined retrieval and the needs of full-text retrieval of related literatures based on latent semantic search. Realize intelligent full-text retrieval with semantic extension and semantic connotation extension in real sense.

The invention discloses an interest and network structure double-cohesion social network community discovering method which comprises the steps of: firstly, archiving content issued by users in a social network, extracting interest characteristics of each user by using an existing interest characteristic extraction method, and further obtaining interest characteristic collection of each user relationship by adopting intersection operation to form a social network R-C model; on the basis, calculating interest characteristic similarity of two user relationships having two common users by adopting an existing similarity calculation method; then, forming a social network weighted undirected graph by regarding a user relationship in the R-C model as a node, regarding whether a common friend exists between two user relationships as a border, and regarding the interest characteristic similarity among the user relationships as a weight value of the border; then, excavating user relationship community by adopting an existing weighted undirected network community discovering algorithm; finally, mapping the user relationship in the user relationship community into associated users directly to form a social network user community.

An instance-level retrieval method and system are provided. A representation of a query image is embedded in a multi-dimensional space using a learned projection. The projection is learned using category-labeled training data to optimize a classification rate on the training data. The joint learning of the projection and the classifiers improves the computation of similarity / distance between images by embedding them in a subspace where the similarity computation outputs more accurate results. An input query image can thus be used to retrieve similar instances in a database by computing the comparison measure in the embedding space.

An object of the present invention is to provide an image retrieval system for computing similarity in the case when components are different between an image feature vector of a query image and the image feature vectors of images to be retrieved when retrieval is carried out by using features of images. A first image feature vector conversion device converts a query image feature vector into a first image feature vector for use in a similarity computation according to similarity computation image feature vector configuration information describing a given configuration of an image feature vector for use in a similarity computation. A second image feature vector conversion device converts an images-to-be-retrieved feature vector into a second image feature vector for use in a similarity computation according to the similarity computation image feature vector configuration information. The image feature vector similarity computation devices compare the first image feature vector with the second image feature vector according to the similarity computation image feature vector configuration information, and compute the similarity.

The invention provides a semantic query expansion algorithm based on emergency ontology. A multilayer emergency ontology model is designed from the angle of ontology, semantic relation among concepts in an emergency field is defined, and accordingly semantics-related concepts can be expanded. Factors affecting the degree of similarity of the concepts are analyzed, a concept similarity computation model with comprehensive consideration in terms of semantic distance, layer factors and coincidence degree of upperseat concepts is created, similarity among concept nodes in an ontology network is more comprehensively quantified, selected concepts can be expanded according to the similarity, homogenization of expansion results is avoided, and final inquiry results can be orderly arrayed according to the similarity. In addition, the semantic query expansion algorithm further expands a participle word bank used for the ontology based on the field of emergency.

Method, system, and programs for computing similarity. Input data is first received from one or more data sources and then analyzed to obtain an input feature vector that characterizes the input data. An index is then generated based on the input feature vector and is used to archive the input data, where the value of the index is computed based on an improved Johnson-Lindenstrass transformation (FJLT) process. With the improved FJLT process, first, the sign of each feature in the input feature vector is randomly flipped to obtain a flipped vector. A Hadamard transformation is then applied to the flipped vector to obtain a transformed vector. An inner product between the transformed vector and a sparse vector is then computed to obtain a base vector, based on which the value of the index is determined.

The invention discloses a cross-modal retrieval algorithm based on mixed hypergraph learning in subspace, based on the cross-modal public subspace learning of canonical correlation analysis. The similarities inside and among modals are calculated by a mapping of the public subspace. A mixed relation matrix is calculated by the similarities inside and among the modals. A mixed hypergraph model is built by the extraction of the relation matrix. And the cross-modal retrieval and sample sequencing are conducted by adoption of hypergraph learning finally. With aiming at cross-modal heterogeneous variations and high order relation among samples, the algorithm instance applies the hypergraph model combined with the cross-modal public subspace learning to the cross-modal retrieval, so that the model is capable of considering the similarity among the modals and the similarity inside the modals simultaneously and giving consideration to the high order relation among a plurality of the samples meanwhile, improving the final precision ratio and the final recall ratio of the cross-modal retrieval. The algorithm is capable of effectively improving the performance of the cross-modal retrieval and greatly enhancing the precision ratio and the recall ratio of the cross-modal retrieval.

The invention discloses a perception based insulator recognition and defect diagnosis method. The method comprise the following steps: step (1), an electric transmission line knowledge model is adopted for perceiving multiple parallel line structures of an insulator; step (2), an area where the insulator is located is recognized, an overlapped output area is detected as follows: an overlapped area is combined, and a potential insulator area is corrected to a rectangle in any angle; and step (3), insulator chip-missing defect diagnosis based on blocking characteristic quantity is performed, and the glass insulator self-destruction chip-missing defect is diagnosed by adopting two steps of insulator area blocking and interblock similarity computation. The perception based insulator recognition and defect diagnosis method has the benefits that influence of background on recognition of the insulator area can be effectively precluded, and the accurate rate of insulator defect recognition is greatly improved.

The invention relates to a word sense disambiguation method and device based on a word vector. The method comprises the following steps that: data preprocessing: carrying out processing, including punctuation removal, word segmentation and the like, on a document and a semantic item; training the word vector: using a word vector training tool to train the word vector; carrying out context vector representation: obtaining the word vector, and adopting a local weighting method to calculate the context vector; carrying out semantic item vector representation: obtaining the word vector of each word of the semantic item, and carrying out calculation to obtain a semantic item vector; carrying out similarity calculation: calculating a cosine similarity between the context vector and each semanticitem vector; carrying out semantic item distribution frequency calculation: carrying out statistics on the distribution frequency of each semantic item of an ambiguous term in a dataset; and carryingout final score statistics: calculating the cosine similarity between the context and each semantic item and the comprehensive score of each piece of semantic item frequency, wherein the semantic item with a highest score is an optimal word meaning.

This invention relates to a computer information search system based on the understanding of natural languages and a search method, in which, the search is started by the interrogative sentence input by a user and the system outputs a sequential answer according to the related program of the semantic meaning, first of all, articles from the internet and data from the content database are processed by the HNC sentence analysis module to get a being selected solution sentence repository with labels, then the interrogative sentences input by the user is processed by the HNC analysis module to get the HNC structure to enter into the interrogative analysis module for analysis to generate an equal semantic target sentence mode sequence then concept similarity computation is carried out to the words and expression blocks in the being selected solution sentence and target solution sentence mode in the repository by a sentence mode matching module to compare the being selected and target sentences to get the marks of accuracy for the result of the sentence mode match, the semantic relation structure identification match result and the solution to array in terms of the correctness and feeds back the result.

The invention provides a Storm-based real-time recommendation method and a system therefor. The real-time recommendation method comprises the steps of: A, performing offline similarity calculation on video data and behavior data of users separately to obtain a similarity model of video, and performing statistical analysis on the behavior data to obtain the statistical data of user behavior; B, performing user modeling according to the video data in the Storm end to generate interest vectors and combining the interest vectors and the similarity model to obtain individualized recommendation results based on the user behavior; C, integrating the statistical data, the similarity model and the individualized recommendation results to obtain a recommendation list. The method combines similarity and interest, thereby improving the accuracy of individualized analysis of user preferences and guaranteeing the recommendation quality.

The invention relates to the technical field of mode recognition, and in particular to an attribute learning and Riemannian manifold combined pedestrian re-recognition method. According to the method,a model with stronger generalization ability is disclosed through researching representations and semantic attributes of pedestrians; and through importing pedestrian attribute labels, using a convolutional neural network model and using multi-target loss functions, the model not only needs to predict pedestrian IDs, but also needs to predict various correct pedestrian attributes. Pedestrian pictures are converted into output tensors of a convolution layer through the trained convolutional neural network, so as to respectively calculate a covariance descriptor of each pedestrian picture. Thecovariance descriptors are utilized to carry out feature fusion so as to eliminate feature redundancy, so that measurements of Riemannian manifolds where the descriptors are located are researched andthen more correct similarity calculation is realized.