52results about How to "Avoid sparsity" patented technology

The invention relates to the field of internet electronic commerce, in particular to a commodity similarity calculation method and a commodity recommending system based on image similarity. The method includes: preprocessing a target image, to be specific, removing image differences caused by changes in light conditions such as brightness and chromatic aberration; processing the target image to detect a foreground frame; converting a community image in the foreground frame into pixel images different in scale by means of bilinear interpolation, and acquiring attribute features, in different dimensions, of the commodity image in the foreground frame under different scales; calculating attribute feature similarities, under different scales, between an attribute feature vector of the commodity image in the foreground frame and an attribute feature vector of a commodity sample image; according to a decision forest model and the attribute feature similarities under different scales, calculating commodity image similarities, under the pixel images of different scales, between the commodity image in the foreground frame and the commodity sample image; using the commodity image as a uniform identifier of a commodity on different commercial platforms. The commodity similarity calculation method and the commodity recommending system have the advantage that reliability of the system is greatly improved.

The invention discloses an emotion analysis method which aims at network microblogs. The emotion analysis method which aims at the problem that in the traditional emotion analysis, expression of diversification emotion of emergencies is not enough is based on a multi-dimensional vector representation model. The emotion analysis method comprises extracting emotional words in combination with a clinical psychological emotion detection table and initializing emotion analysis vectors; automatically finding the network new words in the microblogs through a method which is based on large-scale corpuses and statistics, automatically absorbing the network new words and emotional identifiers which can express emotion, continuously expanding the emotion vectors and setting a gradational structure of the emotion vectors; performing multi-dimensional emotion analysis and timing sequence analysis on the microblogs based on an existing vector model.

The present invention discloses a personalized webpage recommendation method based on a topic and a relative entropy. According to the method, firstly, an LDA (latent dirichlet allocation) model is adopted to carry out topic mining on webpage content and user reading behaviors and to calculate a webpage semantic feature vector and a user interest feature vector based on the topic; and then a similarity measuring formula based on the concept of the relative entropy is utilized to calculate similarity between a webpage-to-be-recommended semantic feature vector and the user interest feature vector, and the obtained similarity is used as a decision basis for personalized webpage recommendation. According to the personalized webpage recommendation method based on the topic, a great deal of computing cost based on a collaborative filtering method is avoided; and meanwhile, the topic, instead of a keyword, is adopted to represent webpage content, and thus, the recommendation process and the recommendation results can more comprehensively and accurately reflect conceal information and deep semantic features of the webpage content.

The invention discloses a three-dimensional point cloud model classification method based on convolution neural network, includes selecting Princeton ModelNet to generate training set and data set from training data and test data by selecting required number of models from official website according to ModelNet 10 and ModelNet 40 respectively, selecting training data and test data from official website according to Princeton ModelNet, selecting Princeton ModelNet to generate training set and data set according to model Net 10 and ModelNet 40 respectively, and selecting Princeton ModelNet to generate training data and test data. 2, carry out feature analysis on that point cloud model and constructing a classification framework; S3, ordering the point cloud; S4, two-dimensional visualizing the ordered point cloud data; S5, Constructing CNN network for two-dimensional point cloud image. The invention applies the CNN in the image field directly to the classification of the three-dimensional point cloud model for the first time, 93.97% and 89.75% classification accuracy were obtained on ModelNet 10 and ModelNet 40 respectively, Experimental results show that it is feasible to classify 3D point cloud model by using CNN in image domain. PCI2CNN proposed in this paper can capture 3D feature information of point cloud model effectively and is suitable for classification of 3D point cloud model.

The invention relates to a specific group discovery and expansion method based on microblog data, and belongs to the field of social network analysis and data mining. The specific group discovery and expansion method comprises the following specific steps: collecting relevant group information; carrying out information integration and mapping; aiming at text data to carry out characteristic extraction; calculating a user similarity degree; carrying out the self-detection of a category group; and extracting the attributes of the specific group, judging a category, and carrying out group expansion. The specific group discovery and expansion method artfully avoids the problem that group identification can not be carried out since data is sparse or incomplete when a network model is used, inputs large-scale data calculation and is high in stability.

The invention discloses a POI recommendation method combining travel interest and social preference, and the method comprises the steps: learning a user travel behavior according to the historical POIdata distribution of a user in an LBSN, and predicting a POI accessed by the user in the future travel according to the current position; Constructing social contact associated interest similarity byextracting theme vectors; Constructing a heterogeneous travel information network, and establishing interest similarity of travel behaviors; Determining a similar group by integrating social interestsimilarity analysis and travel behavior similarity analysis; Generating a candidate POI set by combining the predicted POI of the future travel access of the user and the similar groups of the user,and discovering TOP-N POI that users are most likely to go to by the calculation. According to the method, the similar groups of the user are discovered by utilizing social interest and travel preferences while position prediction is considered, more proper interest point recommendation can be comprehensively provided for the user by utilizing the similar groups instead of friend users, and the problem of data sparsity in the LBSN is relieved, so that the recommendation effect can be better improved.

Efficient exploration of natural language conversations associated with dialogue policy learning may be performed using probabilistic distributions. Exploration may comprise identifying key terms associated with the received natural language input utilizing the structured representation. Identifying key terms may include converting raw text of the received natural language input into a structured representation. Exploration may also comprise mapping at least one of the key terms to an action to be performed by the computer system in response to receiving natural language input associated with the at least one key term. Mapping may then be performed using a probabilistic distribution. The action may then be performed by the computer system. A replay buffer may also be utilized by the computer system to track what has occurred in previous conversations. The replay buffer may then be pre-filled with one or more successful dialogues to jumpstart exploration.

The invention discloses a multi-target random dynamic economic dispatch method based on scenario decoupling and asynchronous iteration. The method herein includes the following steps: 1. assigning relevant computing parameters; 2. establishing a multi-target random dynamic economic dispatch model; 3. using scenario decoupling and asynchronous iteration to improve the interior point method to resolve the multi-target random dynamic economic dispatch model. According to the invention, the method uses the scenario method translates the problem of multi-target random dynamic economic dispatch to the problem of large-scale multi-target deterministic dynamic economic dispatch, translates the problem of large-scale multi-target deterministic dynamic economic dispatch to the problem of a series of large-scale single object non-linear planning by means of the normal boundary cross method, conducts resolution by using the nonlinear primal-dual interior point algorithm, and avoids the generation of dense matrix, such that the matrix in the entire computing process is sparse, is better compatible with economy and environmental protection in operating a power grid, and therefore is a dispatch plan with higher operation benefits.

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 belongs to the technical field of recommendation systems, and relates to a heterogeneous graph neural network-based recommendation method, which comprises the following steps of: collecting a data set with social relationships among users, user-commodity interaction historical data and commodity category information, filtering invalid data and carrying out negative sampling; randomly selecting a user set and a related commodity set, and carrying out multi-order graph sampling and mapping; node feature extraction: inputting the constructed graph into a heterogeneous graph neural network for processing to obtain a fusion node embedding vector of the nodes, wherein for the commodity nodes which do not need to be subjected to the re-calibration step, the fusion node embedding vector of the commodity nodes is the commodity fusion embedding vector; re-calibration: re-calibrating the user fusion node embedding vector to obtain a user final expression embedding vector; and performing preference prediction by using the user final representation embedding vector and the commodity fusion embedding vector, and obtaining a recommendation sequence. The method solves the problems of data sparsity and data missing, and has the advantages of being accurate in recommendation and the like.

The invention provides a food and health knowledge graph construction method based on deep learning. A knowledge graph oriented to the field of food safety is constructed and applied from six aspects of information extraction, knowledge representation, knowledge fusion, knowledge storage, knowledge reasoning and knowledge graph application. The functions of efficiently querying food safety data and scientifically analyzing food safety problems are realized. In the information extraction stage, a deep learning method based on manual annotation data set application comprises the steps that entity recognition is achieved based on a BiLSTM-CRF model, and relation extraction is achieved based on a Transformer model. On the basis, a triple type of (entities, relationships and entities) is adopted as input of knowledge graph representation learning, and high-dimensional knowledge is subjected to Embedding through representation learning, so that the data sparsity is effectively solved, the calculation efficiency is improved, and the method can be applied to entity similarity calculation and relationship prediction.

The invention provides a collaborative filtering recommendation method based on elastic dimensional feature vector optimized extraction, and belongs to the technical field of Internet information recommendation. The recommendation method is constructed by using user feature vectors and recommendation object feature vectors, and dimensions in which a user is interested and to which a recommendation object really belongs in each user feature vector and each recommendation object feature vector are elastically obtained respectively by using user assistant vectors and recommendation object assistant vectors. With no professional knowledge and individual information, the collaborative filtering recommendation method is secure and simple; the minimum root-mean-square error is adopted as an optimization constrain condition; in an implementing process, only existing parts in a rating matrix are constrained, but a correct fitting mark can be also made, and the problems of data sparseness and cold starting caused by lack of historical data are solved. The method can be used for obtaining the dimensions which really work in each user feature vector and each recommendation object feature vector, and adaptively adjusting the search direction, so that overfitting of the recommendation method is avoided, and a recommendation result is optimized.

The invention provides a multiple address access method and device for signals in mobile communication. The method comprises the following steps: A1, in a mobile communication system, obtaining a low-density sparse parity-check matrix based on J users, a business data length m and the number of subcarriers K; A2, based on the low-density sparse parity-check matrix, carrying out LDSM extension on m channel coding symbols of each user, so as to obtain m spread spectrum vectors corresponding to each user; and A3, based on the J users, superposing the m spread spectrum vectors for each user on mK carrier resources to obtain J user signals, and sending the J user signals to a receiving end. The method and device provided by the invention has the advantages that the problem that short cycles exist in the traditional SCMA parity-check matrix when the parity-check matrix lacks sparseness due to small dimensions is solved, and an LDSM parity-check matrix with an irregular distribution is designed for the near-far effect, so that the problem of equality is solved, unequal protection for edge users is achieved, and the detection performance and the network transmission quality are improved.

The invention discloses a self-adaptive degree-of-freedom electromagnetic temperature multi-physical field coupling analysis method. The method comprises the steps of establishing an electrical equipment geometric model and initial unstructured grid discretization; establishing an electromagnetic temperature multi-physical field weak coupling numerical model; analyzing finite element calculation format derivation based on an electromagnetic temperature multi-physical-field weak coupling numerical model of an unstructured grid unit; calculating and solving an electromagnetic temperature multi-physical field in the electrician equipment, and carrying out error analysis on a numerical solution; and adjusting the degree of freedom of each physical field, and solving again according to the adjustment result until the precision of the numerical solution meets the requirement to finish analysis. According to the method, independent and flexible adjustment of the freedom degrees of two physical fields is achieved on one set of grids, different requirements of the physical fields for discretization are met with small computing resources, grid sparsity and subsequent complex operation of actual operation are avoided, a grid mapping function and errors possibly introduced by the grid mapping function are prevented from being used, and the calculation efficiency of electromagnetic temperature coupling analysis calculation is effectively improved.

The invention discloses a deep learning-based user interest point recommendation method and system. The method comprises the steps of obtaining historical sign-in data of a user; training a deep learning model based on the historical sign-in data; inputting the latest sign-in data of the user into the trained deep learning model, and outputting predicted user interest points, wherein the deep learning model automatically extracts a POI category preference feature and a POI preference feature of a user, and the two features are expressed as two features Embedding; then, performing Euclidean distance calculation on the two kinds of characteristics Embedding and a candidate set POI Embedding, sorting through scores, and outputting POIs ranked as the top N, wherein N is a set value. Accordingto the invention, the situation that the recommendation accuracy can be influenced by the retrieval space with a huge POI corpus is considered, and the filtering module is linked after the preferenceencoder of the POI category, so that the POI retrieval space is reduced, the CatDM retrieval difficulty is reduced, and meanwhile, the recommendation accuracy is improved.

The invention discloses a comment recommendation system and method based on deep learning, and belongs to the technical field of information dissemination. The problem that an existing recommendationsystem only gives recommendation scores, and consequently the recommendation interpretation capacity is poor is mainly solved. According to the system, comment text feature representation is acquiredby using a character and word-level bidirectional recurrent neural network; acquiring image features by using a convolutional neural network; obtaining attribute feature representation by using a decomposition machine neural network; and a memory mechanism is used for solving the cold start problem of users and products in the recommendation system; and obtaining a relationship between the user and the product by using a bilinear tensor mechanism to jointly generate scores and comments. The method is mainly used for comment recommendation in information dissemination.

The invention provides a collaborative filtering method, a collaborative filtering device and a collaborative filtering system, and belongs to the technical field of house information processing. Themethod comprises the following steps: determining a first house resource set, and forming a first preference data set corresponding to a selected user from preference data of house resources in the first house resource set; determining a position area range, selecting a part of users according to the position area range, determining a second house resource set, and forming a second preference dataset corresponding to the part of users from the preference data of the house resources in the second house resource set; and obtaining a trained vector decomposition model through the second preference data set, obtaining a feature vector set corresponding to the house resources in the second house resource set by utilizing the second preference data set and the trained vector decomposition model, calculating the similarity of the feature vector set, and forming a similarity set after the calculation is completed. The method and system are used for determining the recommended house resourceswith the user preference characteristics through similarity calculation.

The invention relates to the technical field of data processing, in particular to a knowledge base construction method and device, electronic equipment and a readable storage medium. The method comprises the steps of obtaining a first knowledge base of a target domain, and performing data enhancement processing on the first knowledge base; wherein the first knowledge base is determined according to the video data sample in the target domain; wherein the data enhancement processing comprises the step of adopting a word frequency inverse document frequency algorithm; and determining the first knowledge base after the data enhancement processing as a target knowledge base. By adopting the knowledge base construction method provided by the invention, sparsity of video training data is avoidedthrough enhancement processing on the video data knowledge base, and the knowledge base construction method is particularly suitable for construction of the knowledge base with main data such as education as video data, so that a knowledge spectrogram of a target field taking the video data as a main part is established; and an effective knowledge base is provided for downstream video recommendation by utilizing the knowledge spectrogram.

The invention relates to a preprocessing optimization method for spatial line features of a crowdsourcing fragment map. The preprocessing optimization method comprises the steps of sorting shape points in a single lane line in road fragment data according to coordinate sizes or mileage; performing bevel filtering on a lane line formed by the shape points; smoothing a curve of a lane line formed bythe shape points; and breaking the lane lines formed by the shape points according to the distances between the adjacent shape points. The method is used as a pre-process of a crowdsourcing fusion processing process to pre-process a fragmented map collected by crowdsourcing, so that the processed data meets the requirements of subsequent fusion optimization.

The invention discloses a collaborative filtering scheme recommendation method fusing local similarity and global similarity, and belongs to the technical field of scheme recommendation. Based on a collaborative filtering algorithm, the collaborative filtering scheme recommendation method makes a breakthrough in the aspect of similarity calculation, including: firstly, calculating global similarity by adopting all user project scores, effectively utilizing the data information, being able to solve data sparsity; and secondly, dividing the scores into positive scores, negative scores and combination of the two scores by the aid of common score information of the users, calculating the local similarity, so that common information between the users can be fully and effectively utilized, and the two kinds of similarity are fused, and prediction accuracy can be effectively improved.

The invention relates to the technical field of network data search, and particularly provides an intelligent customer service system constructed based on a knowledge graph, which constructs a knowledge graph for questions and answer files by determining more accurate positions of question and answer pairs by utilizing characteristics of the question and answer pairs. Besides the constructed knowledge graph, the invention further provides a new method for effectively using the knowledge graph; the characteristics of the question and answer document are composed of questions and answers and used for making two dimensions of the map meaningful, the knowledge graph is expanded in the transverse direction and the vertical direction, especially in the vertical expansion period; the structure ofa subsequent layer is kept stable, and a merging mechanism is provided to avoid sparsity. And the LabelSOM selects the feature words of each neuron for navigation, and extracts a typical QA documentto enable a user to quickly understand all contents.

The invention provides a method for detecting a drivable area of an automatic driving vehicle, and the method comprises the steps: obtaining original point cloud information through employing a laserradar, and carrying out the coordinate conversion; longitudinally layering the point cloud information to filter out ground point cloud and impurities; longitudinally increasing the virtual point cloud according to the obtained point cloud distance resolution to obtain second point cloud information; mapping the point cloud into a polar coordinate grid, judging whether the point cloud exists in the grid or not, if not, adding a virtual point as point cloud data, and selecting a point, closest to the vehicle, in the grid as a candidate point. According to the method, the detection of the drivable area can be realized under the condition of weak light, and meanwhile, the obtained boundaries are all reliable boundaries.

The invention discloses an article recommendation method and device and a storage medium, and belongs to the field of information recommendation. The method comprises the following steps: determining first prediction scores of a plurality of articles through a collaborative filtering model according to article behavior data of a target user; determining k first articles from the plurality of articles according to the first prediction scores of the plurality of articles; determining first entity vectors of the k first articles according to the k first articles and the knowledge graph vector set; determining a second prediction score of the k first articles according to the first entity vectors of the k first articles and the article behavior data of the target user; and if a mean square error between the first prediction scores and the second prediction scores of the k first articles is smaller than or equal to an error threshold value, recommending the k first articles to the target user. According to the recommendation algorithm, on the basis of determining the similarity of the articles according to the behavior data of the articles, the similarity among the article attributes is fully considered, and the recommendation accuracy is further improved.

The invention discloses a method and a device for recommending tourist attractions, which relate to the technical field of recommendation. The method and the device are used to solve tourist attraction recommendation problems. The method comprises: a set of attraction images acquired in the attraction is determined as a target domain image, the style image set obtained from the Internet search engine by using the feature style keywords of the scenic spots is taken as the auxiliary domain image, The distribution difference between the target domain image and the auxiliary domain image is expressed by the maximum mean difference, and the objective function of the optimal image classification is determined according to the maximum mean difference function and Laplace support vector machine, and the image style of the scenic spots is obtained. According to the style proportion of all the images, if the style proportion exceeds the threshold, it is determined to contain the style of the scenic spots; The user preference model is established according to the model based on explicit interaction determination and the model based on implicit mining determination. User preference models andattraction styles determine the list of tourist attraction recommendations through the cosine distance formula.

The invention relates to a security detection method based on an Ethernet IP industrial control protocol. The detection method comprises the following steps: 1, establishing a network data packet capture module; 2, establishing a data analysis module; 3, establishing a data learning module; 4, establishing a behavior legality judgment module; 5, establishing a behavior abnormality response module.According to the technical scheme, related commands and function calls are deeply analyzed according to protocol characteristics of Ethernet IP, and analysis results can be directly processed by somevisual tools, so that a user can have intuitive feeling on circulating content in an industrial control network.

The invention belongs to the technical field of data mining, and discloses an acupuncture clinical data preprocessing control system and method, and an information data processing terminal. The systemcomprises a preprocessing module, a word segmentation module, an extraction module and a conversion module. The method comprises the steps of: cleaning original data through reduction, conversion andelimination of same-effect synonyms, and giving a clinical behavior relationship to preprocess related information of an original medical record; performing word segmentation processing on the related information of the original medical record; adopting a limited implicit theme description document, wherein each theme is composed of a plurality of vocabularies, and related information of an original medical record is extracted and processed; and converting the text medical record recorded in the natural language into data easy to apply a mining algorithm. According to the invention, the medical diagnosis rules with the very high medical value can be mined; the diagnosis level of doctors is improved, and beneficial exploration is made for traditional Chinese medicine informatization construction; and moreover, a data set faced by a data mining algorithm is dynamically increased, and the data volume is developed towards mass data along with the increase of users.

The invention discloses a knowledge social recommendation method, system and device based on a graph neural network, and the method comprises the steps: respectively mining influence factors of a user and a project from a social network and a knowledge graph, calculating embedded vectors of the user and the project, carrying out the modeling and decoding again, fusing the knowledge graph and the social network, and constructing a recommendation model. According to the method, system and equipment, data sparsity and cold start encountered during single information recommendation are avoided, the problems of score data sparsity for users and projects and cold start for recommending new user projects in an existing recommendation system are solved, meanwhile, static performance and independence for processing the users and the projects are avoided, and the recommendation performance is improved.

The invention provides a method, a device, and an application for sequentially creating a multi-dimensional feature map of a personal scene. The method comprises the following steps: presetting a multi-level scene tag library, and presetting a corresponding scene logic relationship, an association probability calculation rule between scenes, and a sequential relationship and chain serial expression relationship between the scenes; automatically acquiring and / or calculating a multi-level scene label value of the user by using a base station or / and a satellite positioning system used by a mobilephone of the user, correspondingly storing the value in the preset multi-level scene label library, and generating a user personal scene according to the preset scene logic relationship; andThe invention provides a method, a device, and an application for sequentially creating a multi-dimensional feature map of a personal scene. The method comprises the following steps: presetting a multi-level scene tag library, and presetting a corresponding scene logic relationship, an association probability calculation rule between scenes, and a sequential relationship and chain serial expression relationship between the scenes; automatically acquiring and / or calculating a multi-level scene label value of the user by using a base station or / and a satellite positioning system used by a mobile phone ofthe user, correspondingly storing the value in the preset multi-level scene label library, and generating a user personal scene according to the preset scene logic relationship; based on multiple independent user personal scenes, calculating a correlation probability among multiple independent user personal scenes in a preset time period according to the association probability calculation rule between scenes, and generating the user personal scene map according to the sequential relationship and chain serial expression relationship between the preset scenes. According to the method, the device, and the application for sequentially creating the multi-dimensional feature map of the personal scene, the automatic creation of a personal scene map is realized, the personal scene map is appliedto realize user behavior prediction, and personalized recommendation is realized according to a prediction result.based on multiple independent user personal scenes, calculating a correlation probability among multiple independent user personal scenes in a preset time period according to the association probability calculation rule between scenes, and generating the user personal scene map according to the sequential relationship and chain serial expression relationship between the preset scenes. According to the method, the device, and the application for sequentially creating the multi-dimensional feature map of the personal scene, the automatic creation of a personal scene map is realized, the personal scene map is applied to realize user behavior prediction, and personalized recommendation is realized according to a prediction result.