208results about How to "Reduce sparsity" patented technology

The invention discloses a short text classification method based on a convolution neutral network. The convolution neutral network comprises a first layer, a second layer, a third layer, a fourth layer and a fifth layer. On the first layer, multi-scale candidate semantic units in a short text are obtained; on the second layer, Euclidean distances between each candidate semantic unit and all word representation vectors in a vector space are calculated, nearest-neighbor word representations are found, and all the nearest-neighbor word representations meeting a preset Euclidean distance threshold value are selected to construct a semantic expanding matrix; on the third layer, multiple kernel matrixes of different widths and different weight values are used for performing two-dimensional convolution calculation on a mapping matrix and the semantic expanding matrix of the short text, extracting local convolution features and generating a multi-layer local convolution feature matrix; on the fourth layer, down-sampling is performed on the multi-layer local convolution feature matrix to obtain a multi-layer global feature matrix, nonlinear tangent conversion is performed on the global feature matrix, and then the converted global feature matrix is converted into a fixed-length semantic feature vector; on the fifth layer, a classifier is endowed with the semantic feature vector to predict the category of the short text.

The invention provides a Mongolian-Chinese machine translation method for enhancing semantic feature information based on a Transformer model. The method comprises the following steps: firstly, starting from the language characteristics of Mongolian, finding out the characteristics of the additional components of the Mongolian in terms of stem, affixes and lattices, and merging the language characteristics into the training of a model; secondly, distributed representation for measuring the similarity between the two words is taken as a research background, and the influence of depth, density and semantic coincidence degree on the concept semantic similarity is comprehensively analyzed; in the translation process, a Transformer model is adopted, and the Transformer model is a multi-layer encoder which performs position encoding by using a trigonometric function and is constructed on the basis of an enhanced multi-head attention mechanism. A decoder architecture, which completely dependson the mechanism of attention to draw the global dependency between the input and the output, eliminates recursion and convolution.

The invention discloses a multiattribute collaborative filtering recommendation method oriented to a social network. The multiattribute collaborative filtering recommendation method includes utilizing mass data information of the social network to collect user, friend and item list information, and establishing an original user-item scoring matrix; utilizing a thought of acquiring a middle average value from nine numbers, and performing prediction filling on a sparse matrix; calculating inter-user attracting similarity through a user-item bipartite graph; calculating interaction similarity, linearly combining the attracting similarity with the interaction similarity to acquire comprehensive similarity among users, and searching to acquire a nearest neighbor set of a target user; performing prediction scoring on items to be recommended by the target user according to the nearest neighbor set of the target user, and generating a Top-N recommendation set. By the method, calculating rules of inter-user similarity in a conventional collaborative filtering method are improved, huge impedance brought to the filtering recommendation method and a recommendation system by sparseness of a scoring matrix is reduced, and accuracy of the recommendation system is improved.

The invention discloses an association-rule recommending method based on self-adaptive multiple minimum supports. The method comprises the following steps of firstly, establishing a commodity-classifying hierarchical tree according to commodity classification, and classifying concrete commodities according to the classifying hierarchical tree; next respectively carrying out minimum-support threshold-value setting on each concrete commodity and the upper-layer class of a concrete-commodity layer, and then mining frequent item sets and generating rules by utilizing a multiple-minimum-support association-rule expanding algorithm on the basis of the support threshold-value setting, wherein the threshold-value setting relates to the influences of time factors, concrete-commodity price factors and concrete-commodity brand factors; finally generating recommendation for each user by adopting a TOP-N recommending method. When personalized recommendation is made for the user by the association-rule recommending method, the characteristics of different objects can be better embodied by considering the influences of many factors on the multiple-minimum-support threshold-value setting for the concrete commodities and the classes; meanwhile, a data-sparsity problem and a cold-starting problem in a recommending system are relieved, so that the personalized recommendation can be more accurately made for the user.

The invention discloses a recommended system and a method with facing social network for context awareness based on the tensor decomposition, and relates to the field of the data mining and the information retrieval. Firstly, the method makes use of a social network massive data set to collect users and projects and contexts, to pay attention to the list information, to establish an original the user-the project-the context mark matrix, to calculate the users similarity, and to establish a user-user similarity matrix; Secondly, aim at the extreme sparsity of the original mark matrix, a sparse mark matrix is predicated and filled by using the tensor decomposition; Thirdly, aim at a problem that the user similarity matrix is sparse, a sparse user similarity matrix is predicated and filled by using the matrix decomposition; Finally, according to some similar interest tendencies of some similar users in the social network, a social normalization item is taken to optimizing the mark matrix. The method deals with the problem that a traditional predicated mark matrix does not consider that the context information and the relationship between users have an effect on marking. Also, the method deals with an obstruction which is caused by the sparsity of the mark matrix brings to the recommended system, thus the accuracy of the recommended system is improved. The method can be widely applied to the fields of the social network, the electronic commerce and the like.

The invention relates to a dynamic recommendation method capable of adapting to user interest changes based on time information. The method comprises the steps that a user-object explicit rating matrix is built; a user-object implicit rating matrix is built; a user-object comprehensive rating matrix is built; similarity between two users is calculated; K users close to the target user similarity are obtained as the neighbour set of target users; a monotone decreasing exponential time function is selected as a rating weighting function, and the weighting factor of each user in the rating weighting function is calculated according to different interest change trends embodied by the scores of the users; a TOP-N recommendation method is adopted, and N objects with very top predication scores are recommended to the users. According to the method, the changes of the user interest along with time are considered, and the more precise personalized object recommendation service is provided for the users.

A book recommending method based on the Markov chain comprises the steps: (1) inquiring reading history data of a user, and calculating an initial state probability vector of the user based on the simple Bias algorithm, (2) inquiring a current read book and all unread books of the user, and respectively calculating a transition probability matrix from the current read book to each unread book according to like and dislike of the user on the current read book and the unread books of the user, (3) respectively calculating a state probability vector from the current read book to each unread book according to the initial state probability vector of the user and the transition probability matrix from the current reading book to each unread book, and recommending unread books to the user based on the state probability vector which includes the state probability of like of the user on the unread books. The book recommending method belongs to the field of network application technology, and is capable of recommending books in a personalized mode according to dynamic behaviors of the user.

The invention discloses a meta-path-based link prediction method for aligned heterogeneous social networks, mainly solves the problem of link prediction (mainly including friend relationship prediction and position prediction) in the aligned heterogeneous social networks, and relates to related technologies of the aligned heterogeneous social networks, feature selection, link prediction and the like. The method mainly comprises six steps of (a) modeling the aligned heterogeneous social networks; (b) automatically extracting a meta-path; (c) defining a meta-path-based eigenvalue calculation method; (d) dividing data sets according to timestamps; (e) proposing a two-stage gradual forward greedy feature selection algorithm for performing feature selection; and (f) training a decision tree classifier based on a feature selection result to perform the link prediction. Based on the method, the aligned heterogeneous social networks consisting of Foursquare and Twitter are subjected to the link prediction; and the method has practical application values for entity recommendation, accurate marketing, criminal gang discovery and the like in the social networks.

The invention provides a Bayesian model-based commodity code classification method and system. Aiming at the problems of sparsity and context loss caused by short text features of commodity name information, the method and system adopt a synonym extension technology for word-segmented sparse words, and increase the word synonym degrees through an external search engine technology; and then, a Bayesian model is adopted to calculating condition probabilities between word sequences and code categories of commodity names so as to obtain a Bayesian code classification mode, and a trained model is used for predicting commodity code categories of input commodity names.

The invention provides an n-gram grammar model constructing method for voice identification and a voice identification system. The method comprises: step (101), training is carried out by using a neural network language model to obtain word vectors, and classification and multi-layer screening is carried out on word vectors to obtain parts of speech; step (102), manual marking is expanded by using a direct word frequency statistic method; and when same-kind-word substitution is carried out, direct statistics of 1-to-n-gram grammar combination units changing relative to an original sentence is carried out, thereby obtaining an n-gram grammar model of the expanding part; step (103), manual marking is carried out to generate a preliminary n-gram grammar model, model interpolation is carried out on the preliminary n-gram grammar model and the n-gram grammar model of the expanding part, thereby obtaining a final n-gram grammar model. In addition, the step (101) includes: step (101-1), inputting a mark and a training text; step (101-2), carrying out training by using a neural network language model to obtain corresponding work vectors of words in a dictionary; step (101-3), carrying out word vector classification by using a k mean value method; and step (101-4), carrying out multi-layer screening on the classification result to obtain parts of speech finally.

The invention discloses a channel estimation method for the sparse characteristic of underwater acoustic block structure, which includes three modules: an OFDM system overall working module; a block orthogonal matching tracking algorithm flow module; and a block sparse compressed sensing channel estimation module. This channel estimation method can effectively reduce signal sparsity, thereby reducing the number of pilots required for estimation, greatly reducing the time required for algorithm estimation, and improving the accuracy of reconstruction.

The invention provides a user track position prediction method based on space-time embedding Self-Attention, and relates to the technical field of user position prediction and space-time data mining.The method comprises the following steps: firstly, reading all user identifiers, POI identifiers, access time and geographical location information of POIs from a user historical track of an originalsign-in record; calculating the space distance cost among all POIs and the time interval and timestamp information among all track points of each user; then establishing a space-time embedded Self-Attention model based on the historical track of the user to obtain probability distribution of the next access point of the user, wherein the established space-time embedded Self-Attention model considers the influence of the geographic position information of the POI and the historical POI access time information of the user on the next access behavior of the user at the same time. According to themethod, the training speed of the track prediction model is remarkably increased, the training time of the track model is greatly shortened, and meanwhile, the prediction precision of the model is also improved.

The invention discloses a recommendation method capable of aiming at classification information with items. In various network applications, users always need to be recommended, context information needs to be used for improving recommendation accuracy and enhancing user experience, but traditional context perception recommendation methods still face the challenge of a data sparsity problem. In order to further alleviate the data sparsity problem, the invention puts forward a novel recommendation method, which combines user rating data with user category preference to carry out article recommendation so as to solve the problem of low rating prediction accuracy when the user rating data is sparse. The method is suitable for large-scale data. An experiment result indicates that the method has a good recommendation effect when the method is compared with a traditional mainstream method.

The invention relates to an L1 / 2-norm-based sparse linear array optimization method. The method is characterized by comprising the basic steps of determining an initial array and a weighting matrix, determining array weight vectors, judging whether the stimulus of head and tail array elements in the array weight vector is greater than a set stimulus minimum delta or not, judging whether an L1 norm for optimizing a difference between the previous and latter array weight vectors is smaller than a set error minimum xi or not, and determining the array element positions and stimulus of the sparse linear array. According to the method, the non-convex optimization problem of solution to an L1 / 2 norm is converted into a series of convex optimization problems of the L1 norm, so that a sparse array with lower sparsity can be obtained to reduce the number of actually required array elements on the premise of substantially keeping calculation unchanged; meanwhile, the head and tail array elements of the array are constrained and adaptively regulated under the condition of given array aperture, so that the problem of deficiency of the head and tail array elements of the sparse array in an iterative convex optimization process is well solved, and the method is particularly applied to the place of optimization of a large-sized antenna array.

The invention discloses a socialized convolution matrix decomposition-based document context sensing recommendation method. The method comprises the steps of firstly capturing context information of an article description document by utilizing a convolutional neural network (CNN), and taking obtained context eigenvector and Gaussian noise as potential vectors of a project; secondly, by utilizing acharacteristic that interests and hobbies of users are more easily influenced by trusted friends (having direct link relationships), determining a potential eigenvector of a target user by calculating an average value of potential eigenvectors of the friends; and finally predicting score information of the user for the project according to a joint probability distribution function of the user andthe project. According to the method, the CNN is seamlessly integrated in matrix decomposition technology-based socialized recommendation (SocialMF) from the perspective of a probability, so that thefriends having a trust relationship with the target user and interests relatively close to those of the target user can be further identified in a learning process, and the purpose of optimizing a recommendation result is achieved.

The present invention discloses a Github open source project recommendation method based on a sparse autoencoder. The method comprises: preprocessing data from three dimensions of a project, a user, and a project-user, and obtaining a user association degree matrix, a project association degree matrix, and a user-project association degree matrix; extracting textual information and clustering projects by analyzing text similarity; combining a collaborative filtering model and a sparse autoencoder to help developers find suitable open source projects; and taking the three matrices obtained by data preprocessing as input, obtaining two potential factor vectors through iterative neural network learning, predicting missing items in the user-project association degree matrix through the inner product of the potential factor vectors, and recommending the top N items with higher scores in the same category according to the clustering information of the open source project. According to the method disclosed by the present invention, suitable projects are recommended to the developers, the time taken by the developers to find interestied projects in massive open source projects is saved, and the performance of the developers participating in the open source projects is improved.

The invention discloses a personalized scenic spot recommendation method based on tourist preference modeling, and the method comprises the steps: collecting data, carrying out the preprocessing, and carrying out the numbering of tourists, scenic spots and other objects; converting the display score into an implicit score, and dividing a positive case scenic spot and a negative case scenic spot; constructing a triple and scenic spot knowledge map, and generating a feature vector and a context feature vector of each scenic spot; generating vector representations of historical tourist tour scenic spots and candidate scenic spots through the KCNN; calculating an influence weight of each historical touring scenic spot of the tourist through the attention network to obtain a preference vector of the tourist to the scenic spot; calculating the scenic spot touring probability of the tourists by using the DNN, and generating scenic spot recommendation lists of the tourists according to the probability from small to large. According to the method, when different influences of historical visiting scenic spots of tourists on the candidate scenic spots are depicted and diversification preferences of the tourists are represented, the attention network is used for calculating the influence weights of the historical visiting scenic spots of the tourists on the candidate scenic spots, so that the recommendation result better conforms to the preferences of the tourists.

The invention relates to a service recommendation method based on trust extension and listwise rank learning, which comprises the steps of firstly representing each user to be the probability distribution of invoked service set arrangement by utilizing service ranking position information and referring to a Plackett-Luce model, and calculating the probabilistic user similarity based on the KL distance; taking a direct trust relationship and an indirect trust relationship between the users into consideration at the same time, calculating the direct trust degree by using a Beta trust model, calculating the indirect trust degree by using transfer characteristics of the trust relationships, obtaining the comprehensive trust degree, and constructing a trusted neighbor set of target users; integrating the trusted neighbor set into a matrix decomposition model, taking a cross entropy between a predicted ranking list and a correct ranking list as a loss function, designing a listwise rank learning algorithm to obtain an optimal ranking model, and outputting a recommendation list which best conform to user interests. The service recommendation method has high recommendation accuracy, and can satisfy potential functional requirements of the users to be greatest extent while ensuring the service recommendation quality.

The invention provides a point of interest recommendation method for fusing social network and image conten.Based on the existing social recommendation model, a new matrix decomposition model which integrates social network and image content is proposed, which can make full use of the picture information in user evaluation information, alleviate the sparsity of data, and improve the accuracy of interest point recommendation. In addition, the invention utilizes the text information, the geographic information and the tag classification information of the interest point evaluation to better solve the cold start and accuracy problems of the interest point recommendation, recommends the most desired interest point to the user, and improves the user's experience degree.

The invention relates to a similarity propagation and popularity dimensionality reduction based mixed recommendation method. According to the similarity propagation and popularity dimensionality reduction based mixed recommendation method, sparse data are processed in two phases; firstly, neighbors of the sparse data are expanded due to constant iteration of similar matrixes of users, resources and Tags through a similarity propagation method and accordingly elements for zero are filled; then a score algorithm in a search engine is introduced to calculate the Tag popularity in consideration of the problem that original data is provided with meaningless rubbish Tags, the tags with the popularity smaller than a certain threshold value are deleted to simplify data to perform dimensionality reduction on the matrix; recommendation results are diversified and the sparsity and cold starting problem can be relieved to some extent due to the fact that the recommendation based on contents and the collaborative filtering recommendation are combined. The similarity propagation and popularity dimensionality reduction based mixed recommendation method has the advantages of solving the problem of data sparsity in the individual recommendation process and being high in recommendation result accuracy, high in accuracy and high in reliability.

The invention relates to a hybrid collaborative filtering movie recommendation model based on a bilateral network structure, and belongs to the field of recommendation systems. The model introduces anidea of deep learning and uses posters and description texts of movies, effective feature expression of users and articles is generated through CNN, a probability matrix decomposition technology is used to introduce potential users and movie models through Gaussian noise parameters, and a deep neural recommendation model with a bilateral network structure is integrated and built. The model is applied to open source dataset of ML_1M expressed by MovieLens_1M from a MovieLens website to perform experiments, so that it can be verified that when the deep learning is used to introduce auxiliary information of movie posters and description texts for building the hybrid collaborative filtering recommendation model, under the condition that the sparse interactive scoring matrix of the users and articles is improved, whether or not the influence of the data sparsity on recommendation accuracy is reduced. In the case of sparse data, the hybrid collaborative filtering movie recommendation modelbased on the bilateral network structure has the advantages that the ability to generate effective feature expression is more obvious, the influence of data sparsity is reduced, and the problem of inaccurate recommendation is effectively improved.

The invention provides a massive data processing method and a massive data processing device, and aims to solve the problem of unremarkable data discrimination effects caused by higher sparsity of original data. The method comprises the following steps of storing the massive original data into an (m*n)th-order original matrix A, wherein m and n are positive integers, and the original data is operating data for each user; when the original matrix A is subjected to singular value decomposition, distributing the original data in the original matrix A to a plurality of processing nodes for processing; reconstructing a first unitary matrix U, a first diagonal matrix S and a second unitary matrix V, which are obtained by the singular value decomposition, to obtain a corresponding reconstructed matrix B; and clustering data in the reconstructed matrix B to discriminate the data of different types. According to the method and the device, the problem of higher sparsity of the original data is radically solved, so that higher data discrimination performance during subsequent massive data processing is ensured.

The invention discloses a support vector regression recommendation method based on context sensing. The method comprises the following steps: constructing a user characteristic attribute matrix; obtaining an item characteristic attribute matrix and an item characteristic attribute preference matrix, and constructing a user preference matrix; constructing a context situation matrix and a grading matrix, and constructing a context-based user preference model according to the context situation matrix, the user characteristic attribute matrix, the user preference matrix and the grading matrix; optimizing the context-based user preference model by virtue of a support vector regression SVR algorithm, so as to obtain an effective grading prediction model; and calculating the grades of items which are not bought by a target user based on the grading prediction model, and recommending front L items with the highest grades to the target user. The invention further discloses a corresponding system. The method and the system can be applied to the recommendation of supplementary services of civil aviation passengers, and services suitable for the passengers can be rapidly and accurately found from numerous services.

The invention discloses an implicit feedback recommendation method based on node2vec and a depth neural network, belonging to the technical field of data processing. The invention comprises the stepsof obtaining a user potential vector and an item potential vector; obtaining a user context prediction and a project context prediction; generating a user preference prediction for the project and training the user preference prediction; performing joint training. The method can well take into account the rich metadata information of users and items, and has high recommendation accuracy and modeltraining efficiency.

The invention discloses a text information-based deep reinforcement learning interactive recommendation method and system, and belongs to the field of interactive personalized recommendation, and themethod comprises the steps: respectively converting commodities and users into commodity vectors and user vectors based on text information, and carrying out the clustering of the users; establishinga recommendation model for each user category based on DDPG, and establishing a global environment simulator; for any recommendation model, constructing an action candidate set Can (ui, t) in the t-thround of interaction; enabling the strategy network to take the state st of the current user as an input, obtaining a strategy vector pt, and selecting an action vector at from the Can (ui, t) according to the pt; enabling the valuation network to take the pt and st as input, and calculating a Q value used for evaluating the advantages and disadvantages of pt; in each round of interaction, enabling the environment simulator to calculate a feedback reward value and update the state of the current user; and outputting a feedback reward value to the value estimation network, correcting the valueestimation network, reversely conducting the Q value to the strategy network, and adjusting the strategy network to obtain a better strategy vector. The recommendation efficiency and the recommendation accuracy can be improved.

The invention discloses an improved one-class collaborative filtering method based on socialized information fusion. The improved one-class collaborative filtering method includes the steps of 1, constructing triples to respectively express user project information, project tag information, user tag information and group friend information; 2, computing preference similarity of friends in the group of the user; 3, computing feature similarity of the user's preference with his unselected project; 4, extracting and adding negative examples of the user; 5, subjecting user project matrix to joint probability matrix decomposition to obtain user feature matrix and project feature matrix; 6, acquiring the front N projects with highest prediction scores of each user to form a recommendation list of the user. Negative example extraction is carried out on the basis of project tag information and quality of positive examples selected by the user, socialized information of user groups and project tags are fused to the probability matrix decomposition to realize joint probability matrix decomposition, and accordingly, recommendation results to one-class data are acquired, and problems of height unbalancedness and sparsity of data caused in implementing of the conventional one-class collaborative filtering methods are solved effectively.

The invention discloses a short-text data stream classification method based on topic models and concept drift detection. The method includes: 1, acquiring an external corpus from a knowledge libraryto construct the LDA topic model; 2, dividing a short-text data stream into data blocks according to a sliding window mechanism, and using the LDA topic model to expand short text in the data blocks to obtain an expanded data stream; 3, constructing the online BTM topic model for each data block in the expanded short-text data stream, and obtaining a topic representation of each piece of short text; 4, selecting data blocks of Q topic representations to construct a classifier to use the same to predict a class label of a newly arrived data block; 5, dividing the data blocks of the Q topic representations into category clusters according to class label distribution, and calculating semantic distances between the category clusters and the newly arrived data block to judge whether concept drift occurs; and 6, updating the classifier according to a concept drift situation. The method can be used for the short-text data stream classification problem of unceasingly changed class label distribution.

The invention discloses the sequence abnormity detection method of a telecommunication phone fraud based on sliding time window aggregation and belongs to the data mining, machine learning and business intelligence field. The method comprises the following steps of firstly, constructing a training user data set, and recalling all the call records of a called user to form a call sequence of each called user; using a cos similarity function, calculating a sequence structure similarity, counting a characteristic similarity and carrying out linearly combination to obtain an additivity similarity;then, through a K-Means clustering model, acquiring a K-class user so as to form an independent sequence training data set, and forming K training sets through a sliding time window; and finally, training an iForest model on each training set to obtain K abnormal detection models. Each called user identifies an abnormity through a corresponding abnormity detection model. When a maximum value is higher than a threshold h, the called user is a high-risk called user. The K-Means model and the abnormity detection model are updated every fixed period of time. In the invention, a data sparsity problem is alleviated and an abnormal characteristic based on a group is discovered.

The invention discloses a friend recommending method in a position social network based on positions and time. The method mainly is used for solving the problem of low recommending precision resulting from the face that the existing friend recommending scheme ignores time information and sign-in position semantic information in sign-in information. The method comprises following steps of 1, building a communication system framework composed of users, positioning facilities and a position server; 2, sending the longitudes and latitudes of the positions and current time information to a social network server by the users, thus generating the sign-in information; 3, preprocessing massive stored sign-in information by the social network server; 4, calculating the word frequency-inverse document frequency value of each user to each place class by the social network server; 5, calculating similarities among the users by the social network server; and recommending friends to the users with relatively high similarities. According to the method, the available data sparsity of a recommending system is mitigated; the recommending precision is improved; and the method can be applied in the wireless social network service based on the positions.