260 results about "Machine reading" patented technology

The invention discloses a deep neural network and reinforcement learning-based generative machine reading comprehension method. According to the method, texts and questions are encoded through an attention mechanism-combined deep neural network so as to form question information-fused text vector expressions, and decoding is carried out through a unidirectional LSTM decoder so as to gradually generate corresponding answer texts. According to the reading comprehension method, the advantages of extractive models and generative models are fused, training is carried out by adoption of a multi-taskcombined optimization manner, and a reinforcement learning method is used in the training process, so that benefit is brought to generate more correct and fluent answer texts.

The invention provides a machine reading understanding method based on a multi-head attention mechanism and dynamic iteration, and belongs to the field of natural language processing. The constructionmethod of the machine reading understanding model comprises the following steps: constructing an article and problem coding layer; constructing a recurrent neural network based on bidirectional attention flow; a self-attention layer is constructed and answer output is predicted based on a dynamic iterative decoder. According to the method, answer prediction can be carried out on questions in a machine reading understanding task text; according to the invention, a new end-to-end neural network model is established, and a new idea is provided for exploring a machine reading understanding task.

The present application relates to an answer selection method and apparatus based on an improved attention mechanism and electronic device. The method comprises the following steps: performing word vector conversion on the obtained text data and the problem data respectively to obtain word vector representations of each word of the problem data and the text data respectively; processing the problem data and the text data through a loop neural network model to obtain a problem semantic vector representation and a text semantic vector representation, respectively; processing the problem semanticvector representation and the text semantic vector by stacking the attention mechanism layer to obtain a contextual representation of the fusion problem information; obtaining an answer correspondingto the question data from the text data based on the context representation of the fusion question information and the question semantic vector representation. In this way, the system for machine reading comprehension is optimized through a specific model architecture incorporating an improved attention mechanism to improve the effect of short text answer extraction.

The invention discloses a machine reading model training method and device, a question and answer method and a device, and belongs to the field of natural language processing. The machine reading model training method comprises the following steps: acquiring a training sample, wherein the training sample comprises a sample question and a corresponding sample article, a real initial position and areal final position of the corresponding answer in the sample article; extracting a question feature vector of the sample question and an article feature vector of the sample article, and using a neural network structure to fuse and process the question feature vector and the article feature vector to form a fusion result; inputting the fusion result into a classifier to perform prediction of theinitial position and the final position of the answer; performing error calculation on the predicted initial position and final position, and the real initial position and the real final position of the answer; and optimizing the neural network structure according to the error calculation result. According to the machine reading model training method and device, the question and answer method anddevice in the embodiment of the invention, the corresponding answer can be directly extracted from the entire associated article through end-to-end deep learning.

The invention provides a sheet-like anti-counterfeiting material, and the sheet-like anti-counterfeiting material adopts a special-shaped structure except for positive quadrilateral; the sheet-like anti-counterfeiting material comprises the following layered structures: a paper-based anti-counterfeiting carrier layer, wherein the paper-based anti-counterfeiting carrier is paper prepared by remixing at least two fibers; and a anti-counterfeiting layer. The invention also provides a preparation method of the sheet-like anti-counterfeiting material, and anti-counterfeiting paper containing the sheet-like anti-counterfeiting material. Based on the sheet-like anti-counterfeiting material of the invention, anti-counterfeiting paper can be obtained with multi-level anti-counterfeiting functions of visual identification, rapid machine-reading, and expert analysis.

A machine reading system is described herein that includes a framework in which grammar rules can be developed using a concise language that combines syntax and semantics. The resulting technology thus reduces the development time for new grammars in a new domain. An enormous amount of information appears in the form of natural language across millions of academic papers and other literature sources. For example, in the biological domain, there is a tremendous ongoing effort to extract individual chemical interactions from these texts, but these interactions are only isolated fragments of larger causal mechanisms such as protein signaling pathways. The proposed rule-based event extraction framework can model underlying syntactic representations of events in order to extract signaling pathway fragments. Though application to the biomedical domain is herein described, the framework is domain-independent and is expressive enough to capture most complex events annotated by domain experts.

The invention discloses a multi-granularity answer sorting multi-document machine reading understanding method, and belongs to the technical field of machine reading understanding application. The method is based on a pre-trained deep learning model. Splitting the document into text fragments through a sliding window, and splicing the text fragments with questions; a plurality of candidate answersgenerated by a plurality of documents are sorted by fusing multi-granularity answer sorting of statistical information, shallow semantic information, deep semantic information and answer ending wordinformation, and the semantic information of different granularities is fully utilized to capture the correlation between a question and the plurality of candidate answers. According to the method, the text representation capability and the generalization capability of a traditional machine reading understanding model are improved by utilizing a pre-trained deep learning model; Meanwhile, the defect that the input length of an existing model for a multi-document scene is limited is overcome, Meanwhile, the answer quality of multi-document machine reading understanding is improved by modeling the correlation between questions and answers from different granularities.

The invention discloses a charging processing method and system for internet sales. The method comprises the following steps: establishing a network commodity promotion service website and recording a corresponding relation between a commodity sales webpage and promoter information, seller information, commodity information, and profit allocation mode information; distributing a global unique identity ID for the corresponding relation and generating a two-dimensional code that can be read and identified by a machine or other commodity promotion carrier; receiving a purchase request that is sent out by a purchaser from a webpage of the promotion service website or receiving a purchase request that is sent by a purchaser after machine reading and identifying of the global unique identity ID of the corresponding relation, and generating an order; obtaining the promoter information, the seller information, the commodity information, and the profit allocation mode information from the promotion service website; and according to the obtained promoter information, seller information, commodity information, and order information, realizing expense collection and automatically realizing profit allocation charging between the promoter and the seller based on the profit allocation mode information.

The invention discloses a machine reading comprehension answer obtaining method based on a multi-round attention mechanism. The method comprises steps of performing word segmentation processing and vectorization processing on the questions and the texts corresponding to the questions respectively to obtain feature vectors, selecting a bidirectional long-short time memory network to encode contextsemantic information of the feature vectors, and performing modeling between the questions and the texts by using an attention mechanism to effectively capture information interaction between the questions and the texts. Attention of an article about a question is calculated through multiple rounds, context semantic information is fused, then BLSTM is used for coding the context semantic information, the above processes are repeated for multiple times, so that an nth text semantic vector is obtained, and a Self-Attention mechanism is used for obtaining a vector representation of the question;by calculating the similarity between the semantic vectors of the questions and the similarity of the semantic vectors, namely one representation of each word in the article in the question space, theaccuracy of predicting answers can be effectively improved, BLSTM and Attention are effectively combined, and the matching accuracy of the questions and the answers returned by text extraction can beimproved.

The invention provides a deep learning machine reading understanding training method based on curriculum learning, and the method employs a BERT pre-training language model to build (articles, questions and options) triples into a sequence, and does not need to carry out the independent operation of each tuple. Four option sequences forming a question are input into a network, a fine adjustment process the same as BERT is carried out, a maximum probability option is selected as a prediction answer through a full connection layer and a softmax classification layer, and parameters of a model arereversely updated by maximizing the logarithm probability of a correct answer, so that the model learns text information. A three-stage training framework is firstly finely adjusted in a simple dataset; text knowledge can be learned in sequence from shallow to deep by conducting fine adjustment on a common data set, and finally the test effect after training on a difficult data set is 2.5% higher than the accuracy of fusion learning (fine adjustment is conducted on a set formed by mixing a simple data set and the common data set, and then training is conducted on the difficult data set).

The invention discloses a machine reading understanding method, equipment, a storage medium and a device. The method comprises the steps of obtaining a to-be-understood paragraph and a plurality of corresponding target problems; and carrying out multi-thread processing on the to-be-understood paragraph and the plurality of corresponding target questions, obtaining interaction information semanticsbetween the paragraph to be understood and each target problem through an embedded layer, a coding layer and an interaction layer of the understanding model of a preset machine in sequence; based onartificial intelligence, enabling the interaction information semantics to pass through a screening layer of the preset machine reading understanding model; obtaining a valuable sentence vector whichis highly associated with each target question; and enbabling the valuable sentence vector to passes through an answer layer of the preset machine reading understanding model to obtain a predicted answer range of each target question, and performing answer prediction through the preset machine reading understanding model. The accuracy and efficiency of answer prediction are improved, the predictedanswer range is sent to a target terminal, and the user experience is improved.

The invention provides an intelligent dialogue method, a device and a terminal based on machine reading comprehension. The method comprises the following steps: obtaining a text corresponding to a question put forward by a user; word segmentation and vectorization being performed on the problem and the text to obtain the problem vector and the text vector; inputting a problem vector and a text vector into the attention model to obtain a first vector and a second vector, wherein the first vector is used for indicating the degree of influence of the problem on noticing any word in the text, andthe second vector is used for indicating the degree of influence of the text on generating the problem; an answer start point and an answer end point being determined in the text according to the first vector and the second vector, and a paragraph between the answer start point and the answer end point being determined as the answer of the question. The method of the invention does not need to preset a 'question-answer' pair in advance, can flexibly answer various questions of a user, overcomes the defect that the prior art needs to continuously maintain the question database, and reduces thedata update cost.

The embodiment of the invention provides a question answer acquisition method and system based on machine reading understanding. The method comprises the following steps: inputting a question and a corresponding document set into a trained neural network model, and obtaining an output result of the trained neural network model; determining an answer corresponding to the question from the documentset according to the output result; wherein the trained neural network model is obtained through training according to a training set, and the training set comprises a plurality of sample problems anda sample document set and a sample label set corresponding to each sample problem. According to the method and the system provided by the embodiment of the invention, the question and the corresponding document set are input into the trained neural network model, and the answer corresponding to the question is determined from the document set according to the output result of the trained neural network model. The method has the advantages that the shortage of a machine reading understanding model of the description type problem is filled, the characteristics of multiple documents are effectively utilized, more document information is reserved, and the answer of the description type problem can be extracted more accurately.

The invention discloses a topic-considered machine reading understanding model generation method and system. According to the present invention, the potential topic information in the training sampledata is extracted, and the topic information is utilized to supervise the training of a reading understanding model, so that the effect of the reading understanding model is improved. According to themodel disclosed by the invention, a plurality of topics corresponding to the training samples are extracted before model training, and the topic information of the samples is utilized to improve theeffect of the machine reading understanding work. The basic process of the method comprises the following steps of processing each training sample, and finding out a vector representation capable of representing the sample; clustering the samples, and solving a mean value of the similar sample vectors as the vector representation of the topic; during matching and outputting, using an attention mechanism for representing the higher weight of the words with higher similarity with the topic vector of the sample for the vector. In addition, the training data can obtain a better effect after beingsubjected to better data cleaning, and better topic vector representation can be obtained after noise is reduced.

The invention discloses a machine reading comprehension method based on multi-task joint training, and a computer storage medium. The machine reading comprehension method comprises a model construction and training stage and a test stage, wherein the model construction and training stage comprises the following steps: 1, constructing a training set; 2, establishing a machine reading comprehensionmodel which comprises a coding layer, an attention matching layer and a decoding layer; 3, dividing the training set into W sub-sample sets according to the number of samples, and training the machinereading comprehension model by using the W sub-sample sets to obtain W machine reading comprehension models; and the testing stage comprises the steps: cutting off a to-be-tested article, inputting the processed article and questions into the trained W machine reading comprehension models, obtaining W predicted answer starting positions, W predicted answer ending positions and W corresponding distribution probabilities, and calculating the average value to serve as the starting position, the ending position and the corresponding distribution probabilities of a final answer. The machine reading comprehension method can solve the problems that an existing machine reading comprehension method is low in Chinese text accuracy and the like.

The invention relates to the fields of deep learning, natural language processing and the like, and particularly relates to a syntactic relationship enhanced machine reading understanding multi-hop reasoning model and method. The model comprises a text coding module, an association element relationship graph construction module, a multi-hop reasoning module, an answer generation module and an answer prediction module. According to the invention, syntactic relationships are fused into the graph construction process, an associated element relationship graph is constructed, multi-hop reasoning iscarried out by utilizing a graph attention network based on the relationship graph, and answer support sentences are mined; meanwhile, a multi-head self-attention mechanism is introduced to further mine text clues of viewpoint type questions in the article, and an automatic solution method of the viewpoint type questions is improved; and finally, a plurality of tasks are subjected to joint optimization learning, so that when the model answers the questions, fact description for supporting the answers can be given, the interpretability of the model is improved, and meanwhile, the existing answering method for viewpoint type questions is improved.

A method of creating a visual access token is described. The method comprises: receiving an input referencing private information without revealing the private information; retrieving the referenced private information using the referencing input; encoding the retrieved private information as part of a visual image for use as a visual access token; and presenting the visual access token on a display of a portable communication device for machine reading.

The invention provides a candidate answer screening method for global machine reading comprehension modeling, which belongs to the technical field of computer information screening. All the paragraphscorresponding to a question are taken as a candidate answer fragment locating range in the method. Firstly, the F1 values between the text fragments of paragraphs are obtained, and the best candidateanswer fragment is selected by using F1; and on the other hand, after the features between the paragraphs and the question are extracted, correlation scoring is carried out by using a logistic regression model, a selected candidate answer paragraph set is obtained according to the scores, whether the paragraph where the best candidate answer fragment is located is in the candidate answer paragraph set is determined, and the paragraph where the best candidate answer fragment is located is forcibly put in the first place of the candidate answer paragraph set. Finally, the best candidate answerfragment and the candidate answer paragraph set are output. The method has the advantage that the efficiency of training and prediction is improved.

The invention provides a machine reading comprehension method based on a threshold convolution neural network. The method comprises the following steps: constructing a threshold convolution neural network model, comprising an input layer, a threshold convolution layer and an answer layer; The input layer is used for encoding the target article, and the encoded article vector sequence, the questionvector sequence and the answer vector sequence are transmitted to the threshold convolution layer. The threshold convolution layer generates text, question and answer expressions with high-level semantic information through interactive mode, and transmits these expressions to the answer layer. Finally, the answer layer makes reasoning decision and makes prediction. Determine the target article, import the threshold convolution neural network model for machine reading comprehension, and derive the prediction results. The machine reading comprehension method based on the threshold convolution neural network provided by the invention effectively simplifies the neural network model, greatly reduces the training and testing time, improves the processing efficiency, and improves the sense of user experience. Keep the long-term dependency of the text and accurately predict the answer information.

A verifiable display is provided that enables the visual content of the display to be detected and confirmed in a variety of ambient lighting conditions, enviroments, and operational states. In particular, the verifiable display has a display layer that is capable of visually setting an intended message for human or machine reading, with the intendended message being set using pixels. Depending on the operational condition of the display and the ambient light, for example, the message that is actually displayed and perceivable may vary from the intended message. To detect what message is actually displayed, a light detection layer in the verifiable display detects the illumination state of the pixels, and in that way is able to detect what message is actually being presented by the display layer.

The invention provides a multi-task chapter-level event extraction method based on a multi-headed self-attention mechanism. The method comprises the following steps: converting single sentence-level event extraction into chapter-level event extraction of a packaged sentence set; carrying out the word embedding representation by utilizing a pre-trained language model BERT; taking all word embedding and position embedding in a single sentence as input, employing a convolutional neural network model for coding, and capturing the most valuable features in the sentence in combination with a segmented maximum pool strategy; utilizing a multi-head self-attention model to obtain chapter representation and attention weight fused with full-text semantic information; utilizing a classifier to obtain a predicted event type; taking event types as prior information, linking the event types to an input sequence for event element extraction, and extracting all related elements in the sequence by using a pre-training model in combination with a machine reading understanding method. The method can be used for text-level event extraction tasks, and the breakthrough of converting a sequence labeling problem into a machine reading understanding problem is achieved.

The invention discloses an answer acquisition method based on machine reading understanding for open domain questions and answers. A BERT-based semantic coding module and an information interaction attention network are adopted to deeply capture potential semantic representations of problems and documents, effectively extract and fuse information between the problems and the documents, and captureglobal features of the problems and the documents; and an answer acquisition module based on Pointer Networks is adopted, and the attention weight is used as a pointer, so that the start and stop positions of the predicted answer can be positioned more accurately. The invention provides a reading comprehension-based answer acquisition method for open domain questions and answers. Empirical evaluation is performed on a CMRC 2018 data set. Experimental results show that the method can reach the standard level of open domain questioning and answering tasks, and excellent performance is obtained.

The invention provides a two-dimensional product which can realize the storage of information data for a hand-drawing or a reference pattern and a word structure and the two-dimensional product comprises a hand-drawing product, a code generator, a decoder and a system component. The hand-drawing code technology of the invention enables information to be precisely stored in a hand-drawing or specified pattern structure and a word profile without compromising the overall characteristics and the original expressions of the original pattern or word by keeping the characteristics of the original pattern in culture and art. Therefore, it is possible to add in-depth information storage and analysis technology on the basis of meaning expressions of the pattern and word so that a user could play an important role in the design of the code sample; the generation and production of the code becomes more people oriented; and the application scope of the pattern code product is enriched with personal signature, company trademark and other word pattern. The code of the invention can be directly accessed and analyzed both online and offline. All of the patterns and words that the pattern code technology assigns are provided with the dual information attribute featuring people reading and machine reading so that the pattern information that the people read is consistent with what the machine reads on the same carrier.

The invention belongs to the technical field of natural language processing, particularly relates to a machine reading understanding method, system and device based on external knowledge enhancement,and aims to solve the problem that the existing machine reading understanding method does not utilize graph structure information among triples, so that the answer prediction accuracy is relatively low. The method of the system comprises the following steps: generating context representations of entities in a question and an original text; based on an external knowledge base, obtaining a triple set of each entity in the question and original text and a triple set of adjacent entities of each entity in the original text; based on the triple set, obtaining knowledge sub-graphs of each entity through an external knowledge graph; updating the fused knowledge sub-graph through a graph attention network to obtain knowledge representation; and splicing the context representation and the knowledgerepresentation through a sentry mechanism, and obtaining answers to the to-be-answered questions through a multilayer perceptron and a softmax classifier. According to the method, the graph structureinformation among the triples is utilized, so that the answer prediction accuracy is improved.

The invention discloses a reading understanding method and system based on collaborative attention and self-adaptive adjustment, which belong to the technical field of machine reading understanding. The method is characterized by comprising the following steps: S1, inputting a document word vector and a problem word vector, and training the document word vector and the problem word vector, the word vector comprising two granularities of character-level embedding and word embedding; s2, calculating a similarity weight of the problem and the document by using cosine similarity, and performing adaptive adjustment on document word embedding according to the similarity weight; s3, encoding the document word vector and the problem word vector through a multi-layer bidirectional gating loop unit;s4, using a collaborative attention mechanism for the document and the question, and obtaining a document vector representation query-aware with question perception and a question vector representation para-aware with document perception; s5, learning an internal dependency relationship between the document and the question by using a self-attention mechanism, and obtaining a new semantic vectorrepresentation; s6, using the attention as a pointer, predicting the starting position and the ending position of the answer, and extracting an answer sequence according to the answer span.

The invention discloses a Chinese machine-reading system. The Chinese machine-reading system comprises a data grabbing module, a data processing module, a data extracting module, a knowledge base, a data integration module and a use interface, wherein the data extracting module comprises a wiki content extracting module, a template extracting module, an entity extracting module, a relation extracting module and a template matching module. Compared with the prior art, an open extracting method is used, the extracting field is not limited, unstructured text information widely existing on the Internet can be read, and the system is suitable for being popularized and used and can automatically adapt to evolution of Chinese language.

The invention discloses a generating method of a gama value test card of a display device and a measuring method of gama value. The methods mainly include determining the gray scale that the generated Gamma test card of the display device corresponds to and brightness of a standard picture area shown as the Gamma test card of the display device, determining gray scale voltage of the picture area to be tested according to the gray scale that the standard picture area in a display panel corresponds to, determining brightness of the picture area to be tested and displayed by the display panel under the effect of the gray scale voltage, comparing the brightness of the standard picture area and the picture area to be tested and determining the Gamma value of the display device under the gray scale. The measuring method is suitable for acquiring test results by utilizing a special test device in machine-reading mode and suitable for obtaining the test results by observing test results through naked eyes, and has the advantages of being convenient and quick to use, accurate in test result and the like.

The invention discloses an article title detection method based on a hierarchical hybrid network and a federated learning strategy, and the model comprises a title encoder which is used for carrying out feature extraction on article titles and effectively encoding article title texts into title vectors; The content encoder is used for carrying out feature extraction on the content text and effectively encoding the content text into a document vector; The associated information extractor is used for associating the title vector with the document vector by using a machine reading understanding related technology so as to obtain an associated vector of the title vector and the document vector; And the classification network is used for carrying out title codonopsis pilosula classification based on the title feature vectors, the document vectors and the association vectors, and a better title codonopsis pilosula detection effect can be obtained by utilizing the association information between the document titles and the document contents.

The invention relates to a machine reading comprehension method, an apparatus, a computer device and a storage medium. The method comprises the following steps of: obtaining a document paragraph set corresponding to a core statement from a database according to a core statement and a preset document positioning method; performing Text segmentation on the core sentence and the document paragraph set to obtain the core sentence after word segmentation and the document paragraph set after word segmentation. According to the corresponding relationship between the core sentence and the answer position, the answer paragraphs corresponding to the core sentence after word segmentation are determined from the document paragraph collection after word segmentation. The target answer paragraph corresponding to the core sentence after word segmentation is determined according to the probability value of each answer paragraph. This method can not only deal with the unstructured data, but also improve the accuracy of the target answer paragraph.