725 results about "Semantic vector" patented technology

A system and method are presented for the comparative analysis of textual documents. In an exemplary embodiment of the present invention the method includes accessing two or more documents, performing a linguistic analysis on each document, outputting a quantified representation of a semantic content of each document, and comparing the quantified representations using a defined metric. In exemplary embodiments of the present invention such a metric can measure relative semantic closeness or distance of two documents. In exemplary embodiments of the present invention the semantic content of a document can be expressed as a semantic vector. The format of a semantic vector is flexible, and in exemplary embodiments of the present invention it and any metric used to operate on it can be adapted and optimized to the type and / or domain of documents being analyzed and the goals of the comparison.

An automatic file clustering algorithm enables documents within a file system to be displayed in a semantic view. The file clustering algorithm maps all words and documents into an appropriate semantic vector space, clusters the documents at a predetermined level of granularity, and assigns a meaningful descriptor to each resulting cluster. The documents are displayed to the user in a hierarchy in accordance with the resulting clusters. This results in a virtual file system with a semantic organization, that allows the user to navigate by content.

The invention discloses a Chinese named entity recognition method based on BERT-BiGRU-CRF. The method comprises three stages: in the first stage, preprocessing mass text corpora, and pre-training a BERT language model; in the second stage, preprocessing the named entity recognition corpus, and encoding the named entity recognition corpus through the trained BERT language model; and at the third stage, inputting the encoded corpus into a BiGRU+CRF model for training, and performing named entity recognition on the to-be-recognized statement by using the trained model. Construction of the Chinesenamed entity recognition method based on BERT-BiGRU-CRF is carried out, semantic representation of characters is enhanced through a BERT pre-training language model, semantic vectors are dynamicallygenerated according to contexts of the characters, and the ambiguity of the characters is effectively represented. Compared with a method based on fine tuning of a language model, the method has the advantages that training parameters are reduced, and the training time is saved.

The invention relates to the field of knowledge extraction and particularly relates to a knowledge extraction method and system based on a memory neural network and a device. The invention aims to solve the problem of information redundancy existing in the prior art. The method comprises a step of obtaining a possible relationship type in an input text and a semantic coding vector by using a convolutional neural network under the premise of giving a predefined relationship type, a step of carrying out semantic encoding by using a two-way long and short time memory neural network and obtaininga semantic vector, a step of taking the relationship type as an initial value of the two-way long and short time memory network and a first label in a decoding module and thus fusing the relationshiptype information into encoding information and label information of the decoding module, and a step of obtaining a label sequence by using a decoding module of a single-way long and short time memorynetwork and then obtaining structured information by parsing the label sequence. According to the knowledge extraction method and system and the device, the efficiency of structured information extraction is greatly improved, and the problem of information redundancy existing in the prior art is solved.

The invention belongs to the technical field of text semantic processing and semantic webs, and particularly relates to a knowledge graph management method and system based on semantic space mapping. The method comprises the steps of semantic vector construction, semantic space mapping and knowledge graph management, wherein the step of knowledge graph management comprises three sub-steps of semantic clustering, semantic duplication eliminating and semantic annotation. A text unit describing edge / nodal points of a knowledge graph is projected to a semantic space, and vector representation of the edge / nodal points on the semantic space is obtained by vector accumulation; on the basis, multiple management tasks of the knowledge graph are achieved. The system correspondingly comprises a semantic vector construction module, a semantic space mapping module and a knowledge graph management module. The defects that a conventional knowledge graph management method is sensitive to factors such as word deformation, synonym variation and grammatical form variation are overcome, the situation of difference of the number of words can be easily handled in a vector accumulation mode, and further knowledge graph management tasks such as semantic clustering, semantic duplication eliminating and semantic annotation are easily achieved.

The invention provides a text processing method and device based on ambiguity entity words. The method comprises the steps of obtaining the context of a text with ambiguities to be eliminated and at least two candidate entities represented by the text with the ambiguities to be eliminated, generating a semantic vector of the context through a trained word vector model, generating a first entity vector of at least two candidate entities through a trained non-supervision neural network model, calculating the similarity between the context and each candidate entity, and determining a target entity represented in the context of the text with the ambiguities to be eliminated. On the basis of the learned non-supervision neural network model of the entity text semantics and the relationship between the entities, a first entity vector of the generated candidate entities includes the text semantics of the candidate entities and the relation between the entities, entity information of the text with the ambiguities to be eliminated is completely described, the similarity is calculated with the context semantic vectors, the target entity is determined, and the ambiguity eliminating accuracy ofthe text with the ambiguities to be eliminated is improved.

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 geographical science domain named entity recognition method, which is used for recognizing geographical science core term entities and geographical location entities. The method mainly comprises three steps of (1) establishing a geographical science domain dictionary, and using a new word discovery algorithm to identify new words in the geographical science domain in an unsupervised way; (2) training and testing based on a conditional random field (CRF) model and a multichannel convolutional neural network (MCCNN) model; (3) carrying out error correcting and fusion on entities recognized by the models by using a rule-based method. According to the geographical science domain named entity recognition method, the new words of the domain are identified as the dictionary in an unsupervised way by using the new word discovery algorithm, so that the work distinguishing effect is improved. The semantic vectors of the words are learnt from large-scale unmarked data in an unsupervised way, and basic characteristics of the words are synthesized and are taken as the input characteristics of the MCCNN model, so that manual selection and construction of the characteristics are avoided. The predicting results of the two models are fused by means of a custom rule, so that the problem of error marking in a recognition process can be corrected.

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

An apparatus and method are disclosed for producing a semantic representation of information in a semantic space. The information is first represented in a table that stores values which indicate a relationship with predetermined categories. The categories correspond to dimensions in the semantic space. The significance of the information with respect to the predetermined categories is then determined. A trainable semantic vector (TSV) is constructed to provide a semantic representation of the information. The TSV has dimensions equal to the number of predetermined categories and represents the significance of the information relative to each of the predetermined categories. Various types of manipulation and analysis, such as searching, classification, and clustering, can subsequently be performed on a semantic level.

Embodiments of the invention disclose a search recall method and device, a server and a storage medium. The search recall method comprises the following steps of: obtaining a retrieval word input by a user; calculating a semantic vector of the retrieval word by utilizing a pre-trained neural network model; and recalling a target document related to the semantic vector of the retrieval word from a candidate document according to a pre-established index, wherein the index is established according to a semantic vector of the candidate document, and the semantic vector of the candidate document is calculated according to the pre-trained neural network model. According to the method, the problem that existing recall method is lack of semantic generalization ability so as to influence the recall correctness can be solved, so that the search recall correctness is improved.

In a method of semantic querying in a peer-to-peer network, an item of information is mapped into a semantic vector based on the latent semantic indexing algorithm or any IR algorithms that can derive a vector representation. The semantic vector is associated with an address index as a key pair. The key pair is stored in an overlay network formed from the peer-to-peer network such that the stored key pair is proximally located to at least one other key pair having a similar semantic vector.

The invention discloses a translation model establishing method and system. The translation model establishing method comprises the following steps: respectively generating a regular alignment table, a word semantic vector table and a phrase table according to alignment information of a double-language parallel corpus, subsequently generating a source language phrase semantic vector table of a source language semantic space and a target language phrase semantic vector table of a target language semantic space by using the word semantic vector table and the phrase table, and finally training by using phrase semantic vector tables of different semantic spaces, thereby generating a translation model integrated with semantic information. The result shows that phrase semantic information can be integrated in statistic machine translation, the research shows that the relevance of words or phrases to context words or phrases can be reflected in the semantic information, and compared with a conventional translation method based on words or phrases, the translation model is relatively high in translation quality after the phrase semantic information is integrated, so that the translation property of the statistic machine translation is further improved as compared with that of the prior art.

The invention belongs to the technical field of natural language processing, and discloses multi-label text classification processing method and system and an information data processing terminal. Themethod comprises the steps of: obtaining a data set; preprocessing the data set and dividing the data set into a training set and a test set; finely adjusting and extracting global feature vectors ofwords in the text sequence through a BERT pre-training model, and aggregating the global feature vectors by adopting a convolutional neural network to obtain semantic vectors of the words in the textsequence; constructing an attention weight coefficient matrix, and respectively weighting the semantic vector of each word and a weight coefficient vector in the optimal weight coefficient matrix toobtain an attention vector of the label; and normalizing the attention vectors of the tags to obtain the probability of each tag, and selecting several tags with the maximum probability as the category of the text. According to the method, global and local features of the text sequence are extracted, the influence of keywords in the text on tag categories is considered, and the classification accuracy is improved.

The invention discloses a searching method and searching equipment. The searching method comprises the following steps: receiving a search keyword input by a user, acquiring at least one search document relating to the search keyword based on the search keyword, and determining the semantic vector of the search keyword; according to the semantic vector of the search keyword and the semantic vector of each search document, calculating the semantic correlation between each search document and the search keyword; according to the semantic correlation, sequencing the search documents and outputting the search documents to the user according to the sequencing result. The semantic vector of the document of a search engine is acquired according to the semantic vector of a document tile, the semantic vector of a document identifier and the semantic vector of a document category, so that the semantic information of the document can be expressed accurately, the precision of the semantic correlation between the document acquired by calculation and the search keyword is improved, the search precision of the search engine is improved, and the user experience on the search engine is improved.

The invention discloses an automatic text summarization method based on enhanced semantics. The method comprises the following steps of: preprocessing a text, arranging words from high to low according to the word frequency information, and converting the words to id; using a single-layer bi-directional LSTM to encode the input sequence and extracting text information features; using a single-layer unidirectional LSTM to decode the encoded text semantic vector to obtain the hidden layer state; calculating a context vector to extract the information, most useful the current output, from the input sequence; after decoding, obtaining the probability distribution of the size of a word list, and adopting a strategy to select summarization words; in the training phase, fusing the semantic similarity between the generated summarization and the source text to calculate the loss, so as to improve the semantic similarity between the summarization and the source text. The invention utilizes the LSTM depth learning model to characterize the text, integrates the semantic relation of the context, enhances the semantic relation between the summarization and the source text, and generates the summarization which is more suitable for the subject idea of the text, and has a wide application prospect.

In a method of semantic querying in a peer-to-peer network, an item of information is mapped into a semantic vector based on the latent semantic indexing algorithm or any IR algorithms that can derive a vector representation. The semantic vector is associated with an address index as a key pair. The key pair is stored in an overlay network formed from the peer-to-peer network such that the stored key pair is proximally located to at least one other key pair having a similar semantic vector.

The invention discloses a reinforcement dual-channel sequence learning-based dialog reply generation method and system. The method comprises the following steps of: (1) modeling a context to obtain acontext semantic vector; (2) carrying out combined learning on a current dialog and the context semantic vector by utilizing an encoder so as to current a current dialog vector and an encoder vector;(3) inputting the context semantic vector and the current dialog vector into a decoder so as to obtain a first channel dialog reply draft and a decoder vector; (4) inputting the encoder vector, the decoder vector and the first cannel dialog reply draft into an embellishing device to carry out embellishing, so as to generate an embellished dialog reply of a second channel; (5) optimizing a target function by utilizing a reinforcement learning algorithm; and (6) ending model training and generating and outputting a dialog reply. By utilizing the method and system, dialog generation models can grasp global information more deeply, and replies more according with dialog scenes and having substantial contents can be generated.

The invention discloses a multi-label text classification method and system. The method comprises the following steps: obtaining a training set comprising a text sequence and a label space, extractingglobal feature vectors of all words in the text sequence by adopting a long-short time memory network, and aggregating the obtained global feature vectors by adopting a convolutional neural network to obtain a semantic vector of each word in the text sequence; respectively calculating weight coefficients of each label in the note space and all words in the text sequence, constructing an attentionweight coefficient matrix, and processing the attention weight coefficient matrix to obtain an optimal weight coefficient matrix; respectively weighting the semantic vector of each word and the weight coefficient vector in the optimal weight coefficient matrix to obtain an attention vector of the tag; and performing normalization processing on the attention vectors of the tags to obtain the probability of each tag, and selecting several tags with maximum probabilities to classify the text.

The invention discloses an answer selection method for non-fact questions and answers, and belongs to the technical field of intelligent retrieval. The method comprises the steps of processing a question sentence and a to-be-selected answer statement by adopting a convolutional neural network based on an attention mechanism to obtain a first semantic vector and a second semantic vector, wherein the first semantic vector represents a semantic vector of the question sentence, and the second semantic vector represents a semantic vector of the to-be-selected answer statement; and matching the first semantic vector with the second semantic vector, and returning the to-be-selected answer statement corresponding to the second semantic vector with the highest matching degree as a correct answer. According to the method, the semantic vector representation of the sentence is generated by utilizing the convolutional neural network based on the attention mechanism, and a weight is given to an input by an importance degree in the semantic representation process, so that the influence of contents irrelevant to an answer theme is reduced, the high-quality semantic representation is automaticallygenerated, the semantic matching degree between the question sentence and the answer statement is improved, and the correct rate of answer selection is improved.

The embodiment of the invention discloses a searching method and device based on artificial intelligence. The method comprises the steps of obtaining at least one candidate document related to a searching sentence; determining a searching word vector sequence corresponding to a segmented word sequence of the searching sentence and determining a candidate document word vector sequence corresponding to the segmented word sequence of each candidate document in the at least one candidate document; executing the step of similarity calculation on each candidate document in the at least one candidate document; selecting a preset number of candidate documents as searching results from the at least one candidate document according to the sequence of the similarity with the searching sentence from large to small. According to the searching method and device based on the artificial intelligence, the candidate documents and the searching sentence are represented into a semantic vector through a neural network model, then according to similarities between a candidate document semantic vector and a searching sentence semantic vector, the candidate document is ranked, and the relevance of the searching results which are called back and the searching sentence input by a user is improved.

The embodiment of the invention discloses a processing method and device of a user portrait, a server and a storage medium. The processing method of the user portrait includes: acquiring user historyinformation; and converting the user history information into a semantic vector, which corresponds to a user, on the basis of a language model, and using the same as the user portrait of the user. According to the embodiment of the invention, the user portrait is expressed in a manner of the semantic vector, expression is accurate, a data form is unified, operations of introducing the semantic vector under various application scenes for calculation are facilitated, thus a factor of the user portrait is considered, and effects of accurate and concise meaning expression and easy application areachieved.

The invention discloses a plain text oriented enterprise entity classification method. The plain text oriented enterprise entity classification method comprises the steps of S1, carrying out type labeling for the enterprise entities in collected plain text data and regarding the enterprise entities being subjected to type labeling as a training set of an enterprise entity identification module; carrying out type labeling for the enterprise entities in the collected plain text data according to business nature and regarding the enterprise entities being subjected to the type labeling as a training sample set of an enterprise entity classification module; and S2, carrying out enterprise entity identification model training through a condition random field model to obtain an enterprise entity identification model; S3, carrying out semantic vectorization construction for the text data of an original training set; S4, training by regarding the data of the training set after being subjected to type labeling and semantic vectorization as training parameters to obtain an enterprise entity classification model; and S5, classifying the enterprise entity in a to-be-predicted text by utilizing the enterprise entity classification model. According to the plain text oriented enterprise entity classification method, as the obtained semantic vector serves as the feature of the entity, dependence on artificial features and external data is reduced, and the universality and robustness are guaranteed.

The invention discloses a multi-angle and multi-mode fused image description generation method and system, and the method comprises the following steps: receiving a to-be-described image, extracting the global visual features and local visual features of the image, and carrying out the fusion of the global visual features and local visual features, and obtaining fused visual features; using a single-layer long-short-term memory network, the fused visual features serving as input, and obtaining a first sentence of image description; generating a first sentence semantic vector according to the first sentence image description; and generating a next image description sentence by adopting an attention-based long-term and short-term memory network language generation model and taking the localvisual features and the first sentence semantic vector as input, thereby obtaining complete image description. According to the method, two modes of visual features and text semantic features are fused, and an attention mechanism is combined, so that multi-angle comprehensive description of the image is realized.

The invention relates to a multimode intelligent analysis method and a system. The disease description is multimode data, and the multimode data comprises character data, time sequence signal data and visual data; the method can be applied to the diagnosis navigation and diagnosis decision according to the disease situation; the method includes steps of converting visual data to character description through an image-text decoder; converting a time sequence signal to the text description through a signal character decoder; respectively converting the character data, the character description of the visual data and the time sequence signal data to a semantic vector taking a medical treatment mapping knowledge domain as an axis, wherein the medical treatment mapping knowledge domain describes the topological relation among the symptom, laboratorial index, disease, drug and operation; according to the semantic vector, a first convolution neutral network CNN model is applied to perform diagnosis navigation and diagnosis decision. According to the embodiment, the best diagnosis path and the diagnosis result with high accuracy can be acquired.

The invention discloses a semantic meaning-based specific task text keyword extraction method, and belongs to the field of natural language processing. The method comprises the following steps of: firstly, extracting a subject word of a certain specific task in a related text, and converting the subject word into a semantic vector by utilizing a semantic representation technology; secondly, carrying out word segmentation, part-of-speech tagging and screening on a text of a to-be-extracted keyword by utilizing a word segmentation tool; thirdly, converting the screened words into semantic vectors and calculating a similarity between each screened word and the subject word of the specific task; and finally, constructing a word network chart by taking the words as nodes, and calculating the importance degree of each word on the basis of the word similarity so as to extract important words in the word network chart. The method disclosed by the invention comprehensively considers the semantic features and structural features of the words in the texts, and is suitable for the extraction of specific-task oriented text keywords, so as to realize a function of obtaining important information from the texts and provide important technical support for the field of text mining, natural language processing, knowledge engineering and the like.

The invention discloses a semantic vector map construction method based on visual point cloud, a semantic vector map construction device based on visual point cloud and an electronic device. Accordingto an embodiment, a method for constructing a semantic map based on a visual point cloud includes performing object detection on an image acquired by an image acquisition device to acquire pixel objects in the image and attribute information thereof; Determining position information of each pixel target in the image; Combining attribute information and position information of each pixel object togenerate a point cloud with semantics; And a semantic vector map is constructed based on the point cloud with semantics. The semantic vector map construction method of the present application can complete the high-definition map construction at a very low cost by only using images and a small amount of prior information of external sensors.

The invention discloses a network rumor detection method based on a multi-modal relationship. The network rumor detection method comprises the steps of obtaining a to-be-detected image and a related text published on a network platform; extracting visual feature vectors containing different types of objects in the image through a pre-training factor R-CNN model; after the text is preprocessed, performing semantic vector extraction through a GRU; capturing importance degrees of the visual feature vector and the semantic vector through an attention mechanism, and realizing cross-modal association between the image and the text so as to update the visual feature vector and the semantic vector; moreover, for the visual feature vector and the semantic vector, the relationship of internal dynamic information is modeled through an attention mechanism, so that the visual feature vector and the semantic vector are updated; and connecting the visual feature vector and the semantic vector obtained by updating the two parts together, and obtaining the probability that the information to be detected is the rumor and the real category through a binary classifier. The method can automatically judge whether the to-be-detected information belongs to the network rumors, and has relatively high detection accuracy.

The invention relates to a text abstract generation method based on a sentence association attention mechanism, and belongs to the technical field of natural language processing. The method comprisesthe following steps: firstly, encoding a document by using a layered bi-directional long short-term memory Bi-LSTM network to obtain sentence semantic vectors, then analyzing an association relationship among sentences by virtue of a gating network to realize sentence-level importance and redundancy evaluation, and finally, providing a decoding algorithm based on a sentence association attention mechanism to generate an abstract. When a neural network abstract generation framework is constructed, sentence relevance analysis is fused, and the evaluation capacity of the model for sentence importance and redundancy in an original text is improved. According to the method, the performance of the generative abstract is effectively improved, and a relatively good effect is achieved on the current ROUGH evaluation index.

The invention discloses a recommendation method and device of media content, a storage medium, and an electronic device. The method comprises the following steps: acquiring a recommendation request, wherein the recommendation request is used for requesting to recommend media content to the target object; acquiring a subject word from the word of the first content in response to the recommendationrequest, wherein the first similarity of the word vector of the subject word and the semantic vector of the first content is greater than or equal to the second similarity, the semantic vector of thefirst content is determined according to the word vector of each keyword in multiple keywords and the weight of each keyword, multiple keywords are the keywords in the words of the first content; selecting the second content matched with the subject word from candidate media content, and recommending the second content to the target object. The technical problem that the recommended media contentis low in accuracy rate in the related technology is solved.