30 results about "Measure word" patented technology

A computer system for publishing content includes a categorization member, a rating module, a user profiling assembly and a suggested-reading engine. The categorization member enables an author-user and reader-users of an authored work to categorize content of the authored work using a fixed taxonomy, keywords, tags and / or keyword combinations. The rating module enables reader-users to rate authored works and makes a determination of quality of each authored work in the system. Rating includes reader-user response activity such as number of times the reader recommends, forwards or otherwise effectively promotes the authored work. The user profiling assembly measures words and word combinations used (written or read) by a user and generates therefrom a profile of the user. The suggested-reading engine is responsive to the rating module and provides a ranked list of suggested reading (authored works) for a user. The suggested reading is based on the user's ratings of authored works, respective relevance of authored works to the user and editorial promotion of certain articles / authored works.

The invention discloses an entity relation joint extraction method and device based on a neural network. The method comprises the following steps: firstly, fusing a pre-trained ERNIE word vector, a pre-trained CWV word vector and relative position information of words by utilizing a single-layer Transform network; secondly, improving a handshake labeling strategy, and introducing a vectorized entity type label to make full use of entity type information; then, obtaining candidate entity relationship triples in the sentences through an annotation decoding method; and finally, indexing articles, sentences, entities and relationships by using UCL according to the characteristics that the UCL national standard can efficiently organize contents and effectively associate information. The entity relationship triple can be directly extracted from the sentence, the problems of redundant entities, nested entities, overlapping relationships and the like are avoided, a small amount of annotation data can be used for quickly checking the entity relationship, and the data can be objectively and normatively indexed.

The invention relates to the field of artificial intelligence, in particular to a semantic matching method and device based on knowledge distillation, computer equipment and a storage medium. The method comprises the steps: obtaining a to-be-processed text and a standard text; respectively processing the to-be-processed text and the standard text to obtain corresponding character vectors, word vectors and character and word combination vectors; inputting the character vectors, the word vectors and the word combination vectors corresponding to the to-be-processed text and the standard text intoa pre-trained target semantic matching model, so as to calculate the target similarity of the to-be-processed text and the standard text through the target semantic matching model, wherein the targetsemantic matching model is obtained by training according to a twin network mode of knowledge distillation; and outputting a standard text corresponding to the to-be-processed text according to the target similarity. In addition, the invention also relates to the blockchain technology, and the standard text and the target semantic matching model can be stored in the blockchain. By adopting the method, the matching efficiency can be improved.

The invention belongs to the technical field of data processing, and discloses a processing system and method for extracting fine-grained typical opinion data of a user, which are used for cleaning the data and filtering out noise data. Performing character segmentation and word segmentation on the cleaned data; a word embedding model is used for training word vectors, word vector representation is carried out, and a corresponding file is generated; defining common attributes of the product; extracting attributes related to the product in the user comments; converting all the extracted attribute texts into vector representation by adopting a word vector weighted averaging method; clustering the texts; and obtaining user typical opinions of different attributes of the product. According tothe method, the attribute texts related to the product in the user comments are extracted, the texts with the same attribute and the same emotional tendency are clustered into the same cluster as muchas possible, and the user typical opinions with different attributes are obtained. The accuracy of the clustering result is effectively improved, the granularity of the clustering result is smaller,and the typical opinions of the user for different attributes of the product are quickly obtained.

The invention discloses a text classification method and system. According to the method, the limitation of Bert based on word coding is considered; internal relation information of characters in words is seriously lost. According to the invention, a word position coding mode is added on the basis of Bert; therefore, the obtained sentence vector is composed of a character vector, a sentence vector, a word vector and position information, sentence vector representation containing information of words is obtained, classification model training is conducted through the sentence vector training model, and finally the model is used for Chinese sentence classification.

The invention relates to the field of artificial intelligence, in particular to a semantic matching method and device based on a multi-task twin network, equipment and a medium. The method comprises the steps of obtaining a to-be-processed text and a standard text; inputting the to-be-processed text into a first feature extractor to obtain a word quantity, a word vector and a word combination vector corresponding to the to-be-processed text, and inputting the standard text into a second feature extractor to obtain a word vector, a word vector and a word combination vector corresponding to thestandard text, wherein the first feature extractor and the second feature extractor are twin networks; inputting the obtained character vector, word vector and word combination vector into a pre-trained target semantic matching model to obtain target similarity between the to-be-processed text and the standard text; and outputting a standard text corresponding to the to-be-processed text accordingto the target similarity. In addition, the invention also relates to the blockchain technology, and the standard text and the target semantic matching model can be stored in the blockchain. The method can improve the accuracy.

The embodiment of the invention discloses a semantic recognition method and device combined with knowledge graph entity information and related equipment. The method comprises the following steps: calculating a graph hidden state vector of each training entity in a training corpus, and an adjacent semantic vector, a word embedding vector and a position vector of each word, and combining the graph hidden state vector, the adjacent semantic vector, the word embedding vector and the position vector to obtain an input vector of each word; through predicting the input vector by an encoder of the Transform model, and obtaining the prediction probability of the semantics to which the input vector belongs; according to the probability between the prediction probability and the real semantics to which the input vector belongs, optimizing model parameters of the encoder to obtain a semantic recognition model; and finally, performing semantic recognition through the semantic recognition model. According to the method, the entity features obtained by the semantic recognition model are more comprehensive, so that the semantic recognition accuracy of the semantic recognition model is improved.

The invention provides a Chinese text word order adjustment and quantifier completion method and system. The method comprises the steps: inputting a word sequence in a Chinese corpus into an N-elementlanguage model, obtaining an N-element word list of the Chinese corpus, carrying out the quantifier labeling of the corpus in the Chinese corpus, and forming a quantifier list, deleting the labeled quantifier in the Chinese corpus, forming a parallel corpus with the Chinese corpus, and training the bidirectional long-short-term memory model by taking the parallel corpus as training data to obtaina quantifier completion model; performing part-of-speech tagging on a to-be-adjusted Chinese text, adjusting a statement structure and a sequence in the Chinese text according to a word sequence adjustment rule to form a text sequence adjustment candidate set composed of a plurality of new texts, performing cluster search in the text sequence adjustment candidate set by utilizing an N-element word list, and selecting words according to probability to obtain a text sequence adjustment result; and generating a statement with the maximum probability based on the Chinese corpus as a text sequencing result, and positioning and filling the missing position of the quantifier in the text sequencing result through a quantifier completion model.

The invention provides an input method and an input system for providing number selection for ordinal words, belonging to the field of input methods. On the basis of the traditional input method, the input system is additionally provided with a word bank containing the ordinal words and measure words and a determination unit for determining whether sequentially continuous ordinal words and measure words exist among characters input by a user or not according to the word bank. The input method comprises the following steps of: firstly, generating the word bank containing the ordinal words and the measure words; acquiring the characters input by the user by utilizing an input interface; determining whether sequentially continuous ordinal words and measure words exist among the characters input by the user or not according to the word bank; and adding the characters in candidate words corresponding to the characters input by the user according to an acquired determined result under the condition that the sequentially continuous ordinal words and measure words exist among the characters input by the user, thereby generating a candidate number and word list. By setting of the word bank containing the ordinal words and the measure words, the numbers and words input by the user and sequencing codes can be effectively identified, and thus an output result is enhanced when the ordinal words and the measure words are in continuation.

The invention relates to a key information identification method based on hierarchical attention and label guide learning, and belongs to the technical field of text mining and information processing. According to the method, a key information recognition framework based on hierarchical attention and label guide learning fusion is adopted, a text representation model is directly applied to limitation of text mining, a word coding layer and a sentence coding layer can fully capture a text organization structure, important words are aggregated into sentence vectors, and then the important sentence vectors are aggregated into text vectors; the word attention layer and the sentence attention layer apply attention mechanisms to a word level and a sentence level respectively, so that more important or secondary important contents can be concerned differently during text representation; a label-guided learning layer is adopted to execute label-based attention coding, text representation is mapped to a label space, and the label-guided learning layer can directly perform joint learning together with context coding. The method has a wide application prospect in the fields of quotation analysis, information retrieval, fine-grained knowledge service and the like.

The occurrence cost of unknown words that are not registered in a morphological analysis dictionary is calculated by applying an occurrence cost regression model, which is a learning model. An information processing device includes a notation feature amount extraction unit that extracts a notation feature amount of a character string, a part-of-speech feature amount extraction unit that extracts a part-of-speech feature amount of the character string, and an occurrence cost estimation unit that receives the notation feature amount and the part-of-speech feature amount and calculates an occurrence cost of the character string by applying an occurrence cost regression model. The occurrence cost regression model is a learning model that estimates the occurrence cost from the notation feature amount and the part-of speech feature amount of a character string, generated by a learning process using registration data of an existing morphological analysis dictionary as teacher data.

A machine translation method and an apparatus are provided. The method includes: obtaining, when translating a word fragment from a first language into a second language, a source representation vector of the word fragment. The source representation vector of the word fragment represents the word fragment in the first language. The method also includes obtaining a historical translation vector of the word fragment by querying historical translation information according to the source representation vector of the word fragment. The historical translation vector of the word fragment represents a historical translation situation corresponding to the word fragment. The method further includes translating the word fragment according to the historical translation vector of the word fragment.

The invention discloses a convolutional neural network relation extraction method fusing different pre-trained word vectors. Relation extraction is an important semantic processing task in the field of knowledge maps. At present, the most advanced method still depends on a pre-trained word vector and a natural language processing (NLP) tool, such as a Glove word vector and a word2vec word vector to obtain sentence representation, and depends on syntactic analysis and a named entity identifier (NER) to obtain advanced characteristics. However, only one word vector is used to solve the problem of polysemy of one word, and introduction of an NLP tool certainly brings errors. In order to solve the problems, the invention provides a convolutional neural network fusing different pre-trained wordvectors, two different pre-trained word vectors and relative entity distance vectors are used as network input, the most basic convolutional neural network is adopted, no natural language processingtool is used, and the method is simple and efficient.

The invention relates to a named entity recognition model training method, a named entity recognition model application method and a named entity recognition model training system, and belongs to the field of rail transit natural language processing, and the model training method comprises the following steps: preprocessing a fault text to obtain a word vector and a word vector; wherein the character vectors comprise the character vector of the named entity recognition task and the character vector of the word segmentation task; wherein the word vectors are used for judging whether two continuous word vectors in one sentence are associated into the same word or not; establishing a named entity recognition model; wherein the named entity recognition model comprises a named entity recognition task sub-model, a word segmentation task sub-model and an adversarial training structure; and alternately inputting the character vector of the named entity recognition task and the character vector of the word segmentation task into the adversarial training structure of the named entity recognition model for training to obtain a trained named entity recognition model. When the named entity recognition model is used for recognizing the category of the named entity, the recognition precision and the recognition effect are very high.

The embodiment of the invention discloses a semantic recognition method, device and related equipment combined with knowledge map entity information. The method includes calculating the hidden state vector of each training entity in the training corpus, as well as the adjacent semantic vector, word embedding vector and position vector of each word, and combining the hidden state vector, adjacent semantic vector, word embedding vector and position vector Obtain the input vector of each word; predict the input vector through the encoder of the Transformer model, and obtain the predicted probability of the semantics of the input vector; according to the probability between the predicted probability and the real semantics of the input vector, the model parameters of the encoder are calculated Optimized to obtain a semantic recognition model; finally, carry out semantic recognition through the semantic recognition model. The method makes the entity features acquired by the semantic recognition model more comprehensive, thereby improving the accuracy of semantic recognition by the semantic recognition model.

The invention discloses an internet text entity recognition method and system, electronic equipment and a storage medium. The method comprises the following steps: inputting a historical internet text into an entity recognition AI model to obtain an initialized full quantizer table; constructing a full quantizer dictionary tree according to the initialized full quantizer table; according to an entity recognition AI model and the full quantizer dictionary tree, performing recognition processing on the real-time sampled internet text to obtain a selected word list; constructing a selected word dictionary tree according to the selected word list; splitting the real-time internet text to be recognized according to preset Chinese sentence segmentation symbols to obtain split sub sentences; matching the split sub-sentences with the selected word dictionary tree to obtain matched sub-sentences; and splicing the matched clauses according to a preset sequence, inputting the spliced matched clauses into an entity recognition AI model to obtain an entity recognition result, and performing category output according to entity categories. The to-be-recognized real-time internet text is screened sentence by sentence according to the selected word list, and only sentences possibly containing entities are left, so that the calculated text quantity is greatly reduced, and the operation cost is reduced.

A computer system for publishing content includes a categorization member, a rating module, a user profiling assembly and a suggested-reading engine. The categorization member enables an author-user and reader-users of an authored work to categorize content of the authored work using a fixed taxonomy, keywords, tags and / or keyword combinations. The rating module enables reader-users to rate authored works and makes a determination of quality of each authored work in the system. Rating includes reader-user response activity such as number of times the reader recommends, forwards or otherwise effectively promotes the authored work. The user profiling assembly measures words and word combinations used (written or read) by a user and generates therefrom a profile of the user. The suggested-reading engine is responsive to the rating module and provides a ranked list of suggested reading (authored works) for a user. The suggested reading is based on the user's ratings of authored works, respective relevance of authored works to the user and editorial promotion of certain articles / authored works.

The invention discloses a 95598 customer appeal processing method and device and a storage medium. The processing method comprises the following steps: S101, data word segmentation: preprocessing voice translation text data of a customer, performing sentence segmentation on the data, and performing word segmentation on words of each sentence to obtain result data after word segmentation; s102, performing data vectorization, and generating word vectors by result data after word segmentation; s103, word vector processing: inputting the generated word vector into a customer appeal recognition algorithm model, extracting the core semantics of the text through a neural network structure, capturing the semantics of different words / words in the whole sentence meaning in the whole sentence meaning, and recognizing the appeal classification of the voice translation text data of the customer according to the semantics; and S104, outputting a result, and sending the result of the appeal classification to a corresponding processing end. According to the method, extraction of text semantics and generation of features can be realized under relatively low computing resources, and then different text semantic features are compared, so that the purpose of judging text contents is achieved.

The embodiment of the invention provides a text knowledge supplementing method and device based on a knowledge graph, and the method comprises the steps: obtaining and splicing a character vector, a word vector and a subject vector of a text, and inputting the character vector, the word vector and the subject vector into a bidirectional GRU to obtain a hidden state; processing the hidden state by using a self-attention mechanism to obtain a feature matrix, and converting the feature matrix into a feature vector through a pooling layer; calling a knowledge graph to carry out conceptualization processing on the text to obtain a concept set comprising concept vectors; calculating a relation weight between the concept vector and the feature vector; calculating the importance weight of the concept vector in the concept set; and adjusting the corresponding relation weights by using the importance weights, and performing weighted calculation on each concept vector according to the adjusted relation weights to obtain concept set features so as to perform knowledge supplement on the text through the concept set features. According to the method, the text features are expanded from the aspects of character granularity, word granularity and text granularity, and the deficiency of text context information is made up by means of the knowledge graph.

The invention provides an extraction method and a retrieval method of customs data product words. The extraction method comprises the following steps: firstly, cleaning up redundant parts in a customsdescription text, and converting the redundant parts into a better processing form; then, heuristically finding out segmentation words in the customs description text, and segmenting product words and description parts; replacing quantifiers and date regularities in the text with space characters or deleting the quantifiers and date regularities; deleting a description part in the text through agrammatical rule, or extracting product word groups from the data by using mutual information and left and right information entropies to obtain product word groups of which the number of words is less than or equal to 5, and adding the product word groups into a lexicon. The retrieval method comprises the steps that firstly, word segmentation is conducted on a text to be retrieved, and then retrieval is conducted in a constructed word bank through a bit map or hash map structure. According to the method, the grammatical structure, the mutual information, the character information and the specific structure information of the customs data are combined, the advantages of various information can be fully combined, and the product words can be accurately extracted and retrieved.

A cloud computing platform for implementing semantic search provided by the embodiment of the present application includes: an original document acquisition module, used to acquire the original document; an index establishment module, used to establish corresponding word vectors according to the original document, and generate an index, based on which the index can be Determine the corresponding original document; the search engine module is used to receive the search information input by the user; the word vector extraction module is used to extract the word vector of the search information; the word vector matching module is used to combine the word vector of the search information with the original document The word vector is matched, the target index of the original document corresponding to the search information is determined, and the target original document is determined according to the target index; the search information feedback module is used to feed back the target original document to the user. The cloud computing platform and balancing method for implementing semantic search in the embodiments of the present application have accurate search results, can effectively solve practical problems, and at the same time speed up the response speed and improve user experience.

The degree of relatedness between words included in an amount of data can be calculated, allowing suitable related words and phrases to be extracted. An inter-word score calculation apparatus includes document data (response history document) wherein documents input from outside are accumulated, term list data wherein predetermined terms are written, a word combination unit that can execute a combination process of amplifying an amplification candidate word, which is a word corresponding to a term written in the term list data and included in a document constituting the document data and adding the amplified word to the document data, and an inter-word score calculation unit calculating a degree of relatedness between words in the document data using a predetermined calculation method, wherein the number of documents accumulated in the document data is smaller than a first predetermined amount, the word combination unit adds the amplification candidate word to the document data.

The invention discloses a first-order parallel and pointwise circumscription theory calculation method with priority. The method includes the steps of firstly, translating any first-order circumscription theory into another first-order theory under the stable semantics within the linear time, wherein the first-order circumscription theory and the first-order theory are equivalent in logic on any structure, and four main translations based on the grammatical level are involved; secondly, putting forward an optimized translation algorithm used for eliminating existential quantifiers for the existential quantifiers in the first-order circumscription theory, wherein by means of the method, the number of introduced auxiliary verbs is decreased, and the theory scale increase caused by translation is reduced; thirdly, obtaining a universal first-order circumscription theory solver which can obtain all extremely-small models in a given theory field on the basis of the translation and the optimized existential quantifier eliminating algorithm. The current situation of shortage of a first-order parallel pointwise circumscription theory solver with priority is eliminated, and the first-order parallel pointwise circumscription theory solver which can conduct efficient calculation can be designed and obtained.

The invention relates to a Chinese text data word vector representation method based on a BIE position word list, and relates to the field of deep learning and named entity recognition, and the method comprises the following steps: S1, generating a total word set and a strong correlation word set, and constructing the BIE position word list; s2, constructing a position-independent word vector by using the original representation of the word vector; s3, condensing word vector representations in the word set based on a word frequency weighted average pooling algorithm; and S4, weighting the BIE position word vector of the word and splicing the weighted BIE position word vector with the original word vector to generate a word vector containing vocabulary position information. According to the method, the position information of the highly correlated vocabularies can be highlighted while the total position information of the vocabularies is fused into the word vectors. And the character vector representation dimension is expanded, so that the Chinese entity recognition result has higher accuracy.

The invention discloses a text sentiment analysis method based on word vector deformation and bidirectional bit order convolution, which comprises the following steps: obtaining an evaluation statement, inputting the evaluation statement into a trained bidirectional standard convolution network model with bit order information, and outputting a text sentiment analysis result through the model; wherein the bidirectional standard convolutional network model with bit sequence information comprises a word embedding layer, a bit sequence information layer, a word vector deformation layer, a convolutional layer, a double-end structure and a classification layer; the word embedding layer is used for converting words in an evaluation statement into word vectors which can be understood by a computer; the bit sequence information layer is used for adding bit sequence information to the word vector to obtain the word vector with the bit sequence information; the word vector deformation layer is used for deforming word vectors into word block matrixes and finally splicing the word block matrixes into sentence matrixes; the convolution layer is used for performing convolution operation on the sentence matrix to obtain word vector features; and the classification layer is used for carrying out classification operation on the spliced word vector features to obtain a final text sentiment analysis result.