144 results about "Label" patented technology

The invention discloses a multi-label text classification method based on statistics and a pre-training language model. The multi-label text classification method comprises the following steps: S1, preprocessing training corpora needing to be classified; S2, establishing a label acquisition model based on a statistical method and a language model; S3, processing the obtained label data; S4, establishing a multi-label classification model based on a pre-training language model, and performing model training by utilizing the obtained label data; S5, performing multi-label classification on the text data to be classified by using the trained multi-label text classification model. According to the method, a statistical method and a pre-trained language model label obtaining method are combined, the ALBERT language model is used for obtaining the semantic coding information of the text, a data set does not need to be manually labeled, and the label obtaining accuracy can be improved.

The invention provides a recommendation processing method and device based on artificial intelligence, equipment and a storage medium. The method comprises the steps of obtaining object features of ato-be-recommended object, and obtaining label word features of label words in an article corresponding to the to-be-recommended object; based on the object characteristics corresponding to the to-be-recommended object and the label word characteristics corresponding to the article, determining the relevancy between the label word and the to-be-recommended object; when the relevancy is greater thana relevancy threshold value, determining that the label word and the to-be-recommended object have an association relationship; determining the label words associated with the to-be-recommended object as user labels of user accounts associated with the to-be-recommended object; and based on the user label of the user account, executing recommendation operation corresponding to the user account. Through the method and the device, the user label of the user account can be automatically obtained so as to carry out targeted recommendation.

The invention provides an event extraction method. The method includes obtaining original corpora, labeling elements in sentences in the form of [label-element], dividing the marked corpus into a training set and a test set; mapping the training set into a vector by a pre-training language model; obtaining word embedding vectors, inputting the word embedding vector into a neural network model; outputting sequence label information by the neural network model; establishing a loss function based on the sequence label information, evaluating the event extraction model by using the test set to finally obtain an event extraction model, adjusting the training set and the test set structure for multiple times, selecting the event extraction model with the best evaluation result as an optimal model, and inputting a new original corpus into the optimal model to obtain an event extraction result; the event extraction method provided by the invention belongs to a supervised neural network extraction method, is combined with the fine-grained contextualized word vector in the field, and more conforms to the scene of event extraction in the professional field.

The invention discloses a vulnerability detection method and device, electronic equipment and a computer readable storage medium, and the method comprises the steps: obtaining a script type label vector of a to-be-detected system through a first classifier according to a system feature vector of the to-be-detected system; obtaining structural data of script codes of the to-be-tested system througha second classifier according to the script category label vector of the to-be-tested system; and carrying out vulnerability detection based on the structure data of the script code of the to-be-detected system. Thus, high-accuracy and comprehensive vulnerability detection can be carried out in a continuously increasing and changing network environment.

The invention belongs to the field of natural language processing, and relates to a method and a system for establishing a text label system and completing iteration, and a storage medium. The methodcomprises the following steps that firstly, a label system is preliminarily built, class labels are concluded, at least one feature word used for describing labels is configured for each label, and aprimary label system and a configuration word bank of the primary label system are formed; the label system is verified and coverage conditions of the current label system and the configuration lexicon thereof are evaluated; the label system is improved and text clustering and information extraction are carried out on the text data which is not covered by the configuration word bank of the currentlabel system again to obtain a new batch of feature words and description words; and the current label system is compared with the configuration word bank thereof based on the text similarity to merge the feature words with high similarity, and class labels are named for newly discovered description words to obtain a latest label system. According to the invention, automatic extraction and integration of labels are realized, the constructed system can be self-perfected and optimized, and manual intervention is reduced.

The invention discloses a label extraction method based on short text clustering. The label extraction method comprises: S1, extracting and obtaining all useful vocabularies of a short text; S2, calculating the similarity among the short texts by utilizing word2vec according to the text characteristics of the useful vocabularies of each short text; S3, assuming that each short text is a cluster center, calculating the similarity between each cluster center and all other short texts, and if the similarity is greater than a preset threshold T1, classifying the corresponding short text into the cluster center; S4, trimming each cluster formed in the S3 for the first time; S5, combining all the clusters trimmed for the first time; S6, performing secondary pruning on the clusters merged in theS5; and S7, extracting the cluster center of each cluster after the second trimming as a label of each cluster. Labels which can represent meanings of all types of short texts and have complete meanings are generated while the short texts are clustered, and the problems that an existing clustering technology depends on center point selection and an existing label extraction technology is incomplete in semantics, does not accord with logic and the like are solved.

The invention provides a text annotation method and device, and the method comprises the steps: receiving a triggering instruction for triggering the annotation of a to-be-annotated target text; displaying a target label list at a first preset position of a page where the target text is located according to the trigger instruction; receiving a target selection instruction for selecting one or more target tags in the label list, and receiving a dragging instruction for dragging the selected one or more target labels into the target text; and labeling the one or more target tags for the targettext according to the dragging instruction. According to the invention, the problem that manual annotation of the text is tedious in the prior art can be solved. By directly selecting the target labelfrom the label list, annotation of the text label can be completed, the manual annotation mode is simplified, the method is different from a common single-label classification text, text classification is expanded into text classification annotation supporting multiple labels, and the text classification of the multiple labels is faster.

The invention provides a text classification method and device, and the method comprises the steps: determining a plurality of basic keywords in a to-be-classified text, determining the importance degree of the basic keywords in the text, when a label keyword located in a preset label classification keyword table is determined in the multiple basic keywords, weighing the basic weight of the labelkeyword to increase the weight of the label keyword, meanwhile, determining related keywords associated with the label keyword based on a preset word vector model, determining the related keywords associated with the label keyword based on a preset word vector model so that missing of low-frequency keywords is prevented, and finally, comprehensively determining the classification result of the text according to the corresponding relationship between the keywords in the tag classification keyword table and the classification result, the basic keywords, the tag keywords, the related keywords andthe respective corresponding weight values so as to improve the accuracy of the classification result.

The invention discloses an information hiding and extracting method based on an HTML (Hypertext Markup Language) label attribute sequence. The method comprises the following steps: step 1, determining to-be-hidden information, and converting the to-be-hidden information into a binary Unicode; step 2, determining representing methods of '0' and '1' corresponding label attributes, wherein '0' is represented by two attributes of two different labels, and '1' is represented by two attributes of a same label; step 3, determining a sequence code K, wherein the Unicode consists of 8 binary numbers; selecting nine '1' (one more selected), randomly inserting a plurality of '0' among the nine '1' so as to form the current-time information hiding sequence code K; step 4, determining nine label attributes according to the sequence code K, sequentially putting the nine label attributes on the '1' positions, supplementing residual label attributes, sequentially putting the residual label attributes on the '0' positions so as to synthetize an HTML file label attribute sequence; and step 5, editing codes according to the HTML file label attribute sequence. The method has the beneficial effects that a database is not needed for support, and the webpage execution efficiency is greatly improved.

The invention relates to a noise sample denoising method and device, computer equipment, a storage medium and a model training method. The noise sample denoising method comprises the steps: firstly obtaining a noise sample sequence and first label data corresponding to the noise sample sequence; then, correcting the first label data through a transfer matrix to obtain corrected second label data,thereby improving the accuracy of the label data; and therefore, according to the noise sample sequence, the first label data and the second label data, generating a sample sequence with correct labels through a conditional variation auto-encoder model, further improving the proportion of accurate label samples in the noise sample sequence, and improving the accuracy of the model.

The embodiment of the invention provides a user classification method and device. The method comprises the steps of obtaining an identifier set formed by application identifiers of applications installed by a target user, a word segmentation set after name word segmentation of the applications, and a label set formed by category labels of the applications; inputting the identifier set, the segmented word set and the label set into a pre-trained neural network model, obtaining a first embedded vector corresponding to each application identifier in the identifier set, obtaining a second embeddedvector corresponding to each segmented word in the segmented word set, and obtaining a third embedded vector corresponding to each category label in the label set; performing weighting processing oneach first embedded vector to obtain a fourth embedded vector; performing weighting processing on each second embedded vector to obtain a fifth embedded vector; performing weighting processing on eachthird embedded vector to obtain a sixth embedded vector; and determining a category corresponding to the target user according to the fourth embedded vector, the fifth embedded vector and the sixth embedded vector. The users can be accurately classified.

The embodiment of the invention provides a text similarity calculation method and device. According to the embodiment of the invention, the method comprises: respectively determining a label keyword set and a non-label keyword set in a first text and a second text; determining a first similarity between a first label keyword set of a first text and a second label keyword set of a second text basedon a preset hierarchical tree used for representing an association relationship between keywords; determining a second similarity between the first non-label keyword set of the first text and the second non-label keyword set of the second text based on a preset semantic model; and finally, determining the text similarity between the first text and the second text according to the first similarityand the second similarity, and calculating the similarity between the texts by extracting keywords in the texts and utilizing the keywords, thereby achieving the purpose of quickly calculating the text similarity.

The embodiment of the invention provides a model adversarial training method and device and a named entity recognition method and device. The adversarial training method comprises steps of during training, replacing a first named entity in the first sample sequence with a corresponding original label character to obtain a second sample sequence, determining a first text fragment containing the replaced original label character from the second sample sequence, and determining a classification label value of the first text fragment as a first value for representing a replaced named entity; determining feature implicit vectors of a plurality of segmented words in the second sample sequence by adopting a feature extraction network; determining a first segment vector of the first text segment based on the feature implicit vectors of the plurality of segmented words in the second sample sequence, and inputting the first segment vector into a first discriminator to obtain a first prediction value; determining a first loss value based on a difference between the first prediction value and the first value; updating the first discriminator by taking minimization of the first loss value as atarget; and updating the feature extraction network by taking maximization of the first loss value as a target.

The invention discloses a method and device for generating codes based on annotations and a storage medium, and relates to the technical field of research and development management, and the method comprises the following steps: obtaining an interface document to be converted into codes; determining a first target class to be converted in the interface document, and determining a second target class; setting a first annotation label and a second annotation label for the first target class, wherein the first annotation label is used for indicating that the first target class is a class to be converted, and the second annotation label is used for indicating that the first target class is converted into a second target class; converting the first target class into the second target class according to the second annotation label; compiling the interface document according to the first annotation tag, the second annotation tag, the first target class and the second target class to obtain a source code file corresponding to the interface document. According to the scheme, additional auxiliary classes do not need to be written to help to generate the conversion codes, introduction dependency is reduced, loss caused by dependency introduction of all links of testing and deploying is reduced, and meanwhile the burden of the virtual machine during running can be reduced.

The invention discloses a label-consistent text backdoor attack method, which comprises the following steps of: generating a trigger word: aiming at a target data set, generating the trigger word through a primitive library; disturbing an original input sample by using a disturbance resisting method and a keyword hiding method based on a black box condition; adding the generated trigger word into the disturbed sentence to generate a poisoning sample through a trigger word replacement method based on a sememe, and training a target model by using a poisoning data set; in the reasoning stage, adding trigger words to test sentences through a sememe replacement method, and therefore the target model is decoy to predict the target category. According to the method, two methods for interfering original input are designed, and a high-quality text with the same tag as an original tag is generated, so that a target model can learn a trigger more easily; secondly, in the trigger generation and addition part, a trigger generation and addition method based on a sememe is provided; the attack success rate is high, and high-quality attack samples are generated.

The invention discloses a label generation method and system for self-adaptive container display, and relates to the field of computers. A specific embodiment of the method comprises the following steps: extracting at least one tag element from a tag text input by a user; judging whether the text length of each label element is greater than a preset label length or not; if yes, a target text with the length smaller than or equal to the preset label length is extracted from the label elements, and label content is determined according to the target text; otherwise, taking the label element as the label content; and generating a label corresponding to each label element according to the label content corresponding to each label element. According to the embodiment of the invention, the label generation method can realize that the part of the label content exceeding the length is not displayed, the personalized demand and the custom setting of a user are met, the convenience, the simplicity and the diversity of label generation are greatly improved, and the label generation system is further optimized.

The invention provides a knowledge distillation method, device and system for a pre-trained language model BERT. The knowledge distillation method comprises a distillation training stage and a fine tuning test stage. The distillation training stage comprises the following steps of obtaining a label-free text, utilizing BERT model coding to obtain a BERT sentence vector of the label-free text, acquiring a BiLSTM sentence vector of the label-free text by using BiLSTM model coding, and based on the BERT sentence vector and the BiLSTM sentence vector, training to obtain an optimal BiLSTM model. The fine tuning test stage comprises the following steps of inputting a labeled data set into an optimal BiLSTM model, and carrying out fine tuning training, and inputting the test set into the fine-tuned BiLSTM model, and calculating an output result. According to the method, a method of distillation first and fine adjustment later is adopted, and the problems that a pre-trained language model is large in BERT parameter number, complex in structure and difficult to deploy in engineering are solved.

The invention relates to an artificial intelligence technology and discloses an intelligent table query method. The method comprises the steps of receiving an original table set and a label set, splitting the original table set to obtain a standard table set, performing part-of-speech coding on the label set to obtain a word vector set, inputting the standard table set and the word vector set intoan intelligent query model for training to obtain a training value, until the training value is smaller than a preset threshold value, the intelligent query model completes training; and receiving query content of a user, extracting the query content based on a keyword extraction technology to obtain a keyword set, performing part-of-speech encoding on the keyword set to obtain a keyword vector set, inputting the keyword vector set into the trained intelligent query model to obtain a table set required by the query content, and outputting the table set. The invention further provides an intelligent table query device and a computer readable storage medium. According to the invention, an accurate and efficient table intelligent query function can be realized.

The invention discloses a label generation method and device, a storage medium and computer equipment, relates to the technical field of information, and mainly aims to improve the label generation precision. The method comprises the steps of obtaining a to-be-generated label text; inputting the to-be-generated tag text into a preset entity recognition model for entity recognition to obtain an entity contained in the to-be-generated tag text; inputting the to-be-generated tag text and the entity into a preset tag generation model for tag identification to obtain each tag contained in the to-be-generated tag text; and based on each tag, determining a target tag in the to-be-generated tag text. The method and the device are suitable for generating labels.

The invention relates to a label-based text translation system, which comprises: a marking module used for adding a start character and an end character to texts in different formats, and identifyinga position needing to be translated through the start character and the end character; the escape module marks the first row of spaces and symbols in the original text, so that the translation moduledoes not process the characters as formats; the translation module identifies an area needing to be translated in the text, compiles a translation program or calls a network interface to translate content needing to be translated in the mark; a lexicon records the proper nouns and is used for recording the one-to-one correspondence relationship between the proper nouns and the replacement character strings. According to the automatic translation system provided by the invention, the label is added into the original document, and a program automatically recognizes the area to be translated through label positioning, so that correct text content is translated in a specific format, and the text translation speed is increased.

The invention relates to a method and a system for automatically converting a standard text into a computable logic rule, which are characterized by comprising the following contents: 1) performing statement segmentation on the standard text, preprocessing a plurality of statements obtained by segmentation, and screening out convertible standard statements; 2) adopting a deep neural network model, and adding semantic tags to words or phrases in the convertible standard statements through a BIO annotation format to obtain statements with tags; (3) analyzing the statements with the labels by adopting a context-independent grammar based on identification of a P-R normal form and a recursive usage method thereof, and converting the statements with the labels into a rule check tree; and 4) converting the rule check tree into a language in a specific format. The method can be widely applied to the crossing field of the building industry and the information technology.