533 results about "IntraText" patented technology

The present disclosure provides a semantic understanding based emoji input method and device, and relates to the input method technology field. The method includes: obtaining a text content according to an input sequence; performing word segmentation on the text content, and extracting text features based on the word segmentation result; constructing an input vector using the text features, performing classification using an emotion classification model to determine an emotion label of the text content; based on a correspondence relationship between the emotion label and emojis of various themes, respectively obtaining an emoji corresponding to the emotion label from each of the various themes; sorting the obtained emojis of the various themes, and displaying the sorted emojis as candidate options in a client. The disclosed invention facilitates users to input an emoji, enhances emoji input efficiency, and provides users with rich and wide scope of emoji resources.

The present invention relates to an information reply method and an apparatus, wherein the information reply method comprises: receiving information to be replied, wherein the information to be replied comprises text content and contact information; based on the text content and the contact information, querying corresponding dialog style information from a database; performing preprocessing on the text content, wherein the preprocessing comprises segmentation processing and stop word removal processing; and based on data after the preprocessing, performing query in a database corresponding to the dialog style information so as to determine reply information. A dialog style is determined through the information to be replied and the proper reply information is queried in the database corresponding to the dialog style based on the data after the preprocessing for a user to select, which can shorten time for the user to reply information and improve user experience.

The invention discloses a text classification model training method, device and equipment and a storage medium and belongs to the field of artificial intelligence. According to the method, on one hand, the adversarial sample is introduced, and the text classification model is trained by using the text sample and the adversarial sample, so the text classification model learns the classification method for the text added with disturbance, and thereby the robustness of the text classification model is improved, and accuracy of text classification is improved; on the other hand, the text classification model can reconstruct the text features of the adversarial samples extracted during classification and restore the text features into text content, so interpretability of the adversarial training method is improved. According to the method, the model parameters are trained by combining the errors between the reconstructed text content and the text content of the text sample, so the text classification model can extract more accurate text features, namely, more accurate feature expression of the text content is obtained, and robustness and accuracy of feature extraction of the text classification model are improved.

The invention provides a cross-modal video retrieval method and system based on a multi-head self-attention mechanism and a storage medium, and the cross-modal video retrieval method comprises a videocoding step, a text coding step and a joint embedding step. Semantic information in the training multi-modal data is fully utilized for training, a multi-head self-attention mechanism is introduced,fine interaction in videos and texts is captured, key information of the multi-modal data is selectively concerned to enhance the characterization capability of the model, data semantics are better mined, and the invention has the advantages of being high in practicability and easy to popularize. Consistency of the distances of the data in the original space and the shared subspace is ensured. Theinvention has the beneficial effects that experiments prove that the similarity of the data in the original space can be effectively maintained, and the retrieval accuracy can be improved.

The invention discloses a false comment identification method based on rolling type cooperative training, and relates to the technical field of false comment identification. The method comprises the steps: obtaining a comment text, and determining the content features of the comment text according to the comment text; obtaining reviewer information corresponding to the comment text, and determining reviewer behavior characteristics according to the reviewer information; performing false comment identification according to the comment text content characteristics and the reviewer behavior characteristics; outputting false comment recognition results. According to the false comment recognition method, unlabeled samples are effectively utilized to assist model learning, multiple features suchas emotion and text representation are fused at the same time, the recognition performance of the model is improved through multi-feature fusion collaborative training, and the accuracy is improved by 3.5% compared with a traditional false comment recognition method.

The invention belongs to the technical field of text content filtration, and particularly relates to a fine-grained semantic detection method of harmful text contents in network. Aiming at an introduced harmful information scene, the method comprises the steps of: constructing a train text set in which independent sentences are used as basic units, thereby establishing a mathematic description of the scene by using a probability topic model; performing information content extraction to a Web page to be detected; performing sentence identification to the text information; calculating a condition probability of each sentence under the model based on the established probability topic model; and accomplishing the fine-grained semantic detection under the set content detection sensitivity. According to the invention, the model construction is hardly affected by the number of the topics, and probability calculation on the sentence and word level is carried out effectively, so that the method is applicable for various application circumstances requiring harmful text content detection; furthermore fine-grained detection to harmful words and sentences of the text content is supported, so that the method improves the detection rate and reduces the misinformation rate effectively, and is beneficial to improving the practicability of text content filtration.

The invention discloses a text abstract generation method and device, equipment and a medium, the method belongs to the field of computer vision, and the method comprises the following steps: obtaining text content; inputting the text content into a coding layer to obtain a hidden layer embedding vector of the text content; wherein the coding layer is obtained based on collaborative training of anextraction type abstract generation model and a generation type abstract generation model; and inputting the hidden layer embedding vector into the extraction type abstract generation model or the generation type abstract generation model for processing, and outputting the text abstract. According to the method and the device, the respective advantages of the extraction type abstract generation mode and the generation type abstract generation mode can be integrated, so that the finally output text abstract can better summarize the content characteristics of the text content.

The invention provides a method, device and equipment for text retrieval and a storage medium, and relates to the field of artificial intelligence such as big data and natural language processing. According to the specific implementation scheme, the method includes using a full-text search engine for obtaining a plurality of candidate texts meeting a retrieval formula; calculating multi-dimensional features of the candidate text according to the keyword of the retrieval formula and the text content of the candidate text; obtaining a correlation score through text similarity operation based on multi-dimensional feature fusion, sorting the multiple candidate texts according to the correlation score, wherein the correlation score is used for representing the text similarity of the candidate texts and the retrieval formula; and according to a preset rule, carrying out secondary sorting and filtering on the plurality of candidate texts to obtain a target text. According to the technical scheme, accurate query and sorting of the texts can be realized semantically, so that the accuracy and efficiency of text retrieval are improved, and the text retrieval experience of a user is improved.

The invention discloses an on-line classroom discussion short text real-time grouping method and system based on text clustering. The method comprises the steps of conducting word-splitting preprocessing and stop-word preprocessing on text data; obtaining all text item keywords, counting all the text item keywords and storing the text item keywords into a keyword table keyTable; conducting frequent item set mining on a preprocessed text set, filtering all sub-item quasi-frequent item sets and conducting coarse cluster classification in combination with a keyword table definition quasi-frequentitem set similarity calculation rule; mapping points, the closest to the cluster center, of all clusters to the text set, calculating TF-IDF values of text word sets in all the clusters and iteratingthe center of mass to be optimal according to the distance; pushing the obtained K clusters in real time in group. Through the combination of the keyword table definition quasi-frequent item set similarity calculation rule, the clustering accuracy of an on-line discussion short text is effectively improved; through a quasi-frequent item set filtering strategy, the clustering efficiency is effectively improved, and a clustering method is accelerated; the text information content discussed on an on-line classroom is automatically classified into multiple themes, and the text content is groupedaccording to the themes.

The invention provides a microblog rumor detection method, which considers an attention mechanism. The method comprises the following steps of collecting a microblog event and a corresponding commentdata set as sample data; preprocessing the sample data, and respectively extracting text contents of original microblogs and comments; pre-training the text by adopting a BERT pre-training model, andgenerating a sentence vector with a fixed length for each sentence of text; constructing a dictionary, and extracting original microblogs and a plurality of corresponding comments to form a microblogevent vector matrix; training the vector matrix by adopting a deep learning method Text CNN-Attention, and constructing a multi-level training model; and performing classification detection on the vector matrix according to the multi-level training model to obtain a rumor detection result corresponding to the social network data. Compared with a traditional rumor detection method, accuracy is improved.

The invention provides a text information recommendation method. The text information recommendation method comprises the following steps: establishing an information recommendation pool; acquiring text content of an article requiring information recommendation; segmenting the article requiring information recommendation into multiple words; predicting multi-dimensional topic distribution of the article requiring information recommendation according to multi-dimensional topic distribution of words in an LDA (latent dirichlet allocation) model base; calculating information correlation between the article requiring information recommendation and articles in the information recommendation pool; sorting related information in the information recommendation pool according to an information correlation calculation result; outputting recommended information according to a sorting result. With adoption of the method, semantic ambiguity of related information and semantic related problems during information recommendation can be solved, information heat and timeliness are taken into consideration, and the click-through rate of users is increased. The invention further provides a system for implementing the text information recommendation method.

The embodiment of the invention discloses a short message editing method, comprising obtaining text contents of a short message to be replied; identifying whether a preset key word exists in the text content exists, and querying and displaying a fast reply list associated with the preset keyword if a preset key word exists in the text content exists; and obtaining a fast reply message selected by a user in the fast reply list to generate a short message to be sent. The embodiment of the invention also discloses a terminal. According to the invention, the efficiency of short message reply can be improved.

The invention discloses a text-based event pushing method and device, electronic equipment and a storage medium, and relates to the technical field of knowledge graphs, deep learning, natural languageprocessing and cloud computing. According to the specific implementation scheme, the method comprises the steps: obtaining text content of the target event type, performing word segmentation processing on the text content; obtaining a word sequence containing a plurality of words; inputting the word vectors of the words in the word sequence into a sequence labeling model corresponding to the target event type to mark event attributes for all words in the word sequence; according to the marked event attributes of the words in the word sequence, generating the description information of the target event type, pushing the description information to the clients paying attention to the target event type so as to be displayed on the clients. The description information of the target event typecan be automatically generated and displayed on the clients, and the efficiency and comprehensiveness of obtaining the related information of the concerned event by the user are improved.

The invention relates to a method and device for a text recoverable watermark based on synonym replacement, belonging to the technical field of copyright protection of computer texts. The water recoverable watermark is a technology of embedding secret information into a text and recovering the original text while extracting corresponding watermark information. In the method and the device, words with synonyms in the text are simulated into pixel value pairs, integer reversible transform is utilized for embedding or extracting copyright information and the original text is recovered while extracting the copyright information. The method and the device have the advantages of recovering the original text while extracting the copyright information. The method and the device can be applied in the fields with high requirements on text contents, such as military affairs, law, literature and the like, not only ensure the copyright of documents, but also prevent the ambiguity interpretation oflegal users.

The invention relates to a rumor detection method based on linear and nonlinear propagation, and belongs to the technical field of natural language understanding. According to the method, unified modeling representation is carried out on rumor nodes by utilizing text content and time information, and rumor detection is automatically carried out in a mode of combining linear and nonlinear propagation characteristics. Firstly, text information and time information contained in rumor nodes are used for carrying out joint representation on mixed features of the rumor nodes; then, node informationis aggregated along the linear time sequence and the nonlinear diffusion structure, expression of a source node is enhanced, and final propagation representation is formed. And finally, authenticity label prediction is carried out by using propagation representation. According to the method, node characteristics of rumors are extracted from two different angles, tree perception representation is obtained from a nonlinear diffusion mode, characteristics of propagation sequences are captured from linear time sequence interaction, and authenticity of the rumors can be accurately predicted.

The invention relates to a document translation method and device, a storage medium and electronic equipment, and the method comprises the steps: displaying an original text display region, a translation region and an editing region, wherein text content in a to-be-translated document is displayed in the original text display region, and a reference translation for the text content is displayed inthe translation region; providing translation suggestions from the reference translation along with the input of the user in the editing area, wherein under the condition that confirmation operationof a user for the translation suggestion is detected, the translation suggestion serves as a translation result to be displayed in the editing area; and under the condition that a non-confirmation operation suggested by the user for the translation is detected, receiving the translation which is input by the user and is different from the suggested translation, and updating the reference translation in the translation area according to the translation input by the user. According to the method, the translation efficiency and the translation quality can be improved.

The invention discloses a numerical value and text mixed inverted index algorithm based on a multilayer-optimization balanced tree. The traditional text inverted index only supports indexing on free texts, wherein the content of a numerical value is generally converted into a text, and then an inverted index is established according to the content of the text. According to the invention, the traditional text inverted index algorithm is optimized and expanded so as to be capable of supporting the indexing on numerical value and text mixed data, and a reasonable balance is realized in the aspects of query performance, index space and establishing performance. The index algorithm disclosed by the invention is suitable for being used in a mixed type data management engine and can be used for increasing the composite query performance of the numerical value and text mixed type data.

The invention discloses an interactive microblog text emotion mining method based on emotion migration perception in a social network, which comprises the following steps of: firstly, carrying out emotion polarity labeling on a microblog text based on single microblog semantics and text contents of interaction history of the single microblog semantics; secondly, adopting a pre-training language model BERT in the field of natural language processing, and extracting statement-level microblog emotion semantic features; secondly, extracting a context-level emotion semantic feature in an interactive social network context by using a long short-term memory (LSTM) network; then, introducing a learning normal form of multi-task learning, establishing an emotion migration perception auxiliary task, designing an emotion association relationship enhanced Attention mechanism by utilizing emotion migration characteristics, extracting emotion influence factors related to the current microblog from interaction history, then fusing the emotion influence factors with emotion semantic characteristics, carrying out emotion polarity classification, and constructing a microblog text emotion recognition model. According to the method, the accuracy of interactive microblog text emotion mining and the generalization ability of the model are greatly improved.

The invention discloses a knowledge reasoning-based big data service label extension method and system, and the method comprises the steps: S1, selecting a to-be-extended label and a label subordinatekeyword set K from a preset service label tree, searching each keyword, and obtaining a result-related URL set C; S2, for each keyword, traversing a corresponding URL in the set C, obtaining text content from each webpage, performing semantic analysis on the text, extracting a semantic object and a semantic relationship associated with the keyword, and constructing a semantic relationship knowledge graph; S3, checking and screening entities in the knowledge graph by utilizing the knowledge graph and combining the ontology rule model and taking the to-be-expanded label as a basis, and adding entity attribute contents passing the check into a result set in a keyword form; S4, judging whether the scale of the result set reaches a preset extension scale or not: if so, ending, otherwise, taking the keywords in the result set as a set K, and returning to the step S1.