WO2023045184A1 - Procédé et appareil de reconnaissance de catégorie de texte, dispositif informatique et support - Google Patents

Procédé et appareil de reconnaissance de catégorie de texte, dispositif informatique et support Download PDF

Info

Publication number
WO2023045184A1
WO2023045184A1 PCT/CN2022/070967 CN2022070967W WO2023045184A1 WO 2023045184 A1 WO2023045184 A1 WO 2023045184A1 CN 2022070967 W CN2022070967 W CN 2022070967W WO 2023045184 A1 WO2023045184 A1 WO 2023045184A1
Authority
WO
WIPO (PCT)
Prior art keywords
text
category
spliced
value
recognition model
Prior art date
Application number
PCT/CN2022/070967
Other languages
English (en)
Chinese (zh)
Inventor
黄振宇
王磊
吴文哲
王媛
王晶璇
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2023045184A1 publication Critical patent/WO2023045184A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G06F16/353—Clustering; Classification into predefined classes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/205—Parsing
    • G06F40/216—Parsing using statistical methods

Definitions

  • the present application relates to the technical field of artificial intelligence, in particular to a text category recognition method, device, computer equipment and media.
  • the machine's accurate recognition of the type of text is inseparable from deep learning of natural language, and deep learning mostly uses deep neural network models, that is, deep neural network models are used to train natural language texts.
  • deep neural network models are used to train natural language texts.
  • neural networks under multi-classification problems, Its recognition accuracy is often affected by more categories of text data, that is, when there are more categories of text data, the recognition accuracy of the neural network will decrease.
  • a text category recognition method comprising: obtaining a target text to be recognized; splicing the target text and each text in a standard set to generate a first spliced text set; inputting each text in the first spliced text set into the pre-trained text one by one In the category recognition model, the predicted value of each text in the first spliced text set is output; wherein, the pre-trained text category recognition model is generated based on spliced text training in the second spliced text set, and the second spliced text set is the training set Each text in the test set is spliced with each text in the standard set; the category of the target text is determined based on the predicted value of each text in the first spliced text set.
  • a text category recognition device includes: a text acquisition module, used to obtain the target text to be recognized; a spliced text set generation module, used to splice the target text and each text in the standard set to generate a first spliced text set;
  • the input module is used to input each text in the first spliced text set one by one into the pre-trained text category recognition model, and output the predicted value of each text in the first spliced text set; wherein, the pre-trained text category recognition model is based on the first
  • the spliced text training in the second spliced text set is generated, and the second spliced text set is generated by splicing each text in the training set, the test set, and each text in the standard set;
  • the category determination module is used for each text in the first spliced text set.
  • the predicted value of a piece of text determines the category of the target text.
  • a computer device includes a memory and a processor, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processor, the processor is made to execute the steps of the above text category recognition method.
  • a medium storing computer-readable instructions.
  • the computer-readable instructions are executed by one or more processors, one or more processors are made to execute the steps of the above-mentioned text category recognition method.
  • the text category recognition device first obtains the target text to be recognized, then splices the target text and each text in the standard set to generate the first spliced text set, and then combines each text in the first spliced text set
  • the texts are input into the pre-trained text category recognition model one by one, and the predicted value of each text in the first spliced text set is output; the pre-trained text category recognition model is generated based on spliced text training in the second spliced text set, and the second spliced
  • the text set is generated by concatenating each text in the training set, the test set, and each text in the standard set, and finally the category of the target text is determined based on the predicted value of each text in the first spliced text set. Since the application establishes a correlation after splicing the training set, the test set and the standard set, after using the correlated data to train the model, the recognition result of the model
  • FIG. 1 is an implementation environment diagram of a text category recognition method provided in one embodiment of the present application
  • FIG. 2 is a schematic diagram of the internal structure of a computer device in an embodiment of the present application.
  • FIG. 3 is a method schematic diagram of a text category recognition method provided in an embodiment of the present application.
  • FIG. 4 is a schematic diagram of a text category recognition model provided in an embodiment of the present application.
  • FIG. 5 is a schematic diagram of the idea of text category recognition provided in one embodiment of the present application.
  • Fig. 6 is a device schematic diagram of a text category recognition device provided by an embodiment of the present application.
  • FIG. 1 is an implementation environment diagram of a text category recognition method provided in an embodiment. As shown in FIG. 1 , the implementation environment includes a server 110 and a client 120 .
  • the server 110 can be a server, which can be an independent server, or can provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services , Content Delivery Network (Content Delivery Network, CDN), and cloud servers for basic cloud computing services such as big data and artificial intelligence platforms, such as server equipment for deploying topic extraction models.
  • cloud services cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services , Content Delivery Network (Content Delivery Network, CDN), and cloud servers for basic cloud computing services such as big data and artificial intelligence platforms, such as server equipment for deploying topic extraction models.
  • CDN Content Delivery Network
  • the server 110 obtains the target text to be recognized from the client 120, and the server 110 inputs each text in the first mosaic text set into the pre-trained text category recognition model one by one, and outputs the first concatenation
  • the predicted value of each text in the text set the server 110 queries the question set associated with the topic result of the description information, and sends the question set to the client 120, so that the client 120 can display it on the display interface, and the server 110 is based on
  • the predicted value of each piece of text in the first mosaic text set determines the category of the target text, and sends the category to the client 120 for display.
  • the client 120 may be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc., but is not limited thereto.
  • the server 110 and the client 120 can be connected through Bluetooth, USB (Universal Serial Bus, Universal Serial Bus) or other communication connection methods, which are not limited in this application.
  • Fig. 2 is a schematic diagram of the internal structure of a computer device in an embodiment.
  • the computer device includes a processor, a medium, a memory, and a network interface connected through a system bus.
  • the medium of the computer device stores an operating system, a database, and computer-readable instructions
  • the database can store control information sequences.
  • the processor can implement a text category recognition method .
  • the processor of the computer equipment is used to provide computing and control capabilities to support the operation of the entire equipment.
  • Computer-readable instructions may be stored in the memory of the computer device, and when the computer-readable instructions are executed by the processor, the processor may execute a text category recognition method.
  • the network interface of the computer device is used for connecting and communicating with the terminal.
  • the structure shown in Figure 2 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation to the computer equipment on which the solution of the application is applied.
  • the specific computer equipment can be More or fewer components than shown in the figures may be included, or some components may be combined, or have a different arrangement of components.
  • the medium is a readable storage medium.
  • the text category recognition method provided by the embodiment of the present application will be described in detail below with reference to accompanying drawings 3 to 5 .
  • the method can be implemented relying on a computer program, and can run on a text category recognition device based on the von Neumann system.
  • the computer program can be integrated in the application, or run as an independent utility application.
  • AI artificial intelligence
  • digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
  • Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation/interaction systems, and mechatronics.
  • Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning/deep learning.
  • FIG. 3 provides a schematic flowchart of a text category recognition method according to the embodiment of the present application.
  • the method of the embodiment of the present application may include the following steps:
  • the target text is the expression of language
  • language is a set of communication instructions with common processing rules for expression.
  • the instructions will be transmitted in the form of vision, sound or touch.
  • This instruction specifically refers to the natural language used by human communication , such as Chinese and English.
  • Text refers to the manifestation of written language, usually a sentence or a combination of multiple sentences with complete and systematic meaning.
  • a text can be a sentence, a paragraph or a chapter.
  • the method of obtaining the target text to be recognized at least includes obtaining from the test set.
  • a text is a word composed of several characters or a sentence composed of several words, or a paragraph composed of several sentences. Users can describe their thoughts through text, and using text to describe can make Complex thoughts become instructions that are easy for others to understand. For text, different expressions can be used to make complex ideas easy to understand, making communication easier to understand.
  • One or more pieces of natural language contained in the target text can be referred to as sentences for short, or simply as sentences, or the text can be split into sentences according to the punctuation in the text, that is, it will end with a period, question mark, exclamation point, comma, etc. content as a sentence.
  • the target text may be text input by the user to the user terminal.
  • the text input to the terminal can be the language text obtained from the Internet, that is, in the actual application scenario, or the language text obtained from the test set, that is, in the model training scenario. There are many ways to generate the target text. There is no limit here.
  • the target text to be recognized before obtaining the target text to be recognized, it also includes collecting multiple description texts from the text library; receiving a labeling instruction for each description text in the multiple description texts, and marking each description text based on Afterwards, multiple annotation texts are generated; the multiple annotation texts are divided into training set, test set, and standard set according to preset percentages.
  • the smallest standard text set also needs to be 20 to 30 sentences, or use methods such as semantic representation to measure the diversity of texts, select a small amount of texts with rich diversity, and ensure that the texts in this set are right for each other.
  • This class has a certain coverage.
  • a target text to be recognized may be determined from the test set.
  • online data information is obtained, and the data information is determined as the target text to be recognized.
  • each piece of target text to be recognized and each standard text set are spliced into a spliced text, assuming that the target text to be recognized is A ⁇ , the standard text set is text D label 1, text E label 2, the spliced text set is: text A ⁇ text D 0; text A ⁇ text E 0.
  • multiple texts to be recognized are obtained by splicing the target text to be recognized with the text in the standard set, which can greatly increase the prediction frequency of the text to be predicted and make the subsequent calculated confidence more accurate.
  • the pre-trained text category recognition model is generated based on spliced text training in the second spliced text set, and the second spliced text set is generated by splicing each text in the training set, the test set and each text in the standard set;
  • the pre-trained text category recognition model can be generated according to the following steps. First, each text in the training set, test set and each text in the standard set are spliced to generate a second spliced text set, and then the text category is created. Identify the model, and then input each piece of spliced text in the second spliced text set into the text category recognition model, output the loss value of the model, and finally generate a pre-trained text category recognition model when the loss value is less than the preset loss threshold.
  • the text category recognition model is a twin model such as Sbert (capable of receiving two input texts at the same time), which is created by using a language expression model (such as word2vec, BERT, GPT, etc.).
  • the loss value is greater than or equal to the preset loss threshold, the loss value is backpropagated, and the step of inputting each piece of spliced text in the second spliced text set into the text category recognition model is continued.
  • the original texts in the training set and test set are: text A label 1; text B label 2; text C label 3.
  • the current standard text in the standard set is: Text D Label 1 Text E Label 2
  • the spliced second spliced text set is:
  • the above-mentioned second spliced text set is input into Sbert's twin model one by one for training, and a pre-trained text category recognition model is generated after the training is completed.
  • the basic network structure of Sbert's twin model is shown in Figure 4.
  • the twin model contains two BERT language models, and each BERT language model corresponds to a pooling layer pooling.
  • the pooling layer pooling can reduce the output vector of the language model. sampling.
  • a text of A is sentenceA
  • a text of B is sentenceB. Both sentenceA and sentenceB allow input to the BERT language model.
  • the model outputs vector A and vector B.
  • Vector A and vector B are respectively input into the pooling layer pooling, and the output vector u and Vector v, and calculate the cosine similarity cosine-sim(u, v) according to the vector u and vector v, and finally make a difference between the cosine similarity and the label value corresponding to the A text and the B text, generate the loss value loss of the model, and reverse Propagate loss until the model converges.
  • each text in the first stitched text set is input into the pre-trained text category recognition model one by one, and the predicted value of each text in the first stitched text set is output.
  • the user pre-sets 5 categories, 20 standard texts are prepared for each category, and there are 100 standard texts in total; then a sentence to be recognized and 100 standard texts are spliced and input into the model, and the model is 100 predicted values are output directly.
  • the threshold value of each category in the preset multiple categories is first obtained, and then according to each The predicted value of the text and the threshold of each category count the counting results of each category, generate a counting result sequence, and then obtain the maximum counting result from the counting result sequence, and then use the standard text corresponding to the maximum counting result and the category of the maximum counting result.
  • the total quantity is used as a quotient to generate the confidence degree of the target text, and finally, according to the confidence degree being greater than a preset value, the category corresponding to the confidence degree is determined as the category corresponding to the target text.
  • the predicted values of each text in the first spliced text set are: a, b, c; the thresholds of the types in preset 3 are A, B, and C respectively; compare a, b, and c with A respectively, for example, a >A will count 1, otherwise it will not count, and the counting result of type A can be obtained. Comparing a, b, c with B respectively, the counting result of type B can be obtained, and comparing a, b, c with C respectively By comparison, the counting result of the C type can be obtained. For example, if the counting results of the last three categories are [10,19,5], then the confidence of the second category is considered to be the highest, and if the 95% confidence is obtained after calculating 19, the target text to be recognized is considered to belong to this category.
  • the traditional classification method looks for the edges of the three ellipses A, B, and C in Figure 5, while this application looks for the boundaries of A, B, and C (black thick line).
  • the text category recognition device first obtains the target text to be recognized, and then splices the target text and each text in the standard set to generate the first spliced text set, and then inputs each text in the first spliced text set one by one in advance
  • the predicted value of each text in the first spliced text set is output;
  • the pre-trained text category recognition model is generated based on the spliced text training in the second spliced text set, and the second spliced text set is the training
  • Each text in the test set and each text in the standard set are spliced and generated, and finally the category of the target text is determined based on the predicted value of each text in the first spliced text set. Since the application establishes a correlation after splicing the training set, the test set and the standard set, after using the correlated data to train the model, the recognition result of the model is more accurate, thereby improving the accuracy of text category recognition.
  • FIG. 6 shows a schematic structural diagram of an apparatus for identifying a text category provided by an exemplary embodiment of the present application, which is applied to a server.
  • the text category recognition device can be implemented as all or a part of the device through software, hardware or a combination of the two.
  • the device 1 includes a text acquisition module 10 , a mosaic text set generation module 20 , a text input module 30 , and a category determination module 40 .
  • Text obtaining module 10 is used for obtaining the target text to be identified
  • the mosaic text set generation module 20 is used for splicing each text in the target text and the standard set to generate the first mosaic text set;
  • Text input module 30 is used for inputting each piece of text in the first mosaic text set one by one in the pre-trained text category recognition model, outputs the predictive value of each text in the first mosaic text set;
  • the pre-trained text category recognition model is generated based on spliced text training in the second spliced text set, and the second spliced text set is generated by splicing each text in the training set, the test set and each text in the standard set;
  • a category determining module 40 configured to determine the category of the target text based on the predicted value of each text in the first mosaic text set.
  • the text category recognition device first obtains the target text to be recognized, and then splices the target text and each text in the standard set to generate the first spliced text set, and then inputs each text in the first spliced text set one by one in advance
  • the predicted value of each text in the first spliced text set is output;
  • the pre-trained text category recognition model is generated based on the spliced text training in the second spliced text set, and the second spliced text set is the training
  • Each text in the test set and each text in the standard set are spliced and generated, and finally the category of the target text is determined based on the predicted value of each text in the first spliced text set. Since the application establishes a correlation after splicing the training set, the test set and the standard set, after using the correlated data to train the model, the recognition result of the model is more accurate, thereby improving the accuracy of text category recognition.
  • a computer device in one embodiment, includes a memory, a processor, and a computer program stored on the memory and operable on the processor.
  • the processor executes the computer program, the following steps are implemented: acquiring the target to be identified text; the target text and each text in the standard set are spliced to generate the first spliced text set; each text in the first spliced text set is input into the pre-trained text category recognition model one by one, and the output of each text in the first spliced text set is Predicted value; wherein, the pre-trained text category recognition model is generated based on the spliced text training in the second spliced text set, and the second spliced text set is generated by splicing each text in the training set, the test set, and the standard set The category of the target text is determined based on the predicted value of each text in the first assembled text set.
  • the processor before the processor executes obtaining the target text to be recognized, it also performs the following operations: collect multiple description texts from the text library; receive a labeling instruction for each description text in the multiple description texts, and based on Multiple annotation texts are generated after each description text is annotated; multiple annotation texts are divided into training set, test set, and standard set according to preset percentages.
  • the processor determines the category of the target text based on the predicted value of each text in the first mosaic text set
  • the following operations are specifically performed: acquiring a threshold value for each of the preset multiple categories; The predicted value of each text in the text set and the threshold value of each category count the counting results of each category to generate a counting result sequence; obtain the maximum counting result from the counting result sequence; the standard corresponding to the maximum counting result and the category of the maximum counting result
  • the total number of texts is used as a quotient to generate the confidence level of the target text; according to the confidence level being greater than a preset value, the category corresponding to the confidence level is determined as the category corresponding to the target text.
  • the processor executes counting the counting results of each category according to the predicted value of each text in the first mosaic text set and the threshold value of each category, and when generating the counting result sequence, specifically performs the following operations: judge the first Whether the predicted value of each text in the spliced text set is greater than the threshold of each category; if so, automatically add one to the initial value of each category; if not, keep the initial value of each category unchanged; where the initial value is 0 ; After the judgment of the predicted value of each text in the first assembled text set is completed, the final initial value of each category is determined as the counting result of each category.
  • the processor when it generates the pre-trained text category recognition model, it specifically performs the following operations: splicing each text in the training set and the test set with each text in the standard set to generate a second spliced text set; creating a text A category recognition model; input each piece of spliced text in the second spliced text set into the text category recognition model, and output a loss value of the model; when the loss value is less than a preset loss threshold, generate a pre-trained text category recognition model.
  • the processor when the processor executes inputting each spliced text in the second spliced text set into the text category recognition model, and outputs the loss value of the model, it specifically performs the following operations: input each spliced text in the second spliced text set into the text In the category recognition model, the first semantic vector and the second semantic vector are output; the category similarity value of each piece of spliced text in the second spliced text set is calculated according to the first semantic vector and the second semantic vector; The label value of the text; the difference between the category similarity value of each spliced text and the corresponding label value is generated to generate the loss value of the model; the loss value of the output model.
  • the processor when the processor generates a pre-trained text category recognition model when the loss value is less than the preset loss threshold, the processor specifically performs the following operations: when the loss value is greater than or equal to the preset loss threshold, reverse the loss value Propagate; continue to execute the step of inputting each stitched text in the second stitched text set into the text category recognition model.
  • the text category recognition device first obtains the target text to be recognized, and then splices the target text and each text in the standard set to generate the first spliced text set, and then inputs each text in the first spliced text set one by one in advance
  • the predicted value of each text in the first spliced text set is output;
  • the pre-trained text category recognition model is generated based on the spliced text training in the second spliced text set, and the second spliced text set is the training
  • Each text in the test set and each text in the standard set are spliced and generated, and finally the category of the target text is determined based on the predicted value of each text in the first spliced text set. Since the application establishes a correlation after splicing the training set, the test set and the standard set, after using the correlated data to train the model, the recognition result of the model is more accurate, thereby improving the accuracy of text category recognition.
  • a medium storing computer-readable instructions is provided.
  • the medium of the computer-readable instructions may be non-volatile or volatile.
  • the computer-readable instructions are stored by one or more When the processor is executed, one or more processors are made to perform the following steps: obtaining the target text to be recognized; splicing the target text and each text in the standard set to generate a first spliced text set; Input the pre-trained text category recognition model one by one, and output the predicted value of each text in the first spliced text set; wherein, the pre-trained text category recognition model is generated based on spliced text training in the second spliced text set, and the second spliced
  • the text set is generated by concatenating each text in the training set, the test set, and each text in the standard set; the category of the target text is determined based on the predicted value of each text in the first spliced text set.
  • the processor before the processor executes obtaining the target text to be recognized, it also performs the following operations: collect multiple description texts from the text library; receive a labeling instruction for each description text in the multiple description texts, and based on Multiple annotation texts are generated after each description text is annotated; multiple annotation texts are divided into training set, test set, and standard set according to preset percentages.
  • the processor determines the category of the target text based on the predicted value of each text in the first mosaic text set
  • the following operations are specifically performed: acquiring a threshold value for each of the preset multiple categories; The predicted value of each text in the text set and the threshold value of each category count the counting results of each category to generate a counting result sequence; obtain the maximum counting result from the counting result sequence; the standard corresponding to the maximum counting result and the category of the maximum counting result
  • the total number of texts is used as a quotient to generate the confidence level of the target text; according to the confidence level being greater than a preset value, the category corresponding to the confidence level is determined as the category corresponding to the target text.
  • the processor executes counting the counting results of each category according to the predicted value of each text in the first mosaic text set and the threshold value of each category, and when generating the counting result sequence, specifically performs the following operations: judge the first Whether the predicted value of each text in the spliced text set is greater than the threshold of each category; if so, automatically add one to the initial value of each category; if not, keep the initial value of each category unchanged; where the initial value is 0 ; After the judgment of the predicted value of each text in the first assembled text set is completed, the final initial value of each category is determined as the counting result of each category.
  • the processor when it generates the pre-trained text category recognition model, it specifically performs the following operations: splicing each text in the training set and the test set with each text in the standard set to generate a second spliced text set; creating a text A category recognition model; input each piece of spliced text in the second spliced text set into the text category recognition model, and output a loss value of the model; when the loss value is less than a preset loss threshold, generate a pre-trained text category recognition model.
  • the processor when the processor executes inputting each spliced text in the second spliced text set into the text category recognition model, and outputs the loss value of the model, it specifically performs the following operations: input each spliced text in the second spliced text set into the text In the category recognition model, the first semantic vector and the second semantic vector are output; the category similarity value of each piece of spliced text in the second spliced text set is calculated according to the first semantic vector and the second semantic vector; The label value of the text; the difference between the category similarity value of each spliced text and the corresponding label value is generated to generate the loss value of the model; the loss value of the output model.
  • the processor when the processor generates a pre-trained text category recognition model when the loss value is less than the preset loss threshold, the processor specifically performs the following operations: when the loss value is greater than or equal to the preset loss threshold, reverse the loss value Propagate; continue to execute the step of inputting each stitched text in the second stitched text set into the text category recognition model.
  • the text category recognition device first obtains the target text to be recognized, and then splices the target text and each text in the standard set to generate the first spliced text set, and then inputs each text in the first spliced text set one by one in advance
  • the predicted value of each text in the first spliced text set is output;
  • the pre-trained text category recognition model is generated based on the spliced text training in the second spliced text set, and the second spliced text set is the training
  • Each text in the test set and each text in the standard set are spliced and generated, and finally the category of the target text is determined based on the predicted value of each text in the first spliced text set. Since the application establishes a correlation after splicing the training set, the test set and the standard set, after using the correlated data to train the model, the recognition result of the model is more accurate, thereby improving the accuracy of text category recognition.
  • the aforementioned medium may be a nonvolatile medium such as a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

La présente demande concerne le domaine de l'intelligence artificielle, et divulgue un procédé et un appareil de reconnaissance de catégorie de texte, un dispositif informatique et un support. Le procédé comprend les étapes consistant à : acquérir un texte cible à reconnaître ; épisser le texte cible et chaque texte dans un ensemble standard pour générer un premier ensemble de textes épissés ; entrer chaque texte dans le premier ensemble de textes épissés dans un modèle de reconnaissance de catégorie de texte préentraîné un par un, et délivrer en sortie une valeur prédite de chaque texte dans le premier ensemble de textes épissés, le modèle de reconnaissance de catégorie de texte préentraîné étant généré sur la base de l'entraînement de textes épissés dans un second ensemble de textes épissés, et le second ensemble de textes épissés étant généré par épissage de chaque texte dans un ensemble d'entraînement et un ensemble de test avec chaque texte dans l'ensemble standard ; et déterminer une catégorie du texte cible sur la base de la valeur prédite de chaque texte dans le premier ensemble de textes épissés. Dans la présente demande, un ensemble d'entraînement, un ensemble de test et un ensemble standard sont épissés pour établir une corrélation, et un modèle est entraîné à l'aide de données ayant cette corrélation, amenant le résultat de reconnaissance du modèle à être plus précis.
PCT/CN2022/070967 2021-09-26 2022-01-10 Procédé et appareil de reconnaissance de catégorie de texte, dispositif informatique et support WO2023045184A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111131337.2A CN113836303A (zh) 2021-09-26 2021-09-26 一种文本类别识别方法、装置、计算机设备及介质
CN202111131337.2 2021-09-26

Publications (1)

Publication Number Publication Date
WO2023045184A1 true WO2023045184A1 (fr) 2023-03-30

Family

ID=78970196

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/070967 WO2023045184A1 (fr) 2021-09-26 2022-01-10 Procédé et appareil de reconnaissance de catégorie de texte, dispositif informatique et support

Country Status (2)

Country Link
CN (1) CN113836303A (fr)
WO (1) WO2023045184A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116167336A (zh) * 2023-04-22 2023-05-26 拓普思传感器(太仓)有限公司 基于云计算的传感器数据加工方法、云服务器及介质
CN117033612A (zh) * 2023-08-18 2023-11-10 中航信移动科技有限公司 一种文本匹配方法、电子设备及存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836303A (zh) * 2021-09-26 2021-12-24 平安科技（深圳）有限公司 一种文本类别识别方法、装置、计算机设备及介质
CN115035510B (zh) * 2022-08-11 2022-11-15 深圳前海环融联易信息科技服务有限公司 文本识别模型训练方法、文本识别方法、设备及介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120078631A1 (en) * 2010-09-26 2012-03-29 Alibaba Group Holding Limited Recognition of target words using designated characteristic values
CN108875743A (zh) * 2017-05-15 2018-11-23 阿里巴巴集团控股有限公司 一种文本识别方法及装置
CN112182217A (zh) * 2020-09-28 2021-01-05 云知声智能科技股份有限公司 多标签文本类别的识别方法、装置、设备和存储介质
CN113360654A (zh) * 2021-06-23 2021-09-07 深圳平安综合金融服务有限公司 文本分类方法、装置、电子设备及可读存储介质
CN113360660A (zh) * 2021-07-27 2021-09-07 北京有竹居网络技术有限公司 文本类别识别方法、装置、电子设备和存储介质
CN113836303A (zh) * 2021-09-26 2021-12-24 平安科技（深圳）有限公司 一种文本类别识别方法、装置、计算机设备及介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111382271B (zh) * 2020-03-09 2023-05-23 支付宝(杭州)信息技术有限公司 文本分类模型的训练方法及装置、文本分类方法及装置
CN112464662B (zh) * 2020-12-02 2022-09-30 深圳平安医疗健康科技服务有限公司 医学短语匹配方法、装置、设备及存储介质
CN113157927B (zh) * 2021-05-27 2023-10-31 中国平安人寿保险股份有限公司 文本分类方法、装置、电子设备及可读存储介质
CN113326379B (zh) * 2021-06-30 2024-07-19 中国平安人寿保险股份有限公司 文本分类预测方法、装置、设备及存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120078631A1 (en) * 2010-09-26 2012-03-29 Alibaba Group Holding Limited Recognition of target words using designated characteristic values
CN108875743A (zh) * 2017-05-15 2018-11-23 阿里巴巴集团控股有限公司 一种文本识别方法及装置
CN112182217A (zh) * 2020-09-28 2021-01-05 云知声智能科技股份有限公司 多标签文本类别的识别方法、装置、设备和存储介质
CN113360654A (zh) * 2021-06-23 2021-09-07 深圳平安综合金融服务有限公司 文本分类方法、装置、电子设备及可读存储介质
CN113360660A (zh) * 2021-07-27 2021-09-07 北京有竹居网络技术有限公司 文本类别识别方法、装置、电子设备和存储介质
CN113836303A (zh) * 2021-09-26 2021-12-24 平安科技（深圳）有限公司 一种文本类别识别方法、装置、计算机设备及介质

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116167336A (zh) * 2023-04-22 2023-05-26 拓普思传感器(太仓)有限公司 基于云计算的传感器数据加工方法、云服务器及介质
CN117033612A (zh) * 2023-08-18 2023-11-10 中航信移动科技有限公司 一种文本匹配方法、电子设备及存储介质
CN117033612B (zh) * 2023-08-18 2024-06-04 中航信移动科技有限公司 一种文本匹配方法、电子设备及存储介质

Also Published As

Publication number Publication date
CN113836303A (zh) 2021-12-24

Similar Documents

Publication Publication Date Title
WO2023045184A1 (fr) Procédé et appareil de reconnaissance de catégorie de texte, dispositif informatique et support
US20230100376A1 (en) Text sentence processing method and apparatus, computer device, and storage medium
US11948058B2 (en) Utilizing recurrent neural networks to recognize and extract open intent from text inputs
WO2022142613A1 (fr) Procédé et appareil d'expansion de corpus de formation et procédé et appareil de formation de modèle de reconnaissance d'intention
US10747962B1 (en) Artificial intelligence system using phrase tables to evaluate and improve neural network based machine translation
JP7430820B2 (ja) ソートモデルのトレーニング方法及び装置、電子機器、コンピュータ可読記憶媒体、コンピュータプログラム
US11194963B1 (en) Auditing citations in a textual document
CN111832312B (zh) 文本处理方法、装置、设备和存储介质
CN109857846B (zh) 用户问句与知识点的匹配方法和装置
US20230259707A1 (en) Systems and methods for natural language processing (nlp) model robustness determination
CN107112009B (zh) 用于生成混淆网络的方法、系统和计算机可读存储设备
WO2022174496A1 (fr) Procédé et appareil d'annotation de données basés sur un modèle génératif, dispositif et support de stockage
WO2020149959A1 (fr) Conversion d'une interrogation en langage naturel
US20220414463A1 (en) Automated troubleshooter
CN111368066B (zh) 获取对话摘要的方法、装置和计算机可读存储介质
US11594054B2 (en) Document lineage management system
CN111967253A (zh) 一种实体消歧方法、装置、计算机设备及存储介质
WO2022262080A1 (fr) Procédé de traitement de relation de dialogue, ordinateur et support de stockage lisible
JP2022088540A (ja) ユーザ興味画像の生成方法、装置、電子機器及び記憶媒体
CN112395880B (zh) 结构化三元组的纠错方法、装置、计算机设备及存储介质
US11347928B2 (en) Detecting and processing sections spanning processed document partitions
CN115510203B (zh) 问题答案确定方法、装置、设备、存储介质及程序产品
CN111597336A (zh) 训练文本的处理方法、装置、电子设备及可读存储介质
CN110895924B (zh) 一种文档内容朗读方法、装置、电子设备及可读存储介质
CN116186219A (zh) 一种人机对话交互方法方法、系统及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22871252

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22871252

Country of ref document: EP

Kind code of ref document: A1