CN114328815A - Text mapping model processing method and device, computer equipment and storage medium - Google Patents

Abstract

The embodiment of the application discloses a processing method and device of a text mapping model, computer equipment and a storage medium, and belongs to the technical field of computers. The method comprises the following steps: acquiring sample text information and first label information, and mapping the sample text information based on a text mapping model to obtain predicted text information; and determining second label information based on a first similarity between the predicted text information and the sample text information and a second similarity between the first label information and the sample text information, and training the text mapping model based on the second label information, the predicted text information and the first similarity corresponding to the predicted text information. In the embodiment of the application, the text information most similar to the sample text information is used as the second label information, and the text mapping model is trained based on the predicted text information, the second label text information and the first similarity between the predicted text information and the sample text information, so that the mapping effect of the text mapping model is improved.

Description

Processing method, device, computer equipment and storage medium of text mapping model

technical field

The embodiments of the present application relate to the field of computer technologies, and in particular, to a method, apparatus, computer device, and storage medium for processing a text mapping model.

Background technique

With the development of computer technology, artificial intelligence models are widely used in various scenarios. Among them, in the natural language processing scenario, the text information is usually mapped based on the text mapping model to obtain text information similar to the text information, so as to realize the expansion of the text information, but the mapping effect of the current text mapping model is still poor. .

SUMMARY OF THE INVENTION

Embodiments of the present application provide a method, apparatus, computer device and storage medium for processing a text mapping model, which can improve the mapping effect of the text mapping model. The technical solution is as follows:

In one aspect, a method for processing a text mapping model is provided, the method comprising:

Obtain sample text information and first label information, where the first label information is text information whose similarity with the sample text information is not less than a similarity threshold;

Based on the text mapping model, mapping the sample text information to obtain predicted text information;

Based on a first similarity between the predicted text information and the sample text information and a second similarity between the first label information and the sample text information, second label information is determined, and the second label information is determined. The label information is the text information with a larger similarity between the predicted text information and the first label information and the sample text information;

Based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information, the text mapping model is trained, and the text mapping model is used to map any text information. Similar text information.

In another aspect, there is provided an apparatus for processing a text mapping model, the apparatus comprising:

an acquisition module, configured to acquire sample text information and first label information, where the first label information is text information whose similarity with the sample text information is not less than a similarity threshold;

a mapping module, configured to map the sample text information based on the text mapping model to obtain predicted text information;

A determination module, configured to determine second label information based on a first similarity between the predicted text information and the sample text information and a second similarity between the first label information and the sample text information , the second label information is the text information with a larger similarity between the predicted text information and the first label information and the sample text information;

A training module, configured to train the text mapping model based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information, and the text mapping model is used for mapping Similar text messages out of any text message.

In a possible implementation, the apparatus further includes:

The obtaining module is further configured to obtain a third degree of similarity and a fourth degree of similarity between the predicted text information and the sample text information, where the third degree of similarity indicates the predicted text information and the sample text Differences of words included in the information, and the fourth similarity indicates the semantic similarity between the predicted text information and the sample text information;

The fusion module is used for weighted fusion of the third similarity and the fourth similarity to obtain the first similarity.

In another possible implementation manner, the obtaining module is configured to divide the predicted text information based on the number of at least one character, respectively, to obtain at least one first word set, which belongs to the same first word set. The words contain the same number of characters; based on at least one of the number of characters, the sample text information is divided respectively to obtain at least one second set of words, and the words belonging to the same second set of words contain the same number of characters; determining a first number and a second number, the first number indicating the sum of the number of different words in the first set of words and the second set of words corresponding to each of the number of characters, the second number indicating at least one The total number of words in the first word set and at least one of the second word sets; determining the ratio of the first number to the second number as the difference between the predicted text information and the sample text information; the third similarity between.

In another possible implementation manner, the obtaining module is configured to perform semantic extraction on the predicted text information and the sample text information, respectively, to obtain a first semantic feature of the predicted text information and the sample text information The second semantic feature of ; determining the similarity between the first semantic feature and the second semantic feature as the fourth similarity.

In another possible implementation, the acquisition module is configured to splicing the predicted text information and the sample text information to obtain spliced text information; perform semantic extraction on the spliced text information to obtain the spliced text information A third semantic feature corresponding to the text information; classifying the third semantic feature to obtain a classification result; and determining the classification result as the fourth similarity.

In another possible implementation, the training module is configured to obtain a first loss value corresponding to the predicted text information based on the second label information and the predicted text information; The first similarity and the first loss value of , the text mapping model is trained.

In another possible implementation, the training module is configured to obtain each of the predicted texts based on the difference between the target similarity and the first similarity corresponding to each of the predicted text information, respectively. weight parameter corresponding to the information; based on the weight parameter corresponding to each of the predicted text information, perform a weighted average of the first loss values corresponding to a plurality of the predicted text information to obtain a second loss value; The average value of the first loss values corresponding to the predicted text information is determined as a third loss value; the text mapping model is trained based on the second loss value and the third loss value.

In another possible implementation, the obtaining module includes:

an acquisition unit, configured to acquire a first text information set, where the first text information set includes a plurality of first text information and label information corresponding to each of the first text information;

a determining unit, configured to determine the similarity between each of the first text information and the corresponding label information;

A screening unit, configured to screen out the sample text information and the first label whose similarity is greater than the similarity threshold from the first text information set based on the similarity corresponding to each of the first text information information.

In another possible implementation, the screening unit is configured to, based on the similarity corresponding to each of the first text information, filter out from the first text information set those whose similarity is greater than the similarity threshold. At least one piece of second text information, the second text information and the corresponding label information filtered out form a second text information set; the sample text information and the first label are obtained from the second text information set information, the sample text information is any of the second text information.

In another possible implementation manner, the acquiring module is further configured to train the text mapping model based on the first text information and corresponding label information in the first text information set.

In another possible implementation manner, the determining module is configured to, in response to the largest first similarity among the plurality of first similarities being greater than the second similarity, determine the largest first similarity The predicted text information corresponding to the similarity is determined to be the second label information, and the multiple first similarities are the similarity between multiple predicted text information and the sample text information; or, in response to multiple Each of the first degrees of similarity is not greater than the second degree of similarity, and the first label information is determined as the second label information.

In another possible implementation manner, the mapping module is further configured to map the target text information based on the text mapping model to obtain similar text information of the target text information.

In another aspect, a computer device is provided, the computer device includes a processor and a memory, the memory stores at least one computer program, the at least one computer program is loaded and executed by the processor to achieve the above The operations performed by the processing method of the text mapping model described in the aspect.

In another aspect, a computer-readable storage medium is provided, and at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to realize the text according to the above aspects. Map the action performed by the model's handler method.

In yet another aspect, a computer program product is provided, including a computer program that, when executed by a processor, implements the operations performed by the method for processing a text mapping model according to the above aspects.

The beneficial effects brought by the technical solutions provided in the embodiments of the present application include at least:

In the method, apparatus, computer equipment, and storage medium provided by the embodiments of the present application, according to the similarity between the predicted text information and the first label information and the sample text information, respectively, the predicted text information and the first label information are compared with the sample text information. The most similar text information is used as the second label information, considering not only the difference between the predicted text information and the second label text information, but also the first similarity between the predicted text information and the sample text information, based on the second label information, predicted text information, and the first similarity corresponding to the predicted text information, and train the text mapping model to ensure that the subsequent text mapping model can map similar text information of any text information based on the trained text mapping model, thereby improving the text mapping model. mapping effect.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present application more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments. Obviously, the drawings in the following description are only some implementations of the embodiments of the present application. For example, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without any creative effort.

1 is a schematic structural diagram of an implementation environment provided by an embodiment of the present application;

2 is a flowchart of a method for processing a text mapping model provided by an embodiment of the present application;

3 is a flowchart of another method for processing a text mapping model provided by an embodiment of the present application;

4 is a schematic structural diagram of a text mapping model provided by an embodiment of the present application;

5 is a flowchart of another method for processing a text mapping model provided by an embodiment of the present application;

6 is a schematic diagram of a training text mapping model provided by an embodiment of the present application;

7 is a schematic diagram of a data comparison provided by an embodiment of the present application;

8 is a schematic diagram of an editing interface of a question and answer knowledge base provided by an embodiment of the present application;

9 is a schematic structural diagram of an apparatus for processing a text mapping model provided by an embodiment of the present application;

10 is a schematic structural diagram of a processing apparatus for another text mapping model provided by an embodiment of the present application;

FIG. 11 is a schematic structural diagram of a terminal provided by an embodiment of the present application;

FIG. 12 is a schematic structural diagram of a server provided by an embodiment of the present application.

Detailed ways

In order to make the objectives, technical solutions and advantages of the embodiments of the present application more clear, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

The terms "first", "second", "third", "fourth", etc. used in this application may be used herein to describe various concepts, but these concepts are not limited by these terms unless otherwise specified. These terms are only used to distinguish one concept from another. For example, a first degree of similarity could be referred to as a second degree of similarity, and similarly, a second degree of similarity could be referred to as a first degree of similarity, without departing from the scope of this application.

As used in this application, the terms "at least one", "plurality", "each" and "any", at least one includes one, two or more, multiple includes two or more, and each Each refers to each of the corresponding plurality, and any refers to any one of the plurality. For example, the plurality of predicted text information includes three predicted text information, and each refers to each of the three predicted text information, and any refers to any one of the three predicted text information, It can be the first predictive text information, or the second predictive text information, or the third predictive text information.

Artificial Intelligence (AI) is a theory, method, technology and application system that uses 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. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.

Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, including both hardware-level technology and software-level technology. The basic technologies of artificial intelligence generally include technologies such as sensors, special 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, speech processing technology, natural language processing technology, and machine learning/deep learning.

Natural Language Processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can realize effective communication between humans and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, the language that people use on a daily basis, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question answering, knowledge graph and other technologies.

Machine Learning (ML) is a multi-field interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or realize human learning behaviors to acquire new knowledge or skills, and to reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications are in all fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, teaching learning and other techniques.

The solution provided by the embodiment of the present application, based on the machine learning technology of artificial intelligence, can train a text mapping model, and use the trained text mapping model to map similar text information of any text information, thereby realizing the processing method of the text mapping model .

The processing method of the text mapping model provided by the embodiment of the present application is executed by a computer device. Optionally, the computer device is a terminal or a server. Optionally, the server is an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, Cloud servers for cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms. Optionally, the terminal is a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, etc., but is not limited thereto.

In some embodiments, the computer programs involved in the embodiments of the present application may be deployed and executed on one computer device, or executed on multiple computer devices located at one location, or distributed in multiple locations and communicated through Executed on a plurality of computer devices interconnected by a network, and a plurality of computer devices distributed in multiple locations and interconnected through a communication network can form a blockchain system.

In some embodiments, the computer device is provided as a server. FIG. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present application. Referring to FIG. 1 , the implementation environment includes a terminal 101 and a server 102 . The terminal 101 and the server 102 are connected through a wireless or wired network.

The server 102 is used for training the text mapping model, and after the training of the text mapping model is completed, the text mapping model is stored, and the terminal 101 sends the target text information to the server 102 through the network connection with the server, and the server 102 is used for receiving The target text information sent by the terminal 101 is mapped to the similar text information of the target text information based on the text mapping model, and the similar text information is sent to the terminal 101, and the similar text information is received by the terminal 101.

In a possible implementation manner, a target application provided by the server 102 is installed on the terminal 101, and the terminal 101 can implement, for example, a text mapping function through the target application. Optionally, the target application is a target application in the operating system of the terminal 101 or a target application provided by a third party. For example, the target application is a text mapping application, and the text mapping application has a text mapping function. Of course, the text mapping application can also have other functions, such as a review function, a question-and-answer function, a navigation function, a game function, and the like.

The terminal 101 is used to log in to the target application based on the user ID, and sends a text expansion request to the server 102 through the target application. The text expansion request carries the target text information, and the server 102 is used to receive the text expansion request sent by the terminal 101. Based on the text mapping model, map the The similar text information of the target text information is sent to the terminal 101, and the terminal 101 receives the similar text information.

2 is a flowchart of a method for processing a text mapping model provided by an embodiment of the present application, which is executed by a computer device, where the computer device is a terminal or a server. As shown in FIG. 2 , the method includes:

201. The computer device acquires sample text information and first label information, where the first label information is text information whose similarity with the sample text information is not less than a similarity threshold.

The sample text information is any type of text information, for example, the sample text information includes query sentences, answer sentences, or disease description sentences. The similarity threshold is an arbitrary value, for example, the similarity threshold is 0.8 or 3. The similarity between the first label information and the sample text information is not less than the similarity threshold, indicating that the first label information and the sample text information have the same or similar meaning. For example, the sample text information is "Simple Small Question", and the first label information is "Super Simple Question".

202. The computer device maps the sample text information based on the text mapping model to obtain predicted text information.

Wherein, the text mapping model is used to map out similar text information of any text information. Since the text mapping model is the model to be trained, the current text mapping model may be inaccurate, and the predicted text information mapped based on the text mapping model may be similar to the sample text information.

203. The computer device determines the second label information based on the first similarity between the predicted text information and the sample text information and the second similarity between the first label information and the sample text information, and the second label information is the predicted text information and the text information with a greater similarity between the first label information and the sample text information.

The first similarity is used to represent the similarity between the predicted text information and the sample text information, and the second similarity is used to represent the similarity between the first label information and the sample text information. In the embodiment of the present application, after the predicted text information is obtained based on the text mapping model, according to the similarity between the predicted text information and the first label information and the sample text information, respectively, the sample is selected from the predicted text information and the first label information. The text information with the most similar text information is used as the second label information, that is, the second label information is the text information most similar to the sample text information in the predicted text information and the first label information, so as to ensure the training effect of the subsequent text mapping model. For example, if the first similarity is greater than the second similarity, the second label information is predicted text information, and if the first similarity is not greater than the second similarity, the second label information is the first label information.

204. The computer device trains a text mapping model based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information, where the text mapping model is used to map similar text information of any text information.

In the method provided by this embodiment of the present application, according to the similarity between the predicted text information and the first label information and the sample text information respectively, the text information that is most similar to the sample text information in the predicted text information and the first label information is regarded as the second Label information, considering not only the difference between the predicted text information and the second label text information, but also the first similarity between the predicted text information and the sample text information, based on the second label information, the predicted text information and the predicted text The first similarity corresponding to the information, the text mapping model is trained to ensure that the subsequent text mapping model after training can map similar text information of any text information, thereby improving the mapping effect of the text mapping model.

On the basis of the embodiment shown in FIG. 2 above, the first similarity and the first loss value corresponding to the predicted text information can also be obtained, and based on the first similarity and the first loss value corresponding to the predicted text information, the text mapping model Carry out training, and the training process is detailed in the following examples.

301. The computer device acquires sample text information and first label information, where the first label information is text information whose similarity with the sample text information is not less than a similarity threshold.

302. The computer device maps the sample text information based on the text mapping model to obtain predicted text information.

The predicted text information is the text information that may be similar to the sample text information mapped by the text mapping model.

In a possible implementation manner, the step 302 includes: the computer device maps the sample text information based on the text mapping model to obtain at least one piece of predicted text information.

In this embodiment of the present application, the text mapping model is used to map out at least one similar text information of any text information.

In a possible implementation manner, the step 302 includes: the computer device encodes the sample text information based on the text mapping model to obtain an encoded feature, decodes the encoded feature to obtain a first word, and the first word The word feature of the word is fused with the encoding feature to obtain the first fusion feature, the first fusion feature is decoded to obtain the second word, and the first fusion feature is fused with the word feature of the second word to obtain The second fusion feature decodes the second fusion feature to obtain the third word, and so on, until the nth word is obtained, and the obtained multiple words constitute the predicted text information.

In the embodiment of the present application, in the process of generating the predicted text information corresponding to the sample text information based on the text mapping model, each word of the predicted text information is output in sequence, and when each word is output, it depends on the number of words before the word. words to ensure that the output predicted text information is as similar as possible to the sample text information, so as to ensure the mapping effect of the text mapping model. As shown in Figure 4, the text mapping model includes an encoding layer and a decoding layer. Based on the encoding layer, the sample text information is encoded to obtain encoding features. Based on the decoding layer, the encoding features are decoded to obtain the first word. Based on The decoding layer fuses the word feature of the first word with the encoding feature to obtain the first fusion feature, decodes the first fusion feature, and obtains the second word. Based on the decoding layer, the first fusion feature is Fusion with the word feature of the second word to obtain the second fusion feature, decode the second fusion feature to obtain the third word, and so on, until the nth word is obtained, the obtained multiple words constitute the predicted text information

303. The computer device acquires a third degree of similarity and a fourth degree of similarity between the predicted text information and the sample text information.

The third degree of similarity indicates the difference between words contained in the predicted text information and the sample text information, and can indicate the grammatical similarity between the predicted text information and the sample text information. The fourth similarity indicates the semantic similarity between the predicted text information and the sample text information, and can represent the semantic similarity between the predicted text information and the sample text information.

In a possible implementation manner, a plurality of predicted text information corresponding to the sample text information are obtained based on the text mapping model, then step 303 includes: the computer device obtains the third similarity and the sample text information between each predicted text information and the sample text information. Fourth similarity.

In a possible implementation manner, the process of obtaining the third degree of similarity between the predicted text information and the sample text information includes the following steps 3031-3034:

3031. Divide the predicted text information based on the number of at least one character, respectively, to obtain at least one first word set, and words belonging to the same first word set contain the same number of characters.

The number of characters is any number, for example, the number of characters is 1, 2, or 3. Divide the predicted text information according to the number of at least one type of characters to obtain at least one first word set, each first word set corresponds to a number of characters, and the words belonging to the same first word set contain the same number of characters and belong to The number of characters included in the words of the same first word set is the same as the number of characters corresponding to the first word set. For example, if at least one character number includes 1, 2, and 3, and the predicted text information is "I'm hungry", the predicted text information is divided based on the three character numbers to obtain three first word sets, the first The first word set includes "me", "belly", "child" and "hungry", that is, each word in the first first word set contains 1 character; the second first word set includes "my belly". ”, “belly” and “子 hungry”, that is, each word in the second first word set contains 2 characters; the third first word set includes “my belly” and “belly hungry”, namely Each word in the first set of words contains 3 characters.

In a possible implementation manner, step 3031 includes: for each number of characters, based on the number of characters and the target step size, dividing the predicted text information to obtain a first set of words corresponding to the number of characters.

The target step size is the number of words moved each time the words are divided, and the target step size is an arbitrary value, for example, the target step size is 1. For example, based on N-Gram (a statistical language model), the predicted text information is divided to obtain at least one first word set. For example, starting from the first character of the predicted text information, gradually move the marquee whose length is equal to the number of characters according to the target step to select the characters in the predicted text information. The characters of are regarded as a word, and the plurality of words obtained form the first word set.

3032. Divide the sample text information respectively based on the number of at least one character to obtain at least one second word set, and words belonging to the same second word set contain the same number of characters.

Wherein, each second word set corresponds to a number of characters, and the number of characters included in the words in each second word set is equal to the corresponding number of characters. This step 3032 is the same as the above-mentioned step 3031, and will not be repeated here.

3033. Determine a first number and a second number, where the first number indicates the sum of the number of different words in the first word set and the second word set corresponding to each number of characters, and the second number indicates at least one first word set and at least one word set. The total number of words in a second word set.

For the first word set and the second word set corresponding to any number of characters, determine the number of different words contained in the first word set and the second word set corresponding to the number of characters, and calculate the sum of the numbers corresponding to the number of at least one character , which is determined as the first number.

3034. Determine the ratio of the first number to the second number as the third similarity between the predicted text information and the sample text information.

By dividing the predicted text information into at least one first word set according to the number of at least one character, and dividing the sample text information into at least one second word set, the words in the first word set corresponding to different numbers of characters include characters The number is different, the third similarity between the predicted text information and the sample text information is determined according to the word sets of different granularity levels, and the grammatical similarity between the predicted text information and the sample text information can be obtained, so as to ensure the obtained The accuracy of the third similarity.

In a possible implementation manner, the process of obtaining the fourth similarity between the predicted text information and the sample text information includes the following two methods:

The first way: semantically extract the predicted text information and the sample text information, respectively, to obtain the first semantic feature of the predicted text information and the second semantic feature of the sample text information; The similarity is determined as the fourth similarity.

The first semantic feature is used to represent the predicted text information, and the second semantic feature is used to represent the sample text information. The first semantic feature and the second semantic feature can be represented in any form, for example, the first semantic feature and the second semantic feature can be represented in any form. Both semantic features are represented in the form of feature vectors.

Optionally, both the first semantic feature and the second semantic feature are represented in the form of feature vectors, and the process of determining the fourth similarity includes: determining the product of the first semantic feature vector and the second semantic feature vector as the first semantic feature vector. Four similarities.

The second method: splicing the predicted text information and the sample text information to obtain the splicing text information; performing semantic extraction on the splicing text information to obtain the third semantic feature corresponding to the splicing text information; classifying the third semantic feature to obtain Classification result; determine the classification result as the fourth similarity.

The third semantic feature is used to represent the spliced text information, and the third semantic feature incorporates the feature of the predicted text information and the feature of the sample text information, that is, the third semantic feature can represent the predicted text information and the sample text. information.

By processing the predicted text information and the sample text information by splicing first and then classifying it, the changes between the predicted text information and the core words in the sample text information can be captured, so as to determine semantically. Whether the predicted text information and the sample text information are the same, so as to ensure the accuracy of the obtained fourth similarity.

For example, the predicted text information is "why the label is displayed in gray", and the sample text information is "why the tag is displayed in brown", that is, the core words "gray" and "brown" of the two text information are different. In this way, the obtained fourth similarity is small, so that it can be determined that the meanings of the predicted text information and the sample text information are different.

Optionally, using a self-attention method, semantically extract the spliced text information to obtain a third semantic feature, normalize the third semantic feature to obtain a classification result, and determine the classification result as the fourth similarity. .

In the process of semantically extracting the spliced text information by means of self-attention, the predicted text information and the sample text information can be cross-encoded, so as to ensure that the third semantic feature is obtained through the features of the predicted text information and the The feature obtained by the feature fusion of the sample text information, and then the third semantic feature is normalized to determine whether the predicted text information is the same as the core words in the sample text information, and the classification result indicates the predicted text. The degree of similarity between the information and the core words in the sample text information, so as to determine the fourth degree of similarity.

It should be noted that, in this embodiment of the present application, the above two methods are used to obtain the fourth similarity, while in another embodiment, the above two methods can be combined, and the similarity obtained by the above two methods is Weighted fusion to obtain the fourth similarity. In a possible implementation manner, the similarity obtained according to the first method above is the fifth similarity, the similarity obtained according to the second method above is the sixth similarity, and the fifth similarity is similar to the sixth similarity Degree-weighted fusion to obtain the fourth similarity.

304. The computer device performs weighted fusion on the third similarity and the fourth similarity to obtain a first similarity between the predicted text information and the sample text information.

The first similarity is used to represent the similarity between the predicted text information and the sample text information. Considering the similarity between the first text information and the label information in various ways, the third similarity and the fourth similarity are obtained, and the third similarity and the fourth similarity are weighted and fused to ensure the obtained predicted text The accuracy of the similarity between the information and the sample text information. The present application provides a scoring mechanism combining semantics and grammar. Based on the scoring mechanism, the similarity between the predicted text information and the sample text information is determined, thereby ensuring the accuracy of the determined first similarity.

In a possible implementation manner, the fourth similarity is obtained by weighting the fifth similarity and the sixth similarity, then step 303 includes: weighting the third similarity, the fifth similarity and the sixth similarity Fusion to get the first similarity.

Optionally, the third similarity, the fifth similarity, the sixth similarity and the first similarity satisfy the following relationship:

Score(all)=2*Score( _sbert )+2*Score(common bert)+1*Score(diversity)

Among them, Score(all) is used to represent the first similarity, Score( _sbert ) is used to represent the fifth similarity, Score(common berr) is used to represent the sixth similarity, and Score(diversity) is used to represent the third similarity Spend.

It should be noted that, in this embodiment of the present application, the third similarity and the fourth similarity between the predicted text information and the sample text information are obtained first, and then the third similarity and the fourth similarity are weighted and fused to obtain the prediction. The fourth similarity between the text information and the sample text information, and in another embodiment, it is not necessary to perform steps 303-304, and other methods can be used to obtain the information based on the predicted text information and the sample text information. the first similarity.

305. The computer device determines the second label information based on the first similarity between the predicted text information and the sample text information and the second similarity between the first label information and the sample text information, and the second label information is the predicted text information and the text information with a greater similarity between the first label information and the sample text information.

The first similarity is used to represent the similarity between the predicted text information and the sample text information, and the second similarity is used to represent the similarity between the first label information and the sample text information. By comparing the first similarity and the sample text information The second degree of similarity can determine the text information that is most similar to the sample text information from the predicted text information and the first label information, and use the most similar text information as the second label information. The manner of acquiring the second degree of similarity between the first label information and the sample text information is the same as the manner of acquiring the first degree of similarity between the predicted text information and the sample text information, and will not be repeated here.

In a possible implementation manner, this step 305 includes the following two manners:

The first way: in response to that the largest first similarity among the multiple first similarities is greater than the second similarity, the predicted text information corresponding to the largest first similarity is determined as the second label information, and the multiple first similarity is determined as the second label information. A similarity is the similarity between the plurality of predicted text information and the sample text information.

The second manner: in response to the multiple first degrees of similarity being not greater than the second degrees of similarity, determining the first label information as the second label information.

In the embodiment of the present application, based on the text mapping model, the sample text information is mapped to obtain a plurality of predicted text information, and then the first similarity between each predicted text information and the sample text information and the first label information and the first label information are determined. For the second similarity between the sample text information, the first similarity is compared with the second similarity to determine the maximum similarity, and the information corresponding to the maximum similarity is determined as the second label information. For example, if the maximum similarity is the second similarity, the first label information corresponding to the second similarity is determined as the second label information, and if the maximum similarity is a first similarity, the first similarity The corresponding predicted text information is determined as the second label information.

In a possible implementation manner, the method further includes: when the second label information is predicted text information, updating the first label information corresponding to the sample text information in the second text information to the second label information.

If the first similarity between the predicted text information and the sample text information is greater than the second similarity between the first label information and the sample text information, it means that the predicted text information is more similar to the sample text information, that is, the predicted text information is more similar to the sample text information. It is suitable as the label information of the sample text information. Therefore, the first label information corresponding to the sample text information in the second text information set is updated to ensure that the second text information and the corresponding label information in the second text information set are updated. To ensure that the quality of the second text information set is higher, when the text mapping model is subsequently trained based on the second text information in the second text information set and the corresponding label information, the mapping effect of the text mapping model can be guaranteed.

306. The computer device acquires, based on the second label information and the predicted text information, a first loss value corresponding to the predicted text information.

The difference between the predicted text information and the second label information can represent the accuracy of the text mapping model. Therefore, the first loss value is determined based on the predicted text information and the second label information, so as to facilitate subsequent A text mapping model is trained based on the first loss value.

In a possible implementation manner, the step 306 includes: obtaining a first loss value corresponding to each predicted text information based on the second label information and each predicted text information.

In the embodiment of the present application, based on the text mapping model, a plurality of predicted text information corresponding to the sample text information can be mapped, and then based on each predicted text information and the second label information, the first predicted text information corresponding to each predicted text information can be determined. A loss value for subsequent training of the text mapping model based on the plurality of first loss values.

307. The computer device trains the text mapping model based on the first similarity and the first loss value corresponding to the predicted text information.

Both the first similarity and the first loss value corresponding to the predicted text information can reflect the accuracy of the text mapping model. Based on the first similarity and the first loss value corresponding to the predicted text information, the text mapping model is trained to improve the The mapping effect of the text mapping model.

In a possible implementation, the step 307 includes the following steps 3071-3074:

3071. The computer device obtains a weight parameter corresponding to each predicted text information based on the difference between the target similarity and the first similarity corresponding to each predicted text information, respectively.

Among them, the target similarity is an arbitrary numerical value, which is used to represent the maximum similarity between the expected predicted text information and the sample text information. For example, the target similarity is 5, and the first similarity corresponding to each predicted text information is not greater than 5. The weight parameter indicates whether each predicted text information is accurate, and can also reflect the mapping effect of the text mapping model. Based on the weight parameter, the size of the loss value can be adjusted. The larger the weight parameter, the less accurate the predicted text information. The larger the loss value; the smaller the weight parameter, the more accurate the predicted text information, and the smaller the subsequent calculated loss value.

In a possible implementation manner, the target similarity, the first similarity corresponding to any predicted text information, and the weight parameter satisfy the following relationship:

Among them, rewards is used to represent the weight parameter corresponding to any predicted text information, μ is used to represent the hyperparameter, μ is a constant, score _max is used to represent the target similarity, and score _real is used to represent the first similarity corresponding to the predicted text information Spend.

3072. Based on the weight parameter corresponding to each predicted text information, the computer device performs a weighted average of the first loss values corresponding to the multiple predicted text information to obtain a second loss value.

Taking into account the accuracy of each predicted text information, a weighted average of the first loss values corresponding to the multiple predicted text information is performed through the weight parameter corresponding to each predicted text information to ensure that the obtained second loss value is accurate.

In a possible implementation manner, the weight parameter corresponding to each predicted text information, the first loss value and the second loss value corresponding to multiple predicted text information satisfy the following relationship:

loss _rl = mean(loss _mle-batch ·rewards)

Among them, loss _rl is used to represent the second loss value, loss _mle-batch is used to represent the first loss value corresponding to the predicted text information, rewards is used to represent the weight parameter corresponding to the predicted text information, mean( ) is used to represent the average function of value.

3073. The computer device determines the average value of the first loss values corresponding to the plurality of predicted text information as the third loss value.

The first loss value corresponding to each predicted text information can reflect the accuracy of the text mapping model. Considering the first loss values corresponding to multiple predicted text information comprehensively, the average value of the first loss values corresponding to each predicted text information is calculated. As the third loss value, the text mapping model is subsequently trained with the third loss value to ensure the accuracy of the training text mapping model, thereby improving the mapping effect of the text mapping model.

In a possible implementation manner, the third loss value and the first loss values corresponding to multiple predicted text information satisfy the following relationship:

loss _mle = mean(loss _mle-batch )

Among them, loss _mle is used to represent the third loss value, loss _mle-batch is used to represent the first loss value corresponding to multiple predicted text information, and mean( ) is used to represent the average value.

3074. The computer device trains the text mapping model based on the second loss value and the third loss value.

Since both the second loss value and the third loss value can reflect the accuracy of the text mapping model, the text mapping model is trained based on the second loss value and the third loss value, so as to improve the mapping effect of the text mapping model.

In a possible implementation manner, step 3074 includes: performing weighted fusion on the second loss value and the third loss value to obtain a fourth loss value, and training the text mapping model based on the fourth loss value.

Optionally, the second loss value, the third loss value and the fourth loss value satisfy the following relationship:

loss=λloss _rl +(1-λ)loss _mle

Among them, loss is used to represent the fourth loss value, λ is used to represent the hyperparameter, λ is a constant, loss _rl is used to represent the second loss value, and loss _mle is used to represent the third loss value.

The text mapping model is trained through the second loss value and the third loss value, taking into account the influence of the first similarity between the predicted text information and the sample text information, so that when the predicted text information and the sample text information are sufficiently similar , reduce the loss value to reduce the adjustment range of adjusting the text model, and increase the loss value to speed up the convergence and optimization of the text mapping model when the predicted text information is not similar to the sample text information, so as to ensure that the final training results The mapping effect of the text mapping model.

It should be noted that, in this embodiment of the present application, the text mapping model is trained by predicting the first loss value and the first similarity corresponding to the text information. In other manners, the text mapping model is trained based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information.

It should be noted that the embodiment of the present application is only described by performing an iterative training on the text mapping model based on the sample text information and the corresponding first label information. The text mapping model is trained for multiple iterations. In a possible implementation manner, in the process of iteratively training the text mapping model according to the above steps 301-307, in response to the number of iterations reaching the first value, the training of the text mapping model is stopped; or, in response to the current iteration number of The fourth loss value is smaller than the second value, and the training of the text mapping model is stopped.

The first value is used to represent the maximum number of iterations, and both the first value and the second value are arbitrary values, for example, the first value is 100, and the second value is 0.3.

It should be noted that the embodiment of the present application is only described by performing an iterative training on the text mapping model based on one sample text information and the corresponding first label information. In another embodiment, the text mapping model is trained once. During iterative training, the text mapping model can be trained based on multiple sample text information and the first label information corresponding to each sample text information. For example, obtain a plurality of sample text information and the first label information corresponding to each sample text information from the second text information set, and then according to the above steps 301-306, obtain the predicted text information corresponding to each sample text information, and obtain The first similarity and the first loss value corresponding to each predicted text information are based on the first similarity and the first loss value corresponding to the plurality of predicted text information, and the text mapping model is trained according to the above step 307 .

In a possible implementation manner, after step 307, the method further includes: mapping the target text information based on the text mapping model to obtain similar text information of the target text information.

The target text information is any text information. After training the text mapping model, similar text information of the target text information can be mapped based on the text mapping model.

Optionally, based on the text mapping model, the target text information is mapped to obtain a plurality of similar text information of the target text information.

In the method provided by this embodiment of the present application, according to the similarity between the predicted text information and the first label information and the sample text information respectively, the text information that is most similar to the sample text information in the predicted text information and the first label information is regarded as the second Label information, considering not only the difference between the predicted text information and the second label text information, but also the first similarity between the predicted text information and the sample text information, based on the second label information, the predicted text information and the predicted text The first similarity corresponding to the information, the text mapping model is trained to ensure that the subsequent text mapping model after training can map similar text information of any text information, thereby improving the mapping effect of the text mapping model.

Moreover, the present application provides a scoring mechanism combining semantics and grammar, so as to determine the similarity between the first text information and the corresponding label information, thereby ensuring the accuracy of the determined similarity.

In addition, in the process of training the text mapping model, the label information is updated, so that the updated label information is the text information that is most similar to the sample text information among the predicted text information and the label text information, so as to ensure that the label information is consistent. Accuracy, based on the updated label information, predicted text information and the first similarity corresponding to the predicted text information, to train the text mapping model, which can improve the mapping effect of the text mapping model and avoid the use of fixed labels. training, so that the text mapping model has better generalization ability.

On the basis of the embodiment shown in FIG. 2 , the text mapping model can also be trained in multiple stages, that is, the first stage of the text mapping model is performed based on the first text information and the corresponding label information in the first text information set. After training, the first text information set is screened, and the text mapping model is trained in the second stage based on the second text information set obtained by screening. For details of the training process, refer to the following embodiments.

5 is a flowchart of a method for processing a text mapping model provided by an embodiment of the present application, which is executed by a computer device. As shown in FIG. 5 , the method includes:

501. The computer device acquires a first text information set.

The first text information set includes a plurality of first text information and label information corresponding to each first text information, and the first text information is any type of text information. For example, the first text information includes a query sentence, an answer sentence Or disease description sentences, etc. Label information corresponding to each first text information is text information that may be similar to the first text information.

502. The computer device trains the text mapping model based on the first text information and the corresponding label information in the first text information set.

The text mapping model is an arbitrary network model, for example, the text mapping model is Seq2Seq (Sequence to Sequence, a neural network model), or a Transformer (an attention network model).

The text mapping model is trained according to the first text information and the corresponding label information in the first text information set, so that the trained text mapping model has preliminary text mapping capabilities and can map similar text information of any text information .

In a possible implementation manner, step 502 includes: mapping the first text information based on the text mapping model to obtain predicted text information corresponding to the first text information, and based on the predicted text information corresponding to the first text information and label information, and train the text mapping model.

Based on the text mapping model, the predicted text information corresponding to the first text information is mapped, and the difference between the predicted text information corresponding to the first text information and the label information can reflect the quality of the mapping effect of the text mapping model. The predicted text information and the label information corresponding to the first text information train the text mapping model, so as to improve the mapping effect of the text mapping model.

Optionally, the process of training the text mapping model includes: determining a fifth loss value based on the predicted text information and label information corresponding to the first text information, and training the text mapping model based on the fifth loss value.

The fifth loss value indicates the degree of difference between the predicted text information corresponding to the first text information and the label information, which can reflect the quality of the mapping effect of the text mapping model. For example, using a cross-entropy loss function, the fifth loss value is determined based on the predicted text information and label information corresponding to the first text information. The text mapping model is trained through the fifth loss value to improve the mapping effect of the text mapping model.

In a possible implementation manner, the step 502 includes: performing iterative training on the text mapping model based on a plurality of first text information and label information corresponding to each first text information.

Iteratively trains the text mapping model for many times to improve the mapping effect of the text mapping model as much as possible.

Optionally, in the process of iteratively training the text mapping model, in response to the number of iterations reaching the third numerical value, the training of the text mapping model is stopped; or, in response to the fifth loss value of the current number of iterations being less than the fourth numerical value, Stop training the text mapping model.

The third value is used to represent the maximum number of iterations, and both the third value and the fourth value are arbitrary values, for example, the third value is 100, and the fourth value is 0.3.

503. The computer device determines the similarity between each first text information and the corresponding tag information.

The similarity between each first text information and the corresponding label information can reflect the similarity between the first text information and the corresponding label information.

In a possible implementation manner, the step 503 includes: for any first text information and corresponding label information, obtaining the seventh similarity and the eighth similarity between the first text information and the label information, The seventh similarity and the eighth similarity are weighted and fused to obtain the similarity between the first text information and the label information.

Wherein, the seventh degree of similarity indicates the difference between the words contained in the first text information and the label information, and can indicate the degree of grammatical similarity between the first text information and the label information. The eighth degree of similarity indicates the semantic similarity between the first text information and the label information, and can indicate the degree of semantic similarity between the first text information and the label information. Considering the similarity between the first text information and the label information in various ways, the seventh similarity and the eighth similarity are obtained, and the seventh similarity and the eighth similarity are weighted and fused to ensure that the obtained first The accuracy of the similarity between the text information and the label information. The present application provides a scoring mechanism combining semantics and grammar, and based on the scoring mechanism, the similarity between the first text information and the corresponding label information is determined, thereby ensuring the accuracy of the determined similarity.

Optionally, the process of acquiring the seventh degree of similarity between the first text information and the label information includes:

5031. Divide the first text information based on the number of at least one character, respectively, to obtain at least one third word set, and words belonging to the same third word set contain the same number of characters.

5032. Divide the label information respectively based on the number of at least one character to obtain at least one fourth word set, and words belonging to the same fourth word set contain the same number of characters.

5033. Determine a third number and a fourth number, where the third number indicates the sum of the number of different words in the third word set and the fourth word set corresponding to each number of characters, and the fourth number indicates at least one third word set and at least one different word set. The total number of words in a fourth word set.

5034. Determine the ratio of the third number to the fourth number as the seventh degree of similarity between the first text information and the tag information.

The steps 5031-5034 are the same as the above-mentioned steps 3031-3034, and are not repeated here.

Optionally, the process of acquiring the eighth degree of similarity between the first text information and the label information includes the following two ways:

The first way: perform semantic extraction on the first text information and the label information, respectively, to obtain the fourth semantic feature of the first text information and the fifth semantic feature of the label information; the fourth semantic feature and the fifth semantic feature are obtained. The similarity between the features is determined as the eighth similarity.

The first manner of acquiring the eighth degree of similarity is the same as the above-mentioned first manner of acquiring the fourth degree of similarity, and details are not described herein again.

The second way: splicing the first text information and the label information to obtain the first splicing text information; performing semantic extraction on the first splicing text information to obtain the sixth semantic feature corresponding to the first splicing text information; Six semantic features are classified and processed to obtain a classification result; the classification result is determined as the eighth degree of similarity.

The second manner of acquiring the eighth degree of similarity is the same as the second manner of acquiring the fourth degree of similarity described above, and details are not described herein again.

It should be noted that, in this embodiment of the present application, the above two methods are used to obtain the eighth similarity, while in another embodiment, the above two methods can be combined, and the similarity obtained by the above two methods Weighted fusion, the eighth similarity is obtained. In a possible implementation manner, the similarity obtained according to the first method above is the ninth similarity, the similarity obtained according to the second method above is the tenth similarity, and the ninth similarity is similar to the tenth similarity The degree-weighted fusion is used to obtain the eighth degree of similarity.

504. Based on the similarity corresponding to each first text information, the computer device filters out at least one second text information whose similarity is greater than the similarity threshold from the first text information set, and selects the filtered second text information and the corresponding The tag information constitutes a second set of text information.

The similarity threshold is an arbitrary threshold, for example, the similarity threshold is 0.8, or 3, and so on.

In the embodiment of the present application, the first text information set includes a plurality of first text information and label information corresponding to each first text information, each first text information and the corresponding label information constitute an information combination, and the first text information The quality of the information combination in the information set is different, and there may be a poor information combination, that is, the first text information in the information combination is not similar to the label information. If the text mapping model is trained according to the poor information combination, it will lead to The mapping effect of the text mapping model is poor. Therefore, by screening the text information and the corresponding label information in the first text information set, a second text information set is generated to ensure that each second text information in the second text information set and the corresponding label information are consistent with each other. The similarity between the two is greater than the similarity threshold, that is, the second text information set includes high-quality information combinations, and the subsequent training of the text mapping model based on the second text information and the corresponding label information in the second text information set can ensure that The mapping effect of the text mapping model.

It should be noted that, in this embodiment of the present application, the text mapping model is trained based on the first text information in the first text information set and the corresponding label information, and then the second text information set is filtered based on the first text information set. However, in another embodiment, the above steps 503-504 can be performed first to filter out the second text information set, and then the above step 502 can be performed. The present application does not limit the order of the execution steps.

505. The computer device acquires sample text information and first label information from the second text information set, where the sample text information is any second text information.

In this embodiment of the present application, the second text information set includes at least one second text information and label information corresponding to each second text information, and any second text information is obtained from the second text information set as sample text information , and obtain the first label information corresponding to the sample text information.

It should be noted that, in this embodiment of the present application, the first text information set is first screened to obtain the second text information set, and the sample text information and the first label information are obtained from the second text information set. , there is no need to perform steps 504-505, and other methods can be adopted to screen out the sample text information and the first label information whose similarity is greater than the similarity threshold from the first text information set based on the similarity corresponding to each first text information .

It should be noted that, in the embodiment of the present application, the text mapping model is initially trained based on the first text information set, and then the first text information in the first text information set is screened to construct the second text information set, The sample text information and the first label information are obtained from the second text information set, and in another embodiment, there is no need to first train the text mapping model based on the first text information set, and there is no need to perform steps 501-504. Other methods are used to obtain sample text information and first label information, where the first label information is text information whose similarity with the sample text information is not less than a similarity threshold.

506. The computer device maps the sample text information based on the text mapping model to obtain predicted text information.

507. The computer device determines the second label information based on the first similarity between the predicted text information and the sample text information and the second similarity between the first label information and the sample text information, and the second label information is the predicted text information and the text information with a greater similarity between the first label information and the sample text information.

508. The computer device trains a text mapping model based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information, where the text mapping model is used to map similar text information of any text information.

The steps 506-508 are the same as the above-mentioned steps 302-307, and are not repeated here.

In the method provided by this embodiment of the present application, according to the similarity between the predicted text information and the first label information and the sample text information respectively, the text information that is most similar to the sample text information in the predicted text information and the first label information is regarded as the second Label information, considering not only the difference between the predicted text information and the second label text information, but also the first similarity between the predicted text information and the sample text information, based on the second label information, the predicted text information and the predicted text The first similarity corresponding to the information, the text mapping model is trained to ensure that the subsequent text mapping model after training can map similar text information of any text information, thereby improving the mapping effect of the text mapping model.

In addition, in the process of training the text mapping model, the label information corresponding to the sample text information in the second text information set is updated, so that the updated label information is the predicted text information and the label text information and the sample text information. The most similar text information to ensure the accuracy of the label information, in order to train the text mapping model, it can improve the mapping effect of the text mapping model, avoid training the model with fixed labels, and make the text mapping model more efficient. good generalization ability.

In addition, by screening the initial first text information set, the low-quality information combinations in the first text information set are cleaned up, so as to ensure that the second text information in the screened second text information and the corresponding tag information are indistinguishable. The similarity is greater than the similarity threshold, which improves the quality of the information combination in the second text information set, and the text mapping model is trained with the second text information in the second text information set and the corresponding label information, so as to ensure the training of the text mapping model. The accuracy of the sample information, so as to ensure the training effect of the subsequent training text mapping model.

In addition, the present application adopts multi-stage training for the text mapping model. First, the text mapping model is initially trained based on the first text information set, so that the text mapping model has preliminary text mapping capabilities, and then the screened second text is used. The information set trains the text mapping model to further improve the mapping effect of the text mapping model.

Moreover, the present application provides a scoring mechanism combining semantics and grammar, so as to determine the similarity between the first text information and the corresponding label information, thereby ensuring the accuracy of the determined similarity.

As shown in Figure 6, the text mapping model is trained in two stages. In the first training stage, the text mapping model is trained based on the first text information set. In the second training stage, a scoring mechanism is used, based on the first text information set, filter out the second text information set, and train the text mapping model based on the second text information set. Moreover, in the training process, a scoring mechanism is used to determine the first similarity between the sample text information and the predicted text information and the second similarity between the first label information and the sample text information, based on the first similarity and The second similarity is to determine the second label information, and based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information, the text mapping model is intensively trained.

As shown in FIG. 7 , the processing method of the text mapping model provided by the embodiment of the present application is compared with the processing method of the text mapping model in the related art. Related technology 1 is to train the text mapping model directly based on the original training data set, and related technology 2 is to filter the original training data set, adopt the cross-entropy loss function, and pass the filtered training data set. to train. By comparison, it can be seen that the amount of training data obtained by the method provided in the embodiment of the present application is the smallest, but the amount of similar text information mapped by the text mapping model provided by the embodiment of the present application is large, and the quality of the mapped similar text information is high, that is, The text mapping model provided by the embodiment of the present application has better performance and better mapping effect.

The processing method of the text mapping model provided by the embodiment of the present application can be applied in various scenarios, for example, in the scenario of intelligent question answering, and the question and answer knowledge base can be expanded based on the text mapping model. As shown in Figure 8, the question-and-answer knowledge base of the intelligent dialogue robot is configured with multiple questions and the answer to each question. Editing interface, the user can edit questions and corresponding answers in the question and answer editing interface, map the question input in the question and answer editing interface based on the text mapping model, obtain a plurality of similar questions of the question, and display the prompt identification 801 to Prompt the user to map out multiple similar questions of the current input question. The user clicks the similar question option to display multiple similar questions. The user clicks the save option to associate the input question with the corresponding multiple similar questions, which enriches the Q&A knowledge base. There are multiple questions corresponding to each answer. When the subsequent user has a dialogue with the intelligent dialogue robot, the user enters any question, and the intelligent dialogue robot can find the answer to any question from the Q&A knowledge base, thereby improving the user experience and avoiding It solves the situation that the intelligent dialogue robot cannot answer the questions due to the few questions configured in the Q&A knowledge base.

FIG. 9 is a schematic structural diagram of an apparatus for processing a text mapping model provided by an embodiment of the present application. As shown in FIG. 9 , the apparatus includes:

an acquisition module 901, configured to acquire sample text information and first label information, where the first label information is text information whose similarity with the sample text information is not less than a similarity threshold;

A mapping module 902, configured to map the sample text information based on the text mapping model to obtain predicted text information;

The determining module 903 is configured to determine the second label information based on the first similarity between the predicted text information and the sample text information and the second similarity between the first label information and the sample text information, and the second label information is the prediction Text information with a greater similarity between the text information and the first label information and the sample text information;

The training module 904 is configured to train a text mapping model based on the second label information, the predicted text information and the first similarity corresponding to the predicted text information, and the text mapping model is used to map similar text information of any text information.

In a possible implementation manner, as shown in FIG. 10 , the apparatus further includes:

The obtaining module 901 is further configured to obtain the third similarity and the fourth similarity between the predicted text information and the sample text information, the third similarity indicates the difference between the words contained in the predicted text information and the sample text information, and the fourth similarity The degree indicates the semantic similarity between the predicted text information and the sample text information;

The fusion module 905 is used for weighted fusion of the third similarity and the fourth similarity to obtain the first similarity.

In another possible implementation manner, the obtaining module 901 is configured to divide the predicted text information based on the number of at least one character, respectively, to obtain at least one first word set, and words belonging to the same first word set include the number of characters are the same; based on the number of at least one character, the sample text information is divided respectively to obtain at least one second word set, and the words belonging to the same second word set contain the same number of characters; determine the first number and the second number, the first The number indicates the sum of the number of different words in the first word set and the second word set corresponding to each number of characters, and the second number indicates the total number of words in at least one first word set and at least one second word set; The ratio of the first number to the second number is determined as the third degree of similarity between the predicted text information and the sample text information.

In another possible implementation manner, the obtaining module 901 is configured to perform semantic extraction on the predicted text information and the sample text information respectively, so as to obtain the first semantic feature of the predicted text information and the second semantic feature of the sample text information; The similarity between the semantic feature and the second semantic feature is determined as the fourth similarity.

In another possible implementation, the acquisition module 901 is configured to splicing the predicted text information and the sample text information to obtain the spliced text information; perform semantic extraction on the spliced text information to obtain the third semantic feature corresponding to the spliced text information; The third semantic feature is classified to obtain a classification result; the classification result is determined as the fourth similarity.

In another possible implementation manner, the training module 904 is configured to obtain the first loss value corresponding to the predicted text information based on the second label information and the predicted text information; based on the first similarity and the first loss corresponding to the predicted text information value to train the text mapping model.

In another possible implementation manner, the training module 904 is configured to obtain a weight parameter corresponding to each predicted text information based on the difference between the target similarity and the first similarity corresponding to each predicted text information; For the weight parameter corresponding to each predicted text information, the first loss value corresponding to the multiple predicted text information is weighted and averaged to obtain the second loss value; the average value of the first loss values corresponding to the multiple predicted text information is determined as The third loss value; the text mapping model is trained based on the second loss value and the third loss value.

In another possible implementation manner, as shown in FIG. 10 , the obtaining module 901 includes:

The obtaining unit 9011 is configured to obtain a first text information set, where the first text information set includes a plurality of first text information and label information corresponding to each first text information;

a determining unit 9012, configured to determine the similarity between each first text information and the corresponding label information;

The screening unit 9013 is configured to, based on the similarity corresponding to each first text information, filter out the sample text information and the first label information whose similarity is greater than the similarity threshold from the first text information set.

In another possible implementation, the screening unit 9013 is configured to, based on the similarity corresponding to each first text information, filter out at least one second text information whose similarity is greater than the similarity threshold from the first text information set, The filtered second text information and the corresponding label information constitute a second text information set; sample text information and the first label information are obtained from the second text information set, and the sample text information is any second text information.

In another possible implementation manner, the obtaining module 901 is further configured to train the text mapping model based on the first text information and the corresponding label information in the first text information set.

In another possible implementation manner, the determining module 903 is configured to, in response to the largest first similarity among the multiple first similarities being greater than the second similarity, determine the predicted text information corresponding to the largest first similarity, to determine is the second label information, and the multiple first similarities are the similarities between the multiple predicted text information and the sample text information; or, in response to the multiple first similarities being no greater than the second similarity, the first label is The information is determined to be the second tag information.

In another possible implementation manner, the mapping module 902 is further configured to map the target text information based on the text mapping model to obtain similar text information of the target text information.

It should be noted that: the processing device of the text mapping model provided in the above-mentioned embodiment is only illustrated by the division of the above-mentioned functional modules. The internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus for processing a text mapping model provided in the above embodiment and the embodiment of the method for processing a text mapping model belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

An embodiment of the present application further provides a computer device, the computer device includes a processor and a memory, the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the text mapping model of the above embodiment The operation performed by the processing method.

Optionally, the computer equipment is provided as a terminal. FIG. 11 shows a structural block diagram of a terminal 1100 provided by an exemplary embodiment of the present application. The terminal 1100 may be a portable mobile terminal, such as a smart phone, a tablet computer, an MP3 player (Moving Picture Experts Group Audio Layer III, the standard audio layer 3 of the moving picture experts compression), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic Video Expert Compresses Standard Audio Layer 4) Player, Laptop or Desktop. Terminal 1100 may also be called user equipment, portable terminal, laptop terminal, desktop terminal, and the like by other names.

The terminal 1100 includes: a processor 1101 and a memory 1102 .

The processor 1101 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor 1101 may use at least one hardware form among DSP (Digital Signal Processing, digital signal processing), FPGA (Field-Programmable Gate Array, field programmable gate array), PLA (Programmable Logic Array, programmable logic array) accomplish. The processor 1101 may also include a main processor and a coprocessor. The main processor is a processor used to process data in a wake-up state, also called a CPU (Central Processing Unit, central processing unit); A low-power processor for processing data in a standby state. In some embodiments, the processor 1101 may be integrated with a GPU (Graphics Processing Unit, image processor), and the GPU is used for rendering and drawing the content that needs to be displayed on the display screen. In some embodiments, the processor 1101 may further include an AI (Artificial Intelligence, artificial intelligence) processor, where the AI processor is used to process computing operations related to machine learning.

Memory 1102 may include one or more computer-readable storage media, which may be non-transitory. Memory 1102 may also include high-speed random access memory, as well as non-volatile memory, such as one or more disk storage devices, flash storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1102 is used to store at least one computer program, and the at least one computer program is used to be executed by the processor 1101 to implement the methods provided by the method embodiments in this application. The processing method of the text map model.

In some embodiments, the terminal 1100 may optionally further include: a peripheral device interface 1103 and at least one peripheral device. The processor 1101, the memory 1102 and the peripheral device interface 1103 may be connected through a bus or a signal line. Each peripheral device can be connected to the peripheral device interface 1103 through a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 1104 , a display screen 1105 , a camera assembly 1106 , an audio circuit 1107 , a positioning assembly 1108 and a power supply 1109 .

The peripheral device interface 1103 may be used to connect at least one peripheral device related to I/O (Input/Output) to the processor 1101 and the memory 1102 . In some embodiments, processor 1101, memory 1102, and peripherals interface 1103 are integrated on the same chip or circuit board; in some other embodiments, any one of processor 1101, memory 1102, and peripherals interface 1103 or The two can be implemented on a separate chip or circuit board, which is not limited in this embodiment.

The radio frequency circuit 1104 is used for receiving and transmitting RF (Radio Frequency, radio frequency) signals, also called electromagnetic signals. The radio frequency circuit 1104 communicates with communication networks and other communication devices via electromagnetic signals. The radio frequency circuit 1104 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. Optionally, the radio frequency circuit 1104 includes an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a subscriber identity module card, and the like. The radio frequency circuit 1104 may communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes but is not limited to: World Wide Web, Metropolitan Area Network, Intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), wireless local area network and/or WiFi (Wireless Fidelity, Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1104 may further include a circuit related to NFC (Near Field Communication, short-range wireless communication), which is not limited in this application.

The display screen 1105 is used for displaying UI (User Interface, user interface). The UI can include graphics, text, icons, video, and any combination thereof. When the display screen 1105 is a touch display screen, the display screen 1105 also has the ability to acquire touch signals on or above the surface of the display screen 1105 . The touch signal can be input to the processor 1101 as a control signal for processing. At this time, the display screen 1105 may also be used to provide virtual buttons and/or virtual keyboards, also referred to as soft buttons and/or soft keyboards. In some embodiments, there may be one display screen 1105, which is arranged on the front panel of the terminal 1100; in other embodiments, there may be at least two display screens 1105, which are respectively arranged on different surfaces of the terminal 1100 or in a folded design; In other embodiments, the display screen 1105 may be a flexible display screen, which is disposed on a curved surface or a folding surface of the terminal 1100 . Even, the display screen 1105 can also be set as a non-rectangular irregular figure, that is, a special-shaped screen. The display screen 1105 can be made of materials such as LCD (Liquid Crystal Display, liquid crystal display), OLED (Organic Light-Emitting Diode, organic light emitting diode).

The camera assembly 1106 is used to capture images or video. Optionally, the camera assembly 1106 includes a front camera and a rear camera. The front camera is arranged on the front panel of the terminal, and the rear camera is arranged on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth-of-field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth-of-field camera to realize the background blur function, the main camera It is integrated with the wide-angle camera to achieve panoramic shooting and VR (Virtual Reality, virtual reality) shooting functions or other integrated shooting functions. In some embodiments, the camera assembly 1106 may also include a flash. The flash can be a single color temperature flash or a dual color temperature flash. Dual color temperature flash refers to the combination of warm light flash and cold light flash, which can be used for light compensation under different color temperatures.

Audio circuitry 1107 may include a microphone and speakers. The microphone is used to collect the sound waves of the user and the environment, convert the sound waves into electrical signals, and input them to the processor 1101 for processing, or to the radio frequency circuit 1104 to realize voice communication. For the purpose of stereo collection or noise reduction, there may be multiple microphones, which are respectively disposed in different parts of the terminal 1100 . The microphone may also be an array microphone or an omnidirectional collection microphone. The speaker is used to convert the electrical signal from the processor 1101 or the radio frequency circuit 1104 into sound waves. The loudspeaker can be a traditional thin-film loudspeaker or a piezoelectric ceramic loudspeaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert electrical signals into sound waves audible to humans, but also convert electrical signals into sound waves inaudible to humans for distance measurement and other purposes. In some embodiments, audio circuitry 1107 may also include a headphone jack.

The positioning component 1108 is used to locate the current geographic location of the terminal 1100 to implement navigation or LBS (Location Based Service, location-based service). The positioning component 1108 may be a positioning component based on the GPS (Global Positioning System, global positioning system) of the United States, the Beidou system of China or the Galileo system of Russia.

The power supply 1109 is used to power various components in the terminal 1100 . The power source 1109 may be alternating current, direct current, disposable batteries, or rechargeable batteries. When the power source 1109 includes a rechargeable battery, the rechargeable battery may be a wired rechargeable battery or a wireless rechargeable battery. Wired rechargeable batteries are batteries that are charged through wired lines, and wireless rechargeable batteries are batteries that are charged through wireless coils. The rechargeable battery can also be used to support fast charging technology.

In some embodiments, the terminal 1100 also includes one or more sensors 1110 . The one or more sensors 1110 include, but are not limited to, an acceleration sensor 1111 , a gyro sensor 1112 , a pressure sensor 1113 , a fingerprint sensor 1114 , an optical sensor 1115 , and a proximity sensor 1116 .

The acceleration sensor 1111 can detect the magnitude of acceleration on the three coordinate axes of the coordinate system established by the terminal 1100 . For example, the acceleration sensor 1111 can be used to detect the components of the gravitational acceleration on the three coordinate axes. The processor 1101 can control the display screen 1105 to display the user interface in a landscape view or a portrait view according to the gravitational acceleration signal collected by the acceleration sensor 1111 . The acceleration sensor 1111 can also be used for game or user movement data collection.

The gyroscope sensor 1112 can detect the body direction and rotation angle of the terminal 1100 , and the gyroscope sensor 1112 can cooperate with the acceleration sensor 1111 to collect 3D actions of the user on the terminal 1100 . The processor 1101 can implement the following functions according to the data collected by the gyro sensor 1112: motion sensing (such as changing the UI according to the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

The pressure sensor 1113 may be disposed on the side frame of the terminal 1100 and/or the lower layer of the display screen 1105 . When the pressure sensor 1113 is disposed on the side frame of the terminal 1100, the user's holding signal of the terminal 1100 can be detected, and the processor 1101 can perform left and right hand identification or shortcut operations according to the holding signal collected by the pressure sensor 1113. When the pressure sensor 1113 is disposed on the lower layer of the display screen 1105, the processor 1101 controls the operability controls on the UI interface according to the user's pressure operation on the display screen 1105. The operability controls include at least one of button controls, scroll bar controls, icon controls, and menu controls.

The fingerprint sensor 1114 is used to collect the user's fingerprint, and the processor 1101 identifies the user's identity according to the fingerprint collected by the fingerprint sensor 1114, or the fingerprint sensor 1114 identifies the user's identity according to the collected fingerprint. When the user's identity is identified as a trusted identity, the processor 1101 authorizes the user to perform relevant sensitive operations, including unlocking the screen, viewing encrypted information, downloading software, making payments, and changing settings. The fingerprint sensor 1114 may be disposed on the front, back or side of the terminal 1100 . When the terminal 1100 is provided with physical buttons or a manufacturer's logo, the fingerprint sensor 1114 may be integrated with the physical buttons or the manufacturer's logo.

Optical sensor 1115 is used to collect ambient light intensity. In one embodiment, the processor 1101 can control the display brightness of the display screen 1105 according to the ambient light intensity collected by the optical sensor 1115 . Specifically, when the ambient light intensity is high, the display brightness of the display screen 1105 is increased; when the ambient light intensity is low, the display brightness of the display screen 1105 is decreased. In another embodiment, the processor 1101 can also dynamically adjust the shooting parameters of the camera assembly 1106 according to the ambient light intensity collected by the optical sensor 1115 .

A proximity sensor 1116 , also called a distance sensor, is provided on the front panel of the terminal 1100 . The proximity sensor 1116 is used to collect the distance between the user and the front of the terminal 1100 . In one embodiment, when the proximity sensor 1116 detects that the distance between the user and the front of the terminal 1100 is gradually decreasing, the processor 1101 controls the display screen 1105 to switch from the bright screen state to the off screen state; when the proximity sensor 1116 detects When the distance between the user and the front of the terminal 1100 gradually increases, the processor 1101 controls the display screen 1105 to switch from the screen-off state to the screen-on state.

Those skilled in the art can understand that the structure shown in FIG. 11 does not constitute a limitation on the terminal 1100, and may include more or less components than the one shown, or combine some components, or adopt different component arrangements.

Optionally, the computer device is provided as a server. FIG. 12 is a schematic structural diagram of a server provided by an embodiment of the present application. The server 1200 may vary greatly due to different configurations or performance, and may include one or more processors (Central Processing Units, CPU) 1201 and a Or more than one memory 1202, wherein, at least one computer program is stored in the memory 1202, and the at least one computer program is loaded and executed by the processor 1201 to implement the methods provided by the above method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input/output interface for input and output, and the server may also include other components for implementing device functions, which will not be described here.

Embodiments of the present application further provide a computer-readable storage medium, where at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to implement the text mapping model of the foregoing embodiment. The action performed by the handler method.

Embodiments of the present application further provide a computer program product or computer program, and provide a computer program product, including a computer program, when the computer program is executed by a processor, the processing method for implementing the text mapping model in the above-mentioned embodiments is executed. operate.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above embodiments can be completed by hardware, or can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium can be read-only memory, magnetic disk or optical disk, etc.

The above descriptions are only optional embodiments of the embodiments of the present application, and are not intended to limit the embodiments of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the embodiments of the present application shall be Included within the scope of protection of this application.

Claims (16)

1. A processing method for a text mapping model, wherein the method comprises: Obtain sample text information and first label information, where the first label information is text information whose similarity with the sample text information is not less than a similarity threshold; Based on the text mapping model, the sample text information is mapped to obtain the predicted text information; Based on a first similarity between the predicted text information and the sample text information and a second similarity between the first label information and the sample text information, second label information is determined, and the second label information is determined. The label information is the text information with a larger similarity between the predicted text information and the first label information and the sample text information; Based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information, the text mapping model is trained, and the text mapping model is used to map any text information. Similar text information. 2 . The method according to claim 1 , wherein the method based on the first similarity between the predicted text information and the sample text information and the difference between the first label information and the sample text information. 3 . before determining the second label information, the method further includes: obtaining a third degree of similarity and a fourth degree of similarity between the predicted text information and the sample text information, where the third degree of similarity indicates the difference between the predicted text information and the words contained in the sample text information, The fourth degree of similarity indicates semantic similarity between the predicted text information and the sample text information; The third similarity and the fourth similarity are weighted and fused to obtain the first similarity. 3. The method according to claim 2, wherein the obtaining the third similarity between the predicted text information and the sample text information comprises: Based on the number of at least one character, the predicted text information is divided respectively to obtain at least one first word set, and the words belonging to the same first word set contain the same number of characters; Based on the number of at least one of the characters, the sample text information is divided respectively to obtain at least one second word set, and the words belonging to the same second word set contain the same number of characters; determining a first number and a second number, the first number indicating the sum of the number of different words in the first set of words and the second set of words corresponding to each of the number of characters, the second number indicating the total number of words in at least one of the first set of words and at least one of the second set of words; The ratio of the first number to the second number is determined as the third similarity between the predicted text information and the sample text information. 4. The method according to claim 2, wherein obtaining the fourth similarity between the predicted text information and the sample text information comprises: Semantically extracting the predicted text information and the sample text information, respectively, to obtain a first semantic feature of the predicted text information and a second semantic feature of the sample text information; The similarity between the first semantic feature and the second semantic feature is determined as the fourth similarity. 5. The method according to claim 2, wherein acquiring the fourth similarity between the predicted text information and the sample text information comprises: splicing the predicted text information and the sample text information to obtain spliced text information; performing semantic extraction on the spliced text information to obtain a third semantic feature corresponding to the spliced text information; classifying the third semantic feature to obtain a classification result; The classification result is determined as the fourth similarity. 6 . The method according to claim 1 , wherein the text mapping is performed based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information. 7 . The model is trained, including: obtaining a first loss value corresponding to the predicted text information based on the second label information and the predicted text information; The text mapping model is trained based on the first similarity and the first loss value corresponding to the predicted text information. 7. The method according to claim 6, wherein the training of the text mapping model based on the first similarity and the first loss value corresponding to the predicted text information comprises: Obtain a weight parameter corresponding to each of the predicted text information based on the difference between the target similarity and the first similarity corresponding to each of the predicted text information; Based on the weight parameter corresponding to each of the predicted text information, a weighted average of the first loss values corresponding to a plurality of the predicted text information is performed to obtain a second loss value; determining the average value of the first loss values corresponding to the plurality of predicted text information as the third loss value; The text mapping model is trained based on the second loss value and the third loss value. 8. The method according to claim 1, wherein the acquiring the sample text information and the first label information comprises: obtaining a first text information set, where the first text information set includes a plurality of first text information and label information corresponding to each of the first text information; determining the similarity between each of the first text information and the corresponding label information; Based on the similarity corresponding to each of the first text information, the sample text information and the first label information whose similarity is greater than the similarity threshold are selected from the first text information set. 9 . The method according to claim 8 , wherein, based on the similarity corresponding to each of the first text information, filtering out the similarity greater than the similarity threshold from the first text information set. 10 . The sample text information and the first label information of , including: Based on the similarity corresponding to each of the first text information, at least one second text information whose similarity is greater than the similarity threshold is selected from the first text information set, and the filtered second text is The information and the corresponding label information constitute a second text information set; The sample text information and the first label information are acquired from the second text information set, and the sample text information is any one of the second text information. 10. The method according to claim 8, wherein, before the sample text information is mapped based on the text mapping model to obtain predicted text information, the method further comprises: The text mapping model is trained based on the first text information and corresponding label information in the first text information set. 11. The method according to any one of claims 1-10, wherein the method is based on a first similarity between the predicted text information and the sample text information and the first label information and the The second similarity between the sample text information is determined, and the second label information is determined, including: In response to that the largest first similarity among the plurality of first similarities is greater than the second similarity, determining the predicted text information corresponding to the largest first similarity as the second label information , a plurality of the first degrees of similarity are the degrees of similarity between a plurality of the predicted text information and the sample text information; or, In response to a plurality of the first degrees of similarity being not greater than the second degrees of similarity, the first label information is determined as the second label information. 12. The method according to any one of claims 1-10, wherein, based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information, After training the text mapping model, the method further includes: Based on the text mapping model, the target text information is mapped to obtain similar text information of the target text information. 13. An apparatus for processing a text mapping model, wherein the apparatus comprises: an acquisition module, configured to acquire sample text information and first label information, where the first label information is text information whose similarity with the sample text information is not less than a similarity threshold; a mapping module, configured to map the sample text information based on the text mapping model to obtain predicted text information; A determination module, configured to determine second label information based on a first similarity between the predicted text information and the sample text information and a second similarity between the first label information and the sample text information , the second label information is the text information with a larger similarity between the predicted text information and the first label information and the sample text information; A training module, configured to train the text mapping model based on the second label information, the predicted text information, and the first similarity corresponding to the predicted text information, and the text mapping model is used for mapping Similar text messages out of any text message. 14. A computer device, characterized in that the computer device comprises a processor and a memory, the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the claim The operation performed by the processing method of the text mapping model according to any one of claims 1 to 12. 15. A computer-readable storage medium, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to implement any one of claims 1 to 12. An operation performed by the processing method of the text mapping model according to the claim. 16. A computer program product, comprising a computer program, characterized in that, when the computer program is executed by a processor, the operations performed by the method for processing a text mapping model according to any one of claims 1 to 12 are implemented.

Images