WO2020215683A1

WO2020215683A1 - Semantic recognition method and apparatus based on convolutional neural network, and non-volatile readable storage medium and computer device

Info

Publication number: WO2020215683A1
Application number: PCT/CN2019/117723
Authority: WO
Inventors: 金戈; 徐亮
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-04-26
Filing date: 2019-11-12
Publication date: 2020-10-29
Also published as: CN110222330A; CN110222330B

Abstract

Disclosed are a semantic recognition method and apparatus based on a convolutional neural network, and a non-volatile readable storage medium and a computer device, which relate to the technical field of text processing and can improve the efficiency of semantic recognition. The method comprises: using a first convolutional neural network preset in a semantic recognition model to acquire a text vector of a text to be recognized; using a second convolutional neural network preset in the semantic recognition model to determine, according to the acquired text vector, a named entity in said text; and using a third convolutional neural network preset in the semantic recognition model to determine, according to the acquired text vector and the determined named entity, an entity relationship in said text. The present application is applicable to intelligent question-answering of customer services in an insurance product business.

Description

Semantic recognition method and device based on convolutional neural network, non-volatile readable storage medium, and computer equipment

This application claims priority with the Chinese patent application filed on April 26, 2019 with the Chinese Patent Office, the application number is 201910345595.7, and the application title is "Semantic Recognition Method and Apparatus Based on Convolutional Neural Network, Storage Medium and Computer Equipment". The entire content is incorporated into the application by reference.

Technical field

This application relates to the technical field of processing herein, in particular to a method and device for semantic recognition based on convolutional neural networks, non-volatile readable storage media, and computer equipment.

Background technique

With the development of science and technology, there are more and more ways to identify the relationship between some words and words, and the applicable scenarios are also more and more extensive, such as the upper and lower relationships between some place names, the hierarchical relationship between state institutions, The inclusion relations of the types of items, etc., and these require independent recognition models to realize the recognition of words (namely entities) and the recognition of the relationships between words and words (ie, entity relationships).

The disadvantage of the prior art is that the two independent recognition models used to realize named entity recognition and entity relationship recognition are prone to information redundancy in the process of joint use. The current solution is only It is limited to partially combining the above two independent recognition models based on the cyclic neural network to increase the calculation rate of the network model, thereby improving the efficiency of named entity recognition and entity relationship recognition, but the improvement effect is weak.

Summary of the invention

In view of this, this application provides a semantic recognition method and device based on a convolutional neural network, a non-volatile readable storage medium, and computer equipment. The main purpose is to solve the existing two independent methods for named entity recognition and entity relationship recognition. In the process of joint use of the recognition models, information redundancy is easy to exist between each other, and the calculation rate of the adopted network model is low.

According to one aspect of the present application, there is provided a semantic recognition method based on a convolutional neural network, the method including:

Use the first convolutional neural network preset in the semantic recognition model to obtain the text vector of the text to be recognized;

Use the second convolutional neural network preset in the semantic recognition model to determine the named entity in the text to be recognized according to the obtained text vector;

The third convolutional neural network preset in the semantic recognition model is used to determine the entity relationship in the text to be recognized according to the obtained text vector and the determined named entity.

According to another aspect of the present application, there is provided a semantic recognition device based on a convolutional neural network, the device including:

The first convolutional neural network module is used to obtain the text vector of the text to be recognized by using the first convolutional neural network preset in the semantic recognition model;

The second convolutional neural network module is configured to use the second convolutional neural network preset in the semantic recognition model to determine the named entity in the text to be recognized according to the obtained text vector;

The third convolutional neural network module is used to use the preset third convolutional neural network in the semantic recognition model to determine the entity relationship in the text to be recognized according to the obtained text vector and the determined named entity.

According to another aspect of the present application, there is provided a non-volatile readable storage medium having computer readable instructions stored thereon, and the program is executed by a processor to realize the above-mentioned semantic recognition method based on convolutional neural network.

According to another aspect of the present application, a computer device is provided, including a non-volatile readable storage medium, a processor, and computer readable instructions stored on the non-volatile readable storage medium and running on the processor , When the processor executes the program, the above semantic recognition method based on the convolutional neural network is realized.

With the above technical solutions, the convolutional neural network-based semantic recognition method and device, non-volatile readable storage medium, and computer equipment provided in this application will be used for named entity recognition and entity relationship with existing recurrent neural networks. Compared with the technical solution in which the two independent recognition models for recognition are partially combined, this application uses the first convolutional neural network preset in the semantic recognition model to obtain the text vector of the text to be recognized, and uses the preset first convolutional neural network in the semantic recognition model. The second convolutional neural network determines the named entity in the text to be recognized according to the acquired text vector, and uses the third convolutional neural network preset in the semantic recognition model, according to the acquired text vector and the determined named entity, Determine the entity relationship in the text to be recognized. It can be seen that by using the multi-layer convolutional neural network in the semantic recognition model to realize the recognition of named entities and entity relations, it can effectively avoid the information redundancy problem caused by the existing two independent recognition models in the process of joint use. Improve the efficiency of semantic recognition.

The above description is only an overview of the technical solution of this application. In order to understand the technical means of this application more clearly, it can be implemented in accordance with the content of the specification, and to make the above and other purposes, features and advantages of this application more obvious and understandable. , The following specifically cite the specific implementation of this application.

Description of the drawings

The drawings described here are used to provide a further understanding of the application and constitute a part of the application. The exemplary embodiments and descriptions of the application are used to explain the application and do not constitute an improper limitation of the application. In the attached picture:

FIG. 1 shows a schematic flowchart of a semantic recognition method based on a convolutional neural network provided by an embodiment of the present application;

FIG. 2 shows a schematic flowchart of another semantic recognition method based on a convolutional neural network provided by an embodiment of the present application;

Fig. 3 shows a schematic structural diagram of a semantic recognition device based on a convolutional neural network provided by an embodiment of the present application.

Detailed ways

Hereinafter, the application will be described in detail with reference to the drawings and in conjunction with embodiments. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other if there is no conflict. For the existing two independent recognition models used to realize named entity recognition and entity relationship recognition, in the process of joint use, information redundancy is easy to exist between each other, and the calculation rate of the adopted network model is low. This embodiment provides a semantic recognition method based on a convolutional neural network, which can effectively avoid the information redundancy problem caused by the joint use of the existing two independent recognition models, thereby effectively improving the efficiency of semantic recognition, as shown in Figure 1 As shown, the method includes:

101. Obtain the text vector of the text to be recognized by using the first convolutional neural network preset in the semantic recognition model.

Obtain the text to be recognized, preprocess the text to be recognized to obtain the initialized text vector, and input the initialized text vector into the first convolutional neural network preset in the semantic recognition model to generate a characterization of the text to be recognized Text vector. Among them, the preprocessing can be specifically set according to the actual application scenario, for example, the preprocessing is set as word segmentation processing, that is, the text to be recognized is marked with words as the unit; or the preprocessing is set as word filtering processing, that is, After the word segmentation is performed on the text to be recognized, unimportant words are eliminated, such as auxiliary verbs such as "can, should", and unimportant words such as interjections such as "oh, ah", to improve the semantic recognition of the text to be recognized The efficiency is not specifically limited here.

Among them, taking the preprocessing as the word segmentation processing as an example, the specific word segmentation processing of the text to be recognized is to use the SBME notation method to mark the words in the text to be recognized, that is, to mark the word as S and the beginning of the word as B. The middle part of the word is marked as M, the end of the word is marked as E, and the initial text vector is generated according to the marked text to be recognized.

Before performing semantic recognition on the text to be recognized, construct the semantic recognition model of this application, and obtain the training sample set used to train the semantic recognition model, that is, the training sample set can be used to train the initialized first convolutional neural network, second Convolutional neural network and the third convolutional neural network, thereby obtaining a semantic recognition model. The training sample set includes multiple phrase corpora. The phrase corpus is in a short sentence format, that is, a short sentence is divided into a comma. Specifically, each phrase corpus includes two interrelated words, for example, "China, Shanghai" , And mark the relationship between the two words in each phrase corpus, for example, mark the relationship between the words "China, Shanghai" as the upper and lower relationship to construct a training sample set. In addition, you can also mark the corresponding word attribute for each word in the above multiple phrase corpus, for example, mark China and Shanghai in "China, Shanghai" as place names, or canine in "Canine, Dog" And dogs are marked as animals. In the actual application process, the relationship between two words in the phrase corpus can be set in various ways. For example, mark the relationship between "Copyright Office and Trademark Office" as a parallel relationship, and mark "Copyright Office, Trademark Office" in The term attribute of the Copyright Office and the Trademark Office of is a national institution; the relationship between the mark "Canine family, dog" is an inclusive relationship, and the word attribute of canine family and dog in the mark "Canine family, dog" is animal, etc., here There is no specific limitation on the mutual relationship.

102. Use the second convolutional neural network preset in the semantic recognition model to determine the named entity in the text to be recognized according to the obtained text vector.

The preset second convolutional neural network is used to identify the named entities contained in the text to be recognized, the output result of the preset first convolutional neural network is used as the input of the preset second convolutional neural network, and the preset is input The output of the second convolutional neural network is the named entity contained in the text to be recognized.

Among them, there can be multiple named entities contained in the text to be recognized, that is, the text to be recognized includes multiple words, and a named entity or named entity category is output for each word. The named entity category includes person name, place name, organization name, Named entity categories such as product names and proper nouns.

103. Using the third convolutional neural network preset in the semantic recognition model, determine the entity relationship in the text to be recognized according to the obtained text vector and the determined named entity.

The preset third convolutional neural network is used to identify the entity relationship contained in the text to be recognized, and the preset output results of the first convolutional neural network and the preset second convolutional neural network are used as presets The input of the third convolutional neural network is input to the preset third convolutional neural network, and the output result is the entity relationship between the named entities contained in the text to be recognized.

In practical applications, since the text to be recognized is in a short sentence format, the number of named entities obtained after being processed by the preset second convolutional neural network will not be large. Therefore, the real-name entities that are limited to a few types of The entity relationship obtained after processing by the third convolutional neural network will be relatively certain and accurate. For example, if the preset number of named entities output by the second convolutional neural network is two or three, then according to the output two or three named entities, use the preset third convolutional neural network to output two or three The entity relationship between two named entities, because the text to be recognized is in short sentence format, and the preset third convolutional neural network is only used for the recognition of the relationship between a small number of named entities, so that the recognition efficiency of the text to be recognized is obtained Significant improvement.

For this embodiment, according to the above scheme, the acquired text to be recognized can be hierarchically recognized according to the constructed semantic recognition model, and different convolutional neural networks in the semantic recognition model can be used to realize named entities and entity relationships in the text to be recognized. Compared with the existing technical solution based on the recurrent neural network that partially combines two independent recognition models for named entity recognition and entity relationship recognition, this embodiment can not only improve the recognition efficiency of the text to be recognized, Avoid the information redundancy problem caused by the joint use of the existing two independent recognition models. At the same time, the application scenarios of this embodiment are more broad, that is, it can be simultaneously applied to the recognition of named entities alone, the recognition of entity relationships alone, and At the same time, for the application scenarios of named entity and entity relationship recognition, there is no need to build different semantic recognition models for different needs. The maintenance and optimization of the later model reduce the cost, and it does not affect the model at all while reducing the cost. Semantic recognition efficiency and semantic recognition accuracy.

Further, as a refinement and extension of the specific implementation of the above-mentioned embodiment, in order to fully explain the specific implementation process of this embodiment, another semantic recognition method based on convolutional neural network is provided. As shown in FIG. 2, this method include:

201. Determine the first loss function, the second loss function, and the third loss function according to the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network.

Since the second convolutional neural network is used to identify named entities, and the third convolutional neural network is used to identify entity relationships, the loss functions of the second and third convolutional neural networks are constructed based on cross entropy. The second The loss function of the convolutional neural network is the cross entropy used to identify named entities, and the loss function of the third convolutional neural network is the cross entropy used to identify the relationship.

According to the needs of actual application scenarios, the first loss function, the second loss function, and the third loss function can be set differently according to the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network. For the loss function, the same loss function can also be used. The first loss function, the second loss function, and the third loss function are not specifically set here. In this embodiment, the first loss function and the second loss function are set Same as the third loss function, the calculation formula is:

Among them, x is the data sample in the sample set of the first convolutional neural network, the second convolutional neural network and the third convolutional neural network used for training initialization, p and q are the true probability distributions of the sample set respectively, not true Probability distributions.

202. According to the determined first loss function, second loss function, and third loss function, train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain a preset The first convolutional neural network, the second convolutional neural network and the third convolutional neural network.

To illustrate the specific implementation of step 202, as a preferred embodiment, step 202 may specifically include: determining the loss function of the semantic recognition model according to the determined first loss function, second loss function, and third loss function ; Use the loss function of the semantic recognition model to train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain the preset first convolutional neural network and second convolutional neural network Network and third convolutional neural network.

For example, according to the needs of actual application scenarios, the determined first loss function, second loss function, and third loss function are added and averaged to obtain the loss function of the semantic recognition model. Further, if the actual application scenario is If the number of named entities in the text to be recognized is large, the weight value of the second loss function should be increased accordingly. If there are more entity relationships in the text to be recognized in the actual application scenario, the weight value of the third loss function should be increased accordingly. There is no specific limitation on the calculation method of the loss function of the semantic recognition model.

In the training process of the initialized first convolutional neural network, second convolutional neural network and third convolutional neural network, according to the determined loss function, the convex optimization algorithm is used to automatically update the network parameters in the hidden layer of the neural network, Thus, the preset first convolutional neural network, second convolutional neural network, and third convolutional neural network are obtained. Among them, convex optimization algorithm, also known as convex optimization algorithm, or convex minimization algorithm, is a sub-field of mathematical optimization, which uses the idea of local optimal value that is global optimal value to update the network parameters in the hidden layer of neural network .

The adaptive moment estimation (Adam: Adaptive Moment Estimation) optimization algorithm is a first-order optimization algorithm that can replace the traditional stochastic gradient descent process. According to the training sample set of this application, the Adam optimization algorithm is used to update the network parameters in the hidden layer of the neural network . In Python's tensorflow library, the loss function of the semantic recognition model is optimized by convex function. Specifically, the loss function is minimized as the goal, and the Adam optimization algorithm is used to iteratively update the network parameters in the semantic recognition model to obtain the preset The first convolutional neural network, the second convolutional neural network and the third convolutional neural network. Among them, this application specifically limits the number of convolutional layers in the semantic recognition model.

In practical applications, the specific training process is to compare the named entity recognition result output by the second convolutional neural network with the named entity or labeled word attributes in the training sample set. If the comparison results are inconsistent, it means that the recognition is wrong; and , According to the named entity recognition result output by the second convolutional neural network, compare the entity relationship recognition result output by the third convolutional neural network with the entity relationship marked by the corresponding output named entity recognition result in the training sample set. If the comparison results are inconsistent , It means the recognition error. The loss function of the semantic recognition model is used to correct the erroneous recognition results, and then complete the training of the semantic recognition model, and obtain a semantic recognition model capable of simultaneously performing named entity recognition and entity relationship recognition.

203. Use the word vector dictionary to obtain the word vector and the word vector of the text to be recognized.

The initialized text vector is obtained by performing word segmentation processing on the acquired text to be recognized, and the initialized text vector is used as the input of the first convolutional neural network preset in the semantic recognition model. The preset embedding layer of the first convolutional neural network uses a preset word vector dictionary to convert the initialized text vector into a word vector and a word vector for representing the text to be recognized. Among them, the preset word vector dictionary contains the word vector corresponding to each word in the initialized text vector and the word vector corresponding to each word.

204. Perform a convolution operation on the obtained word vector and word vector to obtain a text vector of the text to be recognized.

The preset first convolutional neural network includes a double-layer one-dimensional full convolution structure, and the word vectors and word vectors from the embedding layer pass through the double-layer one-dimensional full convolution structure to output the text vector of the text to be recognized. Specifically, the convolution kernel is used to perform a convolution operation (ie, dot multiplication) with the word vector and word vector of the text to be recognized, and all the obtained convolution operation results are used as the text vector of the text to be recognized.

For example, set the length of the convolution kernel to 3, that is, use the convolution kernel with dimension 3 to perform convolution operation with the word vector and word vector of the text to be recognized, and use the text vector of the text to be recognized as the preset The input of the second convolutional neural network and the preset third convolutional neural network.

Among them, the preset first convolutional neural network is a shared network structure of the preset second convolutional neural network and the preset third convolutional neural network, thereby realizing the preset second convolutional neural network and the preset The sharing of the underlying parameters in the third convolutional neural network effectively avoids the information redundancy problem caused by the joint use of the existing two independent recognition models, and further improves the efficiency of semantic recognition.

205. Perform a convolution operation on the obtained text vector to obtain a named entity to be determined.

The second convolutional neural network preset in the semantic recognition model is used to perform named entity recognition (NER: Named Entity Recognition) on the obtained text vector to obtain the named entity to be determined. Among them, named entity recognition is also called "proprietary name recognition", which refers to the recognition of entities with specific meaning in the text to be recognized.

The preset second convolutional neural network is a dense connection structure DenseNet. The dense connection structure has a large number of dense connections, which can maximize the information flow between all layers in the neural network. By connecting all layers in the neural network in pairs, The input of each layer of the neural network is the union of the output of all the previous layers, and the feature map output by this layer will also be directly passed to all subsequent layers as input, so as to realize the repeated use of features and reduce redundancy.

In addition, the preset second convolutional neural network includes a two-layer convolution structure, and further convolution operations are performed on the convolution operation result output by the preset first convolutional neural network in the semantic recognition model based on the two-layer convolution structure. , Get the named entity to be determined.

Among them, the preset convolution structure in the second convolutional neural network is a one-dimensional full convolution structure, and the one-dimensional full convolution structure can maintain the same length as the convolution operation result output through it, that is, based on the one-dimensional full convolution The product structure makes the convolution operation result output by the preset first convolutional neural network and the convolution operation result output through the one-dimensional full convolution structure a sequence of equal length.

206. Perform boundary character recognition on the named entity to be determined, and determine the final named entity according to the recognition result.

To illustrate the specific implementation of step 206, as a preferred embodiment, step 206 may specifically include: if the boundary character recognition result of the named entity to be determined is consistent with the preset boundary character recognition result, determining the The named entity is the final named entity; if the boundary character recognition result of the to-be-determined named entity is inconsistent with the preset boundary character recognition result, the to-be-determined named entity is used as a new training sample of the semantic recognition model.

Use the second convolutional neural network preset in the semantic recognition model to perform boundary character recognition according to the obtained SBME mark in the named entity to be determined. Specifically, if the mark in the obtained named entity to be determined is S, that is If the named entity to be determined is a single character, then the single character is recognized; if the recognition result is consistent with the preset boundary character recognition result, the single character is determined to be the final named entity. For example, if the recognized named entity to be determined is "cat", and the recognition result is consistent with the preset boundary character recognition result, then the final recognized named entity is a cat. If the recognition result is inconsistent with the preset boundary character recognition result, it means that the word is not a named entity. For example, if the named entity to be identified is "How", and the recognition result is inconsistent with the preset boundary character recognition result, then "How" is used as a new training sample of the semantic recognition model to further optimize the semantic recognition model. To improve the recognition accuracy of the semantic recognition model.

If the obtained tag in the to-be-determined named entity includes BME or BE, that is, the to-be-determined named entity is multi-character or double-character, then the named entity is recognized according to the tag B and the tag E; if the recognition result is consistent with the preset If the boundary character recognition results are consistent, it is determined that the named entity to be determined is the final named entity. For example, if the mark in the named entity to be determined includes BME, it is recognized that the mark B and mark E in the named entity to be determined correspond to "pre" and "home", and the recognition result is consistent with the preset boundary character recognition result, then Identify the final named entity as a prophet; if the mark in the named entity to be determined includes BE, identify the mark B and mark E in the named entity to be determined corresponding to "work" and "home", and the recognition result is the same as the preset If the boundary character recognition results are consistent, the final named entity is recognized as the writer. If the recognition result is inconsistent with the preset boundary character recognition result, it means that the multi-character or double-character is not a named entity. It may be because more or less characters are recognized, and the recognition result is not a named entity. For example, if the named entity to be identified is "writer", and the recognition result is inconsistent with the preset boundary character recognition result, then "writer" is used as a new training sample of the semantic recognition model, and the semantic recognition model is further refined. Optimized to improve the recognition accuracy of the semantic recognition model.

In practical applications, the preset boundary character recognition result can be the single character of the named entity, and the head and tail of the double character and multi-character, or the word attribute of the word mark in the training sample set, that is, the word attribute of the word. As well as the beginning and end of double-word and multi-word.

In practical applications, the text to be recognized can include one or more named entities. Therefore, according to the text vector of the text to be recognized, the preset activation function softmax in the second convolutional neural network is used to output one or more named entities. The recognition result, that is, the output result corresponds to one or more named entities included in the text to be recognized. Specifically, the second convolutional neural network also includes an activation function softmax. Based on the activation function softmax, the calculation result (ie, the named entity to be determined) obtained through the two-layer convolution structure in the second convolutional neural network is further classified Operate to get the final named entity.

207. Using the third convolutional neural network preset in the semantic recognition model, determine the entity relationship in the text to be recognized according to the obtained text vector and the determined named entity.

The preset third convolutional neural network is a densely connected structure DenseNet. On the basis of the preset first convolutional neural network, a convolutional layer and a pooling layer are constructed, and are fully connected through the activation function softmax The layer outputs the recognition result, and the output result is a multi-classification variable, that is, one or more entity relationships included in the text to be recognized are determined according to the probability values of different classifications.

In practical applications, according to the determined named entity, the corresponding relationship between the named entity in the training sample set and the marked entity relationship is used to determine the entity relationship, and the identified entity relationship is compared with the determined entity relationship. If the recognition results are consistent , The recognized entity relationship is the entity relationship in the text to be recognized; if the recognition results are inconsistent, it means that the recognition is wrong, and the wrong recognition result is adjusted to the corresponding relationship between the named entity in the training sample set and the marked entity relationship. The entity relationship of as the entity relationship in the text to be recognized. Among them, the wrong recognition result is used as a new training sample to realize the training of the semantic recognition model and obtain an optimized semantic recognition model.

208. Use the newly-added training sample to train the semantic recognition model to obtain an optimized semantic recognition model.

When the named entity to be determined is not the final named entity, the named entity to be determined is added to the training sample set for training the semantic recognition model as a new phrase corpus, and the name to be determined is marked in the phrase corpus The word attribute of the entity is a recognition error, so that the semantic recognition model can effectively improve the recognition accuracy of the text to be recognized after optimization training.

By applying the technical solution of this embodiment, the first convolutional neural network preset in the semantic recognition model is used to obtain the text vector of the text to be recognized, and the second convolutional neural network preset in the semantic recognition model is used, according to the obtained The text vector determines the named entity in the text to be recognized, and the third convolutional neural network preset in the semantic recognition model is used to determine the entity relationship in the text to be recognized according to the obtained text vector and the determined named entity. Compared with the existing technical solution based on the recurrent neural network in which two independent recognition models for named entity recognition and entity relationship recognition are partially combined, when this embodiment is applied to customer service intelligent question answering in insurance product business, the system can According to the sentence input by the user, the semantic recognition model is used to realize accurate and rapid recognition of the sentence, thereby providing users with more accurate services and improving user experience.

Further, as a specific implementation of the method in FIG. 1, an embodiment of the present application provides a semantic recognition device based on a convolutional neural network. As shown in FIG. 3, the device includes: a first convolutional neural network module 31, a second The convolutional neural network module 32 and the third convolutional neural network module 33.

The first convolutional neural network module 31 can be used to obtain the text vector of the text to be recognized by using the preset first convolutional neural network in the semantic recognition model; the first convolutional neural network module 31 recognizes the text to be recognized for the device The basic module of named entities and entity relationships in the.

The second convolutional neural network module 32 may be used to use the second convolutional neural network preset in the semantic recognition model to determine the named entity in the text to be recognized according to the text vector obtained by the first convolutional neural network module 31; The second convolutional neural network module 32 is the main functional module of the device for identifying named entities in the text to be recognized, and is also the core functional module of the device.

The third convolutional neural network module 33 can be used to use the preset third convolutional neural network in the semantic recognition model, according to the text vector obtained by the first convolutional neural network module 31 and the second convolutional neural network module 32 The determined named entity determines the entity relationship in the text to be recognized; the third convolutional neural network module 33 is the main functional module of the device that recognizes the entity relationship in the text to be recognized, and is also the core functional module of the device.

In specific application scenarios, the first convolutional neural network module 31 can be specifically used to obtain the word vectors and word vectors of the text to be recognized by using the word vector dictionary; perform convolution operations on the obtained word vectors and word vectors to obtain The text vector of the text to be recognized.

In a specific application scenario, it also includes a training module 34, which can be used to determine the first loss function and the second loss according to the initialized first, second, and third convolutional neural networks. Function and third loss function; according to the determined first loss function, second loss function and third loss function, perform the initialization of the first convolutional neural network, the second convolutional neural network and the third convolutional neural network Train to obtain the preset first convolutional neural network, second convolutional neural network and third convolutional neural network.

In a specific application scenario, the training module 34 may be specifically used to determine the loss function of the semantic recognition model according to the determined first loss function, second loss function, and third loss function; use the semantic recognition model The loss function of training initializes the first convolutional neural network, the second convolutional neural network, and the third convolutional neural network to obtain the preset first, second, and third convolutional neural networks The internet.

In a specific application scenario, the second convolutional neural network module 32 can be specifically used to perform convolution operations on the acquired text vector to obtain the named entity to be determined; perform boundary character recognition on the named entity to be determined, according to the recognition The result determines the final named entity.

In a specific application scenario, the second convolutional neural network module 32 can be specifically used to determine the named entity to be determined if the boundary character recognition result of the named entity to be determined is consistent with the preset boundary character recognition result Is the final named entity; if the boundary character recognition result of the to-be-determined named entity is inconsistent with the preset boundary character recognition result, the to-be-determined named entity is used as a new training sample of the semantic recognition model.

In a specific application scenario, the training module 34 may be specifically used to train the semantic recognition model using the newly added training samples to obtain an optimized semantic recognition model. It should be noted that, for other corresponding descriptions of the functional units involved in the convolutional neural network-based semantic recognition device provided by the embodiment of the present application, reference may be made to the corresponding descriptions in FIG. 1 and FIG. 2, and details are not repeated here.

Based on the above-mentioned method shown in Figure 1 and Figure 2, correspondingly, an embodiment of the present application also provides a non-volatile readable storage medium on which computer readable instructions are stored, and the program is executed when the processor is executed. The above-mentioned semantic recognition method based on convolutional neural network as shown in Fig. 1 and Fig. 2. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, and the software product can be stored in a non-volatile non-volatile readable storage medium (can be CD-ROM, U disk, mobile hard disk) Etc.), including several instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in each implementation scenario of this application.

Based on the methods shown in Figures 1 and 2 and the virtual device embodiment shown in Figure 3, in order to achieve the above objectives, the embodiments of the present application also provide a computer device, which can be a personal computer, a server, or a network. The physical device includes a non-volatile readable storage medium and a processor; the non-volatile readable storage medium is used to store computer readable instructions; and the processor is used to execute computer readable instructions to achieve the above Figure 1 and Figure 2 show the semantic recognition method based on convolutional neural network.

Optionally, the computer device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, and so on. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc., and the optional user interface may also include a USB interface, a card reader interface, and the like. The network interface can optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc. Those skilled in the art can understand that the structure of a computer device provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or arrange different components.

The non-volatile readable storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of computer equipment, and supports the operation of information processing programs and other software and/or programs. The network communication module is used to implement communication between various components in the non-volatile readable storage medium and communication with other hardware and software in the physical device. Through the description of the foregoing implementation manners, those skilled in the art can clearly understand that this application can be implemented by means of software plus a necessary general hardware platform, or by hardware. By applying the technical solution of the present application, compared with the existing technical solution based on the recurrent neural network that partially combines two independent recognition models for named entity recognition and entity relationship recognition, this embodiment can effectively avoid the existing two This independent recognition model causes information redundancy in the process of joint use, thereby effectively improving the efficiency of semantic recognition.

Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of preferred implementation scenarios, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing this application. Those skilled in the art can understand that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the description of the implementation scenario, or can be changed to be located in one or more devices different from the implementation scenario. The modules of the above implementation scenarios can be combined into one module or further divided into multiple sub-modules.

The above serial number of this application is only for description and does not represent the pros and cons of implementation scenarios. The above disclosures are only a few specific implementation scenarios of the application, but the application is not limited to these, and any changes that can be thought of by those skilled in the art should fall into the protection scope of the application.

Claims

A semantic recognition method based on convolutional neural network, which is characterized in that it includes:

Determine the first loss function, the second loss function, and the third loss function according to the initialized first, second, and third convolutional neural networks;

According to the determined first loss function, second loss function, and third loss function, train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain the preset first One convolutional neural network, second convolutional neural network and third convolutional neural network;

Use the first convolutional neural network preset in the semantic recognition model to obtain the text vector of the text to be recognized;

Use the second convolutional neural network preset in the semantic recognition model to determine the named entity in the text to be recognized according to the obtained text vector;

Use the preset third convolutional neural network in the semantic recognition model to determine the entity relationship in the text to be recognized based on the obtained text vector and the determined named entity;

Among them, using the first convolutional neural network preset in the semantic recognition model to obtain the text vector of the text to be recognized includes:

Use the word vector dictionary to obtain the word vector and word vector of the text to be recognized;

Perform a convolution operation on the obtained word vector and word vector to obtain the text vector of the text to be recognized.
The method according to claim 1, wherein according to the determined first loss function, second loss function and third loss function, the initialized first convolutional neural network, second convolutional neural network and first Three convolutional neural networks are trained to obtain the preset first, second, and third convolutional neural networks, which specifically include:

Determine the loss function of the semantic recognition model according to the determined first loss function, second loss function, and third loss function;

Use the loss function of the semantic recognition model to train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain the preset first convolutional neural network and second convolutional neural network And the third convolutional neural network.
The method according to claim 1, wherein the second convolutional neural network preset in the semantic recognition model is used to determine the named entity in the text to be recognized according to the obtained text vector, which specifically includes:

Perform a convolution operation on the obtained text vector to obtain the named entity to be determined;

Perform boundary character recognition on the named entity to be determined, and determine the final named entity according to the recognition result.
The method according to claim 3, wherein, performing boundary character recognition on the named entity to be determined, and determining the final named entity according to the recognition result, specifically includes:

If the boundary character recognition result of the to-be-determined named entity is consistent with the preset boundary character recognition result, determining that the to-be-determined named entity is the final named entity;

If the boundary character recognition result of the to-be-determined named entity is inconsistent with the preset boundary character recognition result, the to-be-determined named entity is used as a new training sample of the semantic recognition model.
The method according to claim 4, further comprising:

The semantic recognition model is trained by using the newly added training samples to obtain an optimized semantic recognition model.
A semantic recognition device based on convolutional neural network, which is characterized in that it comprises:

The training module is used to determine the first loss function, the second loss function and the third loss function according to the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network; Determine the first loss function, second loss function, and third loss function, train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to get the preset first volume Convolutional neural network, second convolutional neural network and third convolutional neural network;

The first convolutional neural network module is used to obtain the text vector of the text to be recognized by using the first convolutional neural network preset in the semantic recognition model;

The second convolutional neural network module is configured to use the second convolutional neural network preset in the semantic recognition model to determine the named entity in the text to be recognized according to the obtained text vector;

The third convolutional neural network module is used to use the preset third convolutional neural network in the semantic recognition model to determine the entity relationship in the text to be recognized according to the obtained text vector and the determined named entity;

Wherein, the first convolutional neural network module specifically includes:

Use the word vector dictionary to obtain the word vector and word vector of the text to be recognized;

Perform a convolution operation on the obtained word vector and word vector to obtain the text vector of the text to be recognized.
7. The device according to claim 6, wherein the first convolutional neural network, the second convolutional neural network, and the first convolutional neural network are initialized according to the determined first loss function, second loss function, and third loss function. Three convolutional neural networks are trained to obtain the preset first, second, and third convolutional neural networks, which specifically include:

Determine the loss function of the semantic recognition model according to the determined first loss function, second loss function, and third loss function;

Use the loss function of the semantic recognition model to train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain the preset first convolutional neural network and second convolutional neural network And the third convolutional neural network.
The device according to claim 6, wherein the second convolutional neural network module specifically comprises:

Perform a convolution operation on the obtained text vector to obtain the named entity to be determined;

Perform boundary character recognition on the named entity to be determined, and determine the final named entity according to the recognition result.
The device according to claim 8, characterized in that, performing boundary character recognition on the named entity to be determined, and determining the final named entity according to the recognition result, specifically includes:

If the boundary character recognition result of the to-be-determined named entity is consistent with the preset boundary character recognition result, determining that the to-be-determined named entity is the final named entity;

If the boundary character recognition result of the to-be-determined named entity is inconsistent with the preset boundary character recognition result, the to-be-determined named entity is used as a new training sample of the semantic recognition model.
The device according to claim 9, characterized in that it specifically further comprises:

The semantic recognition model is trained by using the newly added training samples to obtain an optimized semantic recognition model.
A non-volatile readable storage medium having computer readable instructions stored thereon, characterized in that, when the program is executed by a processor, a method for semantic recognition based on a convolutional neural network is realized, including:

Determine the first loss function, the second loss function, and the third loss function according to the initialized first, second, and third convolutional neural networks;

According to the determined first loss function, second loss function, and third loss function, train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain the preset first One convolutional neural network, second convolutional neural network and third convolutional neural network;

Use the first convolutional neural network preset in the semantic recognition model to obtain the text vector of the text to be recognized;

Use the second convolutional neural network preset in the semantic recognition model to determine the named entity in the text to be recognized according to the obtained text vector;

Use the preset third convolutional neural network in the semantic recognition model to determine the entity relationship in the text to be recognized based on the obtained text vector and the determined named entity;

Among them, using the first convolutional neural network preset in the semantic recognition model to obtain the text vector of the text to be recognized includes:

Use the word vector dictionary to obtain the word vector and word vector of the text to be recognized;

Perform a convolution operation on the obtained word vector and word vector to obtain the text vector of the text to be recognized.
The non-volatile readable storage medium according to claim 11, wherein according to the determined first loss function, second loss function, and third loss function, the initialized first convolutional neural network, second The second convolutional neural network and the third convolutional neural network are trained to obtain the preset first, second, and third convolutional neural networks, which specifically include:

Determine the loss function of the semantic recognition model according to the determined first loss function, second loss function, and third loss function;

Use the loss function of the semantic recognition model to train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain the preset first convolutional neural network and second convolutional neural network And the third convolutional neural network.
The non-volatile readable storage medium according to claim 11, wherein the second convolutional neural network preset in the semantic recognition model is used to determine the named entity in the text to be recognized according to the obtained text vector, Specifically:

Perform a convolution operation on the obtained text vector to obtain the named entity to be determined;

Perform boundary character recognition on the named entity to be determined, and determine the final named entity according to the recognition result.
The non-volatile readable storage medium according to claim 13, wherein performing boundary character recognition on the named entity to be determined, and determining the final named entity according to the recognition result, specifically includes:

If the boundary character recognition result of the to-be-determined named entity is consistent with the preset boundary character recognition result, determining that the to-be-determined named entity is the final named entity;

If the boundary character recognition result of the to-be-determined named entity is inconsistent with the preset boundary character recognition result, the to-be-determined named entity is used as a new training sample of the semantic recognition model.
The non-volatile readable storage medium according to claim 14, characterized in that it specifically further comprises:

The semantic recognition model is trained by using the newly added training samples to obtain an optimized semantic recognition model.
A computer device, including a non-volatile readable storage medium, a processor, and computer readable instructions stored on the non-volatile readable storage medium and running on the processor, characterized in that the processor The method for implementing semantic recognition based on convolutional neural network when executing the program includes:

Determine the first loss function, the second loss function, and the third loss function according to the initialized first, second, and third convolutional neural networks;

According to the determined first loss function, second loss function, and third loss function, train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain the preset first One convolutional neural network, second convolutional neural network and third convolutional neural network;

Use the first convolutional neural network preset in the semantic recognition model to obtain the text vector of the text to be recognized;

Use the second convolutional neural network preset in the semantic recognition model to determine the named entity in the text to be recognized according to the obtained text vector;

Use the preset third convolutional neural network in the semantic recognition model to determine the entity relationship in the text to be recognized based on the obtained text vector and the determined named entity;

Among them, using the first convolutional neural network preset in the semantic recognition model to obtain the text vector of the text to be recognized includes:

Use the word vector dictionary to obtain the word vector and word vector of the text to be recognized;

Perform a convolution operation on the obtained word vector and word vector to obtain the text vector of the text to be recognized.
The computer device according to claim 16, characterized in that, according to the determined first loss function, second loss function and third loss function, the initialized first convolutional neural network, second convolutional neural network and The third convolutional neural network is trained to obtain the preset first, second, and third convolutional neural networks, which specifically include:

Determine the loss function of the semantic recognition model according to the determined first loss function, second loss function, and third loss function;

Use the loss function of the semantic recognition model to train the initialized first convolutional neural network, second convolutional neural network, and third convolutional neural network to obtain the preset first convolutional neural network and second convolutional neural network And the third convolutional neural network.
The computer device according to claim 16, wherein the second convolutional neural network preset in the semantic recognition model is used to determine the named entity in the text to be recognized according to the obtained text vector, which specifically includes:

Perform a convolution operation on the obtained text vector to obtain the named entity to be determined;

Perform boundary character recognition on the named entity to be determined, and determine the final named entity according to the recognition result.
The computer device according to claim 18, wherein, performing boundary character recognition on the named entity to be determined, and determining the final named entity according to the recognition result, specifically includes:

If the boundary character recognition result of the to-be-determined named entity is consistent with the preset boundary character recognition result, determining that the to-be-determined named entity is the final named entity;

If the boundary character recognition result of the to-be-determined named entity is inconsistent with the preset boundary character recognition result, the to-be-determined named entity is used as a new training sample of the semantic recognition model.
The computer device according to claim 19, further comprising:

The semantic recognition model is trained by using the newly added training samples to obtain an optimized semantic recognition model.