WO2024179519A1 - Semantic recognition method and apparatus - Google Patents

Abstract

Provided in the present application are a semantic recognition method and apparatus. The semantic recognition method comprises: acquiring N pieces of first text, and second text, wherein each piece of first text comprises a first keyword in a preset word list, and N is a positive integer greater than 1; performing first text processing on each piece of first text by means of a preset text processing model, so as to obtain M character vectors corresponding to the first keyword in each piece of first text, wherein M is a positive integer greater than 1; according to the M character vectors, determining a first word vector corresponding to each piece of first text; and performing semantic recognition on the second text according to the first word vector corresponding to each piece of first text.

Description

Semantic recognition method and device

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims the priority of the Chinese patent application filed with the China Patent Office on March 1, 2023, with application number 202310181906.7 and invention name “Semantic Recognition Method and Device Thereof”. The entire contents of the Chinese patent application are incorporated herein by reference.

Technical Field

The present application belongs to the field of data processing technology, and specifically relates to a semantic recognition method and device thereof.

Background Art

At present, the relevant technology usually performs semantic recognition on the text according to a pre-set vocabulary to obtain the text category corresponding to the text. The above vocabulary includes keywords and text categories. It is easy to understand that the above vocabulary can represent the mapping relationship between keywords and text categories, and the above keywords can be represented by the word vectors included in the vocabulary. However, the word vectors representing the keywords in the above vocabulary may not be accurate enough, which affects the recognition results of semantic recognition and thus reduces the accuracy of semantic recognition.

Summary of the invention

The purpose of the embodiments of the present application is to provide a semantic recognition method and device thereof, which can solve the problem of low accuracy of semantic recognition.

In a first aspect, an embodiment of the present application provides a semantic recognition method, which includes: obtaining N first texts and second texts; each of the first texts includes a first keyword in a preset vocabulary, where N is a positive integer greater than 1; performing first text processing on each of the first texts through a preset text processing model to obtain M character vectors corresponding to the first keyword in each of the first texts, where M is a positive integer greater than 1; determining a first word vector corresponding to each of the first texts based on the M character vectors; and performing semantic recognition on the second text based on the first word vector corresponding to each of the first texts.

In the second aspect, an embodiment of the present application provides a semantic recognition device, which includes: an acquisition module, used to acquire N first texts and second texts; each of the first texts includes a first keyword in a preset vocabulary, N is a positive integer greater than 1; a first processing module, used to perform first text processing on each of the first texts through a preset text processing model, and obtain M character vectors corresponding to the first keyword in each of the first texts, M is a positive integer greater than 1; a first determination module, used to determine the first word vector corresponding to each of the first texts based on the M character vectors; a first recognition module, used to perform semantic recognition on the second text based on the first word vector corresponding to each of the first texts.

In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect.

In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

In a fifth aspect, an embodiment of the present application provides a chip, comprising a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the method described in the first aspect.

In a sixth aspect, an embodiment of the present application provides a computer program product, which is stored in a storage medium and is executed by at least one processor to implement the method described in the first aspect.

The embodiment of the present application performs first text processing on each first text through a preset text processing model, obtains M character vectors corresponding to the first keyword in each first text, and then determines the first word vector based on the M character vectors. The above-mentioned first word vector is determined based on the M character vectors, which fully considers the context information in the first text, and does not directly determine the text feature vector corresponding to the first text as the word vector as in the related art; the above-mentioned M character vectors are obtained by performing first text processing on each first text through the text processing model, thereby taking into account the semantics of the first keyword in the first text. The above method ensures that the first word vector retains the semantics of the keyword. At the same time, it will not be affected by contextual information, thereby improving the accuracy of semantic recognition in the subsequent process of semantic recognition of the second text according to the first word vector corresponding to each first text.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a semantic recognition method provided by an embodiment of the present application;

FIG2 is an application flow chart of updating a vocabulary table provided in an embodiment of the present application;

FIG3 is a structural diagram of a semantic recognition device provided in an embodiment of the present application;

FIG4 is a structural diagram of an electronic device provided in an embodiment of the present application;

FIG5 is a hardware structure diagram of an electronic device provided in an embodiment of the present application.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and/or" in the specification and claims represents at least one of the connected objects, and the character "/" generally indicates that the objects associated with each other are in an "or" relationship.

In the related art, the text is usually classified by a pre-set vocabulary, and the vocabulary can represent the mapping relationship between keywords and text categories, and the keywords can be represented by word vectors included in the vocabulary. Specifically, after obtaining the text, the text keywords in the text are extracted, and the similarity between the word vectors of the text keywords and the word vectors representing the keywords in the vocabulary is calculated. According to the similarity value, the text category corresponding to the text is determined to achieve semantic recognition of the text.

The word vectors of keywords are usually obtained in the following two ways to build a vocabulary:

Method 1:

Input the keyword into a preset text processing model to obtain the word vector corresponding to the keyword. The above text processing model includes a Bert model, a Markov model or other types of text processing models.

Method 2:

The text including the keyword is input into a preset text processing model to obtain a text feature vector corresponding to the text, and then the text feature vector is determined as the word vector corresponding to the keyword. Optionally, if the text is a sentence, the text feature vector is a sentence vector.

However, the word vector obtained by the above method 1 cannot fully express the semantics of the keyword, which in turn affects the accuracy of the vocabulary. In the process of classifying texts through the vocabulary, the accuracy of the semantic recognition results is also reduced.

The word vector obtained by the above method 2 is easily affected by the context information in the text, that is, the word vector obtained by the method 2 is not sufficient, which in turn affects the accuracy of the vocabulary. In the process of classifying the text through the vocabulary, the accuracy of the semantic recognition result is also reduced.

Based on the above-mentioned technical problems, an embodiment of the present application provides a semantic recognition method. The semantic recognition method provided by the embodiment of the present application is described in detail below through specific embodiments and application scenarios in combination with the accompanying drawings.

The present application embodiment provides a semantic recognition method. Please refer to Figure 1, which is a flow chart of the semantic recognition method provided by the present application embodiment. The semantic recognition method provided by the present application embodiment includes the following steps:

S101, obtaining N first texts and second texts.

In this step, N first texts are obtained, and each first text includes a first keyword, where N is a positive integer greater than 1.

Optionally, the N first texts may be obtained from a preset database, or downloaded from the Internet. The specific method for obtaining the N first texts is not limited herein.

Optionally, the first text mentioned above is a public opinion text used in a public opinion analysis project.

Optionally, the first text may be a sentence.

In this step, a second text is obtained, and the second text can be understood as a text to be classified. The second text includes a second keyword, and optionally, the second keyword can be a manually annotated opinion word or an object word representing an opinion.

S102: Perform first text processing on each of the first texts using a preset text processing model to obtain M character vectors corresponding to the first keyword in each of the first texts.

In this embodiment, a text processing model is pre-set. As mentioned above, the text processing model includes a Bert model, a Markov model or other types of text processing models.

In this step, N first texts are input into the text processing model, and each first text is processed by the text processing model to obtain M character vectors corresponding to the first keyword in each first text. M is a positive integer greater than 1.

The above character vector is also called a token vector, and the above character vector is a vector corresponding to each character in the first keyword.

Among them, the text processing model processes the input text to obtain a processing method of a vector corresponding to each character in the input text, which is called the first text processing method.

S103: Determine a first word vector corresponding to each of the first texts according to the M character vectors.

In this step, after obtaining M character vectors corresponding to the first keyword in each first text, these M character vectors can be used to determine the first word vector corresponding to each first text, wherein the first word vector corresponding to the above first text is used to represent the first keyword in the first text.

S104: Perform semantic recognition on the second texts according to the first word vector corresponding to each of the first texts.

In this step, after obtaining the first word vector corresponding to each first text, that is, N first word vectors, semantic recognition can be performed on the second text based on the N first word vectors. For the specific method of performing semantic recognition on the second text, please refer to the subsequent embodiments.

In the embodiment of the present application, each first text is processed by a preset text processing model to obtain M character vectors corresponding to the first keyword in each first text, and then the M character vectors are processed according to the M Character vector, determine the first word vector. The above-mentioned first word vector is determined based on M character vectors, taking full account of the context information in the first text, rather than directly determining the text feature vector corresponding to the first text as the word vector as in the related art; the above-mentioned M character vectors are obtained by performing first text processing on each first text through a text processing model, thereby taking into account the semantics of the first keyword in the first text. The above-mentioned method ensures that the first word vector will not be affected by the context information while retaining the semantics of the keyword, thereby improving the accuracy of semantic recognition in the subsequent process of semantic recognition of the second text based on the first word vector corresponding to each first text.

Optionally, performing text processing on each of the first texts by using a preset text processing model to obtain M character vectors corresponding to the first keyword in each of the first texts includes:

For each of the first texts, performing first text processing on the first text by using the text processing model to obtain at least two character vectors corresponding to each character in the first text;

Determine M target characters according to the position of the first keyword in the first text; the M target characters are used to represent the first keyword;

An average value of at least part of the character vectors corresponding to each of the target characters is determined as the M character vectors corresponding to the first keyword in the first text.

In this embodiment, after the first text is processed by the text processing model, multiple character vectors corresponding to each character in the first text are obtained; then, based on the position of the character corresponding to the first keyword in the first text, M target characters for representing the first keyword from the multiple characters included in the first text are determined; the average value of at least part of the character vector corresponding to each target character is determined as the character vector corresponding to the target character, and then the M character vectors corresponding to the M target characters are determined as the M character vectors corresponding to the first keyword.

For example, the text processing model is a Bert model including 25 network layers, the first text is represented as a sentence including 10 characters "Cannot charge, charging is too slow", and the first keyword is "too slow".

In this case, the first text is input into the Bert model, and each network layer of the Bert model outputs 10 character vectors, that is, the Bert model outputs a total of 250 character vectors, and each character in the first text corresponds to 25 character vectors. Exemplarily, the output result of the i-th layer of the Bert model can be expressed as [vec_i1, vec_i2, ..., vec_i10].

The first keyword is located at the last three characters in the first text, and then the last three characters in the first text are determined as 3 target characters. Exemplarily, in the i-th layer of the Bert model, the target characters can be represented as [vec_i8, vec_i9, vec_i10].

As described above, each character corresponds to 25 character vectors, that is, each target character also corresponds to 25 character vectors. Optionally, the average value of the 10th to 25th character vectors corresponding to each target character can be determined as the character vector corresponding to the first keyword.

Exemplarily, the character "太" in the first keyword corresponds to 25 character vectors, and these 25 character vectors are the character vectors output by the 1st to 25th network layers of the Bert model respectively; the average value of the character vectors output by the 10th to 25th network layers of the Bert model is calculated, and the average value is determined as a character vector corresponding to the character "太" in the first keyword.

The same sampling method as above is used to determine a character vector corresponding to the character "慢" in the first keyword, and a character vector corresponding to the character "了" in the first keyword, and then the above three character vectors are determined as the three character vectors corresponding to the first keyword.

Optionally, the average value of all character vectors corresponding to each target character may also be determined as the character vector corresponding to the first keyword.

In this embodiment, each first text is processed by a preset text processing model to obtain M character vectors corresponding to the first keyword in each first text, so that the first text including the first keyword and some context information is processed, taking into account the semantics of the first keyword in the first text to avoid semantic loss of the first keyword.

Optionally, determining, based on the M character vectors, a first word vector corresponding to each of the first texts includes:

For each of the first texts, an average value of the M character vectors corresponding to the first keyword in the first text is determined as the first word vector corresponding to the first text.

In this embodiment, the average value of the M character vectors corresponding to the first keyword can be determined as the first word vector corresponding to the first text. It should be understood that in other embodiments, the M character vectors corresponding to the first keyword can be weighted and summed, and the weighted sum result can be determined as the first word vector corresponding to the first text.

In this embodiment, the first word vector corresponding to each first text is determined based on M character vectors, rather than directly determining the text feature vector corresponding to the first text as the word vector as in the related art, so as to avoid the representation of the first word vector being affected by the context information in the first text.

Optionally, performing semantic recognition on the second text according to the first word vector corresponding to each of the first texts includes:

Update a preset vocabulary according to the first word vector corresponding to each of the first texts;

The text category corresponding to the second text is determined through the updated vocabulary.

In this embodiment, a vocabulary is pre-set, the vocabulary includes multiple first keywords and multiple text categories, and the vocabulary is used to represent the mapping relationship between the first keywords and the text categories. The vocabulary can be used to classify texts, as described in the background technology, extract text keywords from the text to be classified, calculate the similarity between the word vector of the above text keyword and the word vector representing the first keyword in the vocabulary, and determine the text category corresponding to the text to be classified according to the similarity value.

In this embodiment, after determining N first word vectors, the above-mentioned preset vocabulary can be updated according to the first word vector corresponding to each first text, and then the text category corresponding to the second text is determined by the updated vocabulary, so as to achieve semantic recognition of the second text.

In other embodiments, the word vector of the second text may also be extracted, and at least one first word vector may be determined based on the similarity between each first word vector and the word vector of the second text, and then the text type corresponding to the first word vector may be determined as the text type corresponding to the second text, thereby achieving semantic recognition of the second text.

If the first keyword includes multiple semantics, and the vocabulary can represent the mapping relationship between the first keyword including multiple semantics and the text category, in this case, the technical solution involved in the following embodiment can be used to update the vocabulary:

Optionally, updating a preset vocabulary according to the first word vector corresponding to each of the first texts includes:

Perform semantic clustering on the N first word vectors to obtain K clusters;

Determine the average value of at least one first word vector included in each of the clusters as the second word vector corresponding to the cluster;

The word list is updated according to the K second word vectors corresponding to the K clusters; the updated word list includes K second word vectors for representing the first keyword, and each second word vector is used to represent a semantics corresponding to the first keyword.

In this embodiment, N first word vectors are clustered, and first word vectors with similar semantics are clustered in one cluster to obtain K clusters; wherein each cluster includes at least one first word vector, and K is a positive integer less than or equal to N.

The average value of at least one first word vector included in each cluster is determined as the second word vector corresponding to the cluster. Specifically, the second word vector corresponding to the cluster can be determined by the following formula:

in, Represents the second word vector, N _I represents the number of second word vectors included in the cluster, Vec_i represents the i-th second word vector, and N_I represents the I-th first text.

After obtaining K clusters and the second word vector corresponding to each cluster, the word list is updated. The updated word list includes K second word vectors for representing the first keyword, that is, the first keyword can be represented by K second word vectors, and each second word vector is used to represent a semantics corresponding to the first keyword. For example, the updated word list includes 3 second word vectors for representing the first keyword, indicating that the updated word list includes 3 semantics of the first keyword.

Optionally, the vocabulary is used to represent a mapping relationship between a first keyword and a text category, and determining the text category corresponding to the second text through the updated vocabulary includes:

In the case where the second keyword included in the second text is not the same keyword as the first keyword, determining the text category corresponding to the second text according to the similarity between the text feature vector corresponding to the second text and each second word vector; the text feature vector corresponding to the second text is obtained by performing second text processing on the second text by the text processing model;

In the case that the second keyword and the first keyword are the same keyword, the text category corresponding to the second text is determined according to the mapping relationship between the first keyword represented by the updated vocabulary and the text category.

It should be understood that the vocabulary is used to characterize the mapping relationship between the first keyword and the text category. In this embodiment, the text can be classified according to the updated vocabulary.

The second keyword is queried in the updated vocabulary. If the updated vocabulary includes the second keyword, that is, the second keyword and the first keyword are the same keyword, the text category corresponding to the second text can be directly determined based on the mapping relationship represented by the updated vocabulary.

If the updated vocabulary does not include the second keyword, that is, the second keyword is not the same keyword as the first keyword, in this case, the second text can be processed by a preset text processing model to obtain a text feature vector corresponding to the second text, wherein the text processing model processes the input text to obtain a text feature vector corresponding to the input text, which is called a second text processing method. Optionally, the second text is a sentence including the second keyword, and the text feature vector corresponding to the second text is a sentence vector.

The similarity between the text feature vector and each second word vector in the vocabulary is calculated. If there is a second word vector whose similarity value with the text feature vector is higher than a preset threshold, the text category that has a mapping relationship with the second word vector is determined as the text category corresponding to the second text.

If there are multiple second word vectors whose similarity values with the text feature vector are higher than the first preset threshold, the second word vector with the highest similarity value among the multiple second word vectors is determined, and the text category that has a mapping relationship with the second word vector is determined as the text category corresponding to the second text.

It should be understood that the above-mentioned method of determining the similarity between the text feature vector and the second word vector includes, but is not limited to, based on cosine similarity between vectors, based on Euclidean distance between vectors or other calculation methods.

In this embodiment, the updated vocabulary is used to perform text classification to achieve semantic recognition of the text. Since the second word vector included in the updated vocabulary can accurately represent the keyword, it will not affect the similarity calculation result between the word vector of the text keyword and the word vector representing the keyword in the vocabulary, thereby improving the accuracy of the semantic recognition result.

If the first keyword includes multiple semantics, but the vocabulary representation includes a mapping relationship between the first keyword with one semantic and the text category, in this case, the technical solution involved in the following embodiment can be used to update the vocabulary:

Optionally, after obtaining N first texts and second texts, the method further includes:

Performing second text processing on the N first texts by using the text processing model to obtain N text feature vectors corresponding to the N first texts one by one;

According to the similarity between the N text feature vectors and the benchmark word vector, the N first texts are screened to obtain L first texts; the benchmark word vector is obtained by performing first text processing on a preset benchmark text through the text processing model;

Determine third word vectors corresponding to L first texts; the third word vector is an average value of L fourth word vectors, and the L fourth word vectors are obtained by performing first text processing on the L first texts based on the text processing model;

Perform semantic recognition on the second text based on L third word vectors corresponding to the first texts.

In this embodiment, the second text processing is performed on N first texts through the text processing model to obtain N text feature vectors. The implementation method of the second text processing is consistent with the implementation method of the second text processing involved in the above embodiment, and will not be repeated here. Optionally, the above first text is a sentence, and the text feature vector is a sentence vector.

The text processing model performs first text processing on the preset reference text to obtain a reference word vector. The reference text includes a first keyword, and the implementation of the first text processing is consistent with the implementation of the first text processing involved in the above embodiment, which will not be repeated here.

The similarity between each text feature vector and the benchmark word vector is calculated, and the first texts corresponding to the text feature vectors whose similarity is higher than a second preset threshold are selected to obtain L first texts.

The L first texts are processed by the text processing model to obtain L fourth word vectors; the average value of the L fourth word vectors is determined as the third word vector, and then the second text is semantically recognized based on the third word vector. Optionally, a preset word list can be updated based on the third word vector, wherein the updated word list includes a third word vector for representing the first keyword, that is, the first keyword can be represented by 1 third word vector, and the third word vector is used to represent a semantics corresponding to the first keyword.

To facilitate understanding of the technical solutions of some embodiments provided in this application, please refer to FIG. As shown, in some embodiments of the present application, a vocabulary is pre-set, and the preset vocabulary can be updated, and then the second text is semantically recognized through the updated vocabulary. The application process of updating the vocabulary provided by the present application is: obtain N first texts including the first keyword.

When the first keyword includes multiple semantics and the vocabulary can represent the mapping relationship between the first keyword including multiple semantics and the text category, the vocabulary is updated in the following manner: determine N first word vectors corresponding to N first texts, perform semantic clustering on the N first word vectors, and obtain K cluster clusters; determine the second word vector corresponding to each cluster cluster; and update the vocabulary according to the K second word vectors corresponding to the K cluster clusters.

In the case where the first keyword includes multiple semantics and the vocabulary representation includes a mapping relationship between the first keyword of one semantic and the text category, the vocabulary is updated in the following manner: second text processing is performed on N first texts through a text processing model to obtain N text feature vectors; based on the similarity between the N text feature vectors and the benchmark word vectors, L first texts are screened and obtained; the third word vectors corresponding to the L first texts are determined; and based on the third word vectors, the vocabulary is updated.

Optionally, performing semantic recognition on the second text according to the L third word vectors corresponding to the first texts includes:

According to the third word vectors corresponding to the L first texts, a preset word list is updated; the updated word list includes the third word vector used to represent the first keyword, and the updated word list is used to represent the mapping relationship between the first keyword and the text category;

In the case where the second keyword included in the second text is not the same keyword as the first keyword, determining the text category corresponding to the second text according to the similarity between the text feature vector corresponding to the second text and the third word vector; the text feature vector corresponding to the second text is obtained by performing second text processing on the second text by the text processing model;

In the case that the second keyword and the first keyword are the same keyword, the text category corresponding to the second text is determined according to the mapping relationship between the first keyword represented by the updated vocabulary and the text category.

In this embodiment, a vocabulary is pre-set, the vocabulary includes a plurality of first keywords and a plurality of text categories, and the vocabulary is used to represent the mapping relationship between the first keywords and the text categories. The vocabulary can be Used to categorize text.

In this embodiment, the preset word list can be updated according to the third word vector. The specific updating method is consistent with the above-mentioned method of updating the word list according to the first word vector, and will not be repeated here. Among them, the updated word list includes a third word vector for representing the first keyword, that is, the first keyword can be represented by 1 third word vector, and the third word vector is used to represent a semantics corresponding to the first keyword.

In this embodiment, the text can be classified according to the updated vocabulary, thereby achieving semantic recognition of the text.

The second keyword included in the second text is queried in the updated vocabulary. If the updated vocabulary includes the second keyword, that is, the second keyword and the first keyword are the same keyword, the text category corresponding to the second text can be directly determined based on the mapping relationship represented by the updated vocabulary.

If the updated vocabulary does not include the second keyword, that is, the second keyword is not the same keyword as the first keyword, in this case, the second text can be processed by a preset text processing model to obtain a text feature vector corresponding to the second text; then the similarity between the text feature vector and the third word vector in the vocabulary is calculated. If the similarity value between the text feature vector and the third word vector is higher than a third preset threshold, the text category that has a mapping relationship with the third word vector is determined as the text category corresponding to the second text.

In this embodiment, the text is classified by the updated vocabulary to achieve semantic recognition of the text. Since the third word vector included in the updated vocabulary can accurately represent the keyword, it will not affect the similarity calculation result between the word vector of the text keyword and the word vector representing the keyword in the vocabulary, thereby improving the accuracy of the semantic recognition result.

The semantic recognition device provided in the embodiment of the present application is described in detail below through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

As shown in FIG3 , the semantic recognition device 300 includes:

An acquisition module 301 is used to acquire N first texts and second texts; each of the first texts includes a first keyword in a preset vocabulary, and N is a positive integer greater than 1;

The first processing module 302 is used to perform first text processing on each of the first texts using a preset text processing model to obtain M character vectors corresponding to the first keyword in each of the first texts. quantity, M is a positive integer greater than 1;

A first determination module 303, configured to determine a first word vector corresponding to each of the first texts according to the M character vectors;

The first recognition module 304 is used to perform semantic recognition on the second text according to the first word vector corresponding to each of the first texts.

Optionally, the first processing module 302 is specifically configured to:

For each of the first texts, performing first text processing on the first text by using the text processing model to obtain at least two character vectors corresponding to each character in the first text;

Determine M target characters according to the position of the first keyword in the first text; the M target characters are used to represent the first keyword;

An average value of at least part of the character vectors corresponding to each of the target characters is determined as the M character vectors corresponding to the first keyword in the first text.

Optionally, the first determining module 303 is specifically configured to:

For each of the first texts, an average value of the M character vectors corresponding to the first keyword in the first text is determined as the first word vector corresponding to the first text.

Optionally, the first identification module 304 is specifically configured to:

According to the first word vector corresponding to each of the first texts, a preset word list is updated; the word list includes the first keyword in each of the first texts;

The text category corresponding to the second text is determined through the updated vocabulary.

Optionally, the first identification module 304 is further specifically configured to:

Perform semantic clustering on the N first word vectors to obtain K clusters; each cluster includes at least one first word vector, and K is a positive integer less than or equal to N;

Determine the average value of at least one first word vector included in each of the clusters as the second word vector corresponding to the cluster;

The word list is updated according to the K second word vectors corresponding to the K clusters; the updated word list includes K second word vectors for representing the first keyword, and each second word vector is used to represent a semantics corresponding to the first keyword.

Optionally, the vocabulary is used to characterize a mapping relationship between the first keyword and the text category;

The first identification module 304 is further specifically configured to:

In the case where the second keyword included in the second text is not the same keyword as the first keyword, determining the text category corresponding to the second text according to the similarity between the text feature vector corresponding to the second text and each second word vector; the text feature vector corresponding to the second text is obtained by performing second text processing on the second text by the text processing model;

In the case that the second keyword and the first keyword are the same keyword, the text category corresponding to the second text is determined according to the mapping relationship between the first keyword represented by the updated vocabulary and the text category.

Optionally, the semantic recognition device 300 further includes:

A second processing module, configured to perform second text processing on the N first texts by using the text processing model to obtain N text feature vectors corresponding to the N first texts one by one;

A screening module, configured to screen the N first texts according to the similarity between the N text feature vectors and the benchmark word vectors to obtain L first texts; the benchmark word vectors are obtained by performing first text processing on a preset benchmark text through the text processing model;

A second determination module is used to determine third word vectors corresponding to L first texts; the third word vector is an average value of L fourth word vectors, and the L fourth word vectors are obtained by performing first text processing on the L first texts based on the text processing model;

The second recognition module is used to perform semantic recognition on the second text according to the L third word vectors corresponding to the first texts.

Optionally, the second identification module is specifically used to:

According to the L third word vectors corresponding to the first texts, a preset word list is updated; the updated word list includes the third word vector used to represent the first keyword, and the updated word list is used to represent the mapping relationship between the first keyword and the text category;

In the case where the second keyword included in the second text is not the same keyword as the first keyword, the text category corresponding to the second text is determined according to the similarity between the text feature vector corresponding to the second text and the third word vector; the text feature vector corresponding to the second text The amount is obtained by performing second text processing on the second text by the text processing model;

In the case that the second keyword and the first keyword are the same keyword, the text category corresponding to the second text is determined according to the mapping relationship between the first keyword represented by the updated vocabulary and the text category.

The embodiment of the present application performs first text processing on each first text through a preset text processing model, obtains M character vectors corresponding to the first keyword in each first text, and then determines the first word vector based on the M character vectors. The above-mentioned first word vector is determined based on the M character vectors, which fully considers the context information in the first text, and does not directly determine the text feature vector corresponding to the first text as the word vector as in the related art; the above-mentioned M character vectors are obtained by performing first text processing on each first text through the text processing model, thereby taking into account the semantics of the first keyword in the first text. The above method ensures that the first word vector will not be affected by the context information while retaining the semantics of the keyword, thereby improving the accuracy of semantic recognition in the subsequent process of semantic recognition of the second text based on the first word vector corresponding to each first text.

The semantic recognition device in the embodiment of the present application can be an electronic device, or a component in the electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or it can be other devices other than a terminal. Exemplarily, the electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a car-mounted electronic device, a mobile Internet device (Mobile Internet Device, MID), an augmented reality (augmented reality, AR)/virtual reality (virtual reality, VR) device, a robot, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc. It can also be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., and the embodiment of the present application does not make specific limitations.

The semantic recognition device in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

The embodiment of the present application provides a semantic recognition device that can implement various aspects of the method embodiment of FIG. 1. To avoid repetition, the process will not be described here.

Optionally, as shown in Figure 4, an embodiment of the present application also provides an electronic device 400, including a processor 401, a memory 402, and a program or instruction stored in the memory 402 and executable on the processor 401. When the program or instruction is executed by the processor 401, each process of the above-mentioned semantic recognition method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

FIG5 is a schematic diagram of the hardware structure of an electronic device implementing an embodiment of the present application.

The electronic device 500 includes but is not limited to: a radio frequency unit 501, a network module 502, an audio output unit 503, an input unit 504, a sensor 505, a display unit 506, a user input unit 505, an interface unit 508, a memory 509, and a processor 510.

Those skilled in the art will appreciate that the electronic device 500 may also include a power source (such as a battery) for supplying power to each component, and the power source may be logically connected to the processor 510 through a power management system, so that the power management system can manage charging, discharging, and power consumption. The electronic device structure shown in FIG5 does not constitute a limitation on the electronic device, and the electronic device may include more or fewer components than shown, or combine certain components, or arrange components differently, which will not be described in detail here.

The input unit 504 is further used to obtain N first texts and second texts;

The processor 510 is further configured to perform first text processing on each first text by using a preset text processing model to obtain M character vectors corresponding to the first keyword in each first text;

Determine, according to the M character vectors, a first word vector corresponding to each first text;

Perform semantic recognition on the second text according to the first word vector corresponding to each of the first texts.

The embodiment of the present application performs first text processing on each first text through a preset text processing model, obtains M character vectors corresponding to the first keyword in each first text, and then determines the first word vector based on the M character vectors. The above first word vector is determined based on the M character vectors, taking full account of the context information in the first text, rather than directly determining the text feature vector corresponding to the first text as the word vector as in the related art; the above M character vectors are obtained through text processing. The model processes each first text, thereby taking into account the semantics of the first keyword in the first text. The above method ensures that the first word vector will not be affected by context information while retaining the semantics of the keyword, thereby improving the accuracy of semantic recognition in the subsequent process of semantic recognition of the second text based on the first word vector corresponding to each first text.

It should be understood that in the embodiment of the present application, the input unit 504 may include a graphics processing unit (GPU) 5041 and a microphone 5042, and the graphics processor 5041 processes the image data of the static picture or video obtained by the image capture device (such as a camera) in the video capture mode or the image capture mode. The display unit 506 may include a display panel 5061, and the display panel 5061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 506 includes a touch panel 5061 and at least one of other input devices 5062. The touch panel 5061 is also called a touch screen. The touch panel 5061 may include two parts: a touch detection device and a touch controller. Other input devices 5062 may include, but are not limited to, a physical keyboard, function keys (such as a volume control key, a switch key, etc.), a trackball, a mouse, and a joystick, which will not be repeated here.

The memory 509 can be used to store software programs and various data. The memory 509 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory 509 may include a volatile memory or a non-volatile memory, or the memory 509 may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and Direct Rambus RAM (DRRAM) The memory 509 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

The processor 510 may include one or more processing units; optionally, the processor 510 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 510.

An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned semantic recognition method embodiment are implemented and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned semantic recognition method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

An embodiment of the present application provides a computer program product, which is stored in a storage medium. The program product is executed by at least one processor to implement the various processes of the above-mentioned semantic recognition method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "includes a..." does not exclude the existence of other identical elements in the process, method, article or device that includes the element. In addition, it should be noted that this application The scope of the methods and devices in the embodiments is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved. For example, the described methods may be performed in an order different from that described, and various steps may be added, omitted, or combined. In addition, features described with reference to certain examples may be combined in other examples.

Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM/RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims (20)

A semantic recognition method, comprising: Obtain N first texts and second texts; each of the first texts includes a preset first keyword, and N is a positive integer greater than 1; Performing first text processing on each of the first texts through a preset text processing model to obtain M character vectors corresponding to the first keyword in each of the first texts, where M is a positive integer greater than 1; Determine, according to the M character vectors, a first word vector corresponding to each of the first texts; Perform semantic recognition on the second text according to the first word vector corresponding to each of the first texts. The method according to claim 1, wherein the performing text processing on each of the first texts using a preset text processing model to obtain M character vectors corresponding to the first keyword in each of the first texts comprises: For each of the first texts, performing first text processing on the first text by using the text processing model to obtain at least two character vectors corresponding to each character in the first text; Determine M target characters according to the position of the first keyword in the first text; the M target characters are used to represent the first keyword; An average value of at least part of the character vectors corresponding to each of the target characters is determined as the M character vectors corresponding to the first keyword in the first text. The method according to claim 1, wherein determining the first word vector corresponding to each of the first texts based on the M character vectors comprises: For each of the first texts, an average value of the M character vectors corresponding to the first keyword in the first text is determined as the first word vector corresponding to the first text. The method according to claim 1, wherein the performing semantic recognition on the second text according to the first word vector corresponding to each of the first texts comprises: According to the first word vector corresponding to each of the first texts, a preset word list is updated; the word list includes the first keyword in each of the first texts; The text category corresponding to the second text is determined through the updated vocabulary. The method according to claim 4, wherein the updating a preset vocabulary according to the first word vector corresponding to each of the first texts comprises: Perform semantic clustering on the N first word vectors to obtain K clusters; each cluster includes at least one first word vector, and K is a positive integer less than or equal to N; Determine the average value of at least one first word vector included in each of the clusters as the second word vector corresponding to the cluster; The word list is updated according to the K second word vectors corresponding to the K clusters; the updated word list includes K second word vectors for representing the first keyword, and each second word vector is used to represent a semantics corresponding to the first keyword. The method according to claim 4, wherein the vocabulary is used to represent the mapping relationship between the first keyword and the text category; The determining, by using the updated vocabulary, a text category corresponding to the second text includes: In the case where the second keyword included in the second text is not the same keyword as the first keyword, determining the text category corresponding to the second text according to the similarity between the text feature vector corresponding to the second text and each second word vector; the text feature vector corresponding to the second text is obtained by performing second text processing on the second text by the text processing model; In the case that the second keyword and the first keyword are the same keyword, the text category corresponding to the second text is determined according to the mapping relationship between the first keyword represented by the updated vocabulary and the text category. The method according to claim 1, wherein after obtaining the N first texts and the second texts, the method further comprises: Performing second text processing on the N first texts by using the text processing model to obtain N text feature vectors corresponding to the N first texts one by one; According to the similarity between the N text feature vectors and the benchmark word vector, the N first texts are screened to obtain L first texts; the benchmark word vector is obtained by performing first text processing on a preset benchmark text through the text processing model; Determine the third word vectors corresponding to the L first texts; the third word vector is the L fourth an average value of word vectors, where the L fourth word vectors are obtained by performing first text processing on the L first texts based on the text processing model; Perform semantic recognition on the second text based on L third word vectors corresponding to the first texts. The method according to claim 7, wherein the performing semantic recognition on the second text according to the L third word vectors corresponding to the first texts comprises: According to the L third word vectors corresponding to the first texts, a preset word list is updated; the updated word list includes the third word vector used to represent the first keyword, and the updated word list is used to represent the mapping relationship between the first keyword and the text category; In the case where the second keyword included in the second text is not the same keyword as the first keyword, determining the text category corresponding to the second text according to the similarity between the text feature vector corresponding to the second text and the third word vector; the text feature vector corresponding to the second text is obtained by performing second text processing on the second text by the text processing model; In the case that the second keyword and the first keyword are the same keyword, the text category corresponding to the second text is determined according to the mapping relationship between the first keyword represented by the updated vocabulary and the text category. A semantic recognition device, comprising: An acquisition module, used to acquire N first texts and second texts; each of the first texts includes a first keyword in a preset vocabulary, and N is a positive integer greater than 1; A first processing module, configured to perform first text processing on each of the first texts through a preset text processing model to obtain M character vectors corresponding to the first keyword in each of the first texts, where M is a positive integer greater than 1; A first determination module, configured to determine a first word vector corresponding to each of the first texts according to the M character vectors; The first recognition module is used to perform semantic recognition on the second text according to the first word vector corresponding to each of the first texts. The device according to claim 9, wherein the first processing module is specifically configured to: For each of the first texts, performing first text processing on the first text by using the text processing model to obtain at least two character vectors corresponding to each character in the first text; Determine M target characters according to the position of the first keyword in the first text; the M target characters are used to represent the first keyword; An average value of at least part of the character vectors corresponding to each of the target characters is determined as the M character vectors corresponding to the first keyword in the first text. The apparatus according to claim 9, wherein the first determining module is specifically configured to: For each of the first texts, an average value of the M character vectors corresponding to the first keyword in the first text is determined as the first word vector corresponding to the first text. The device according to claim 9, wherein the first identification module is specifically configured to: According to the first word vector corresponding to each of the first texts, a preset word list is updated; the word list includes the first keyword in each of the first texts; The text category corresponding to the second text is determined through the updated vocabulary. The device according to claim 12, wherein the first identification module is further specifically configured to: Perform semantic clustering on the N first word vectors to obtain K clusters; each cluster includes at least one first word vector, and K is a positive integer less than or equal to N; Determine the average value of at least one first word vector included in each of the clusters as the second word vector corresponding to the cluster; The word list is updated according to the K second word vectors corresponding to the K clusters; the updated word list includes K second word vectors for representing the first keyword, and each second word vector is used to represent a semantics corresponding to the first keyword. The apparatus according to claim 12, wherein the vocabulary is used to represent a mapping relationship between the first keyword and the text category; The first identification module is further specifically used for: In the case where the second keyword included in the second text is not the same as the first keyword, according to the correlation between the text feature vector corresponding to the second text and each second word vector similarity, determining a text category corresponding to the second text; a text feature vector corresponding to the second text is obtained by performing second text processing on the second text by the text processing model; In the case that the second keyword and the first keyword are the same keyword, the text category corresponding to the second text is determined according to the mapping relationship between the first keyword represented by the updated vocabulary and the text category. The device according to claim 9, wherein the device further comprises: A second processing module, configured to perform second text processing on the N first texts by using the text processing model to obtain N text feature vectors corresponding to the N first texts one by one; A screening module, configured to screen the N first texts according to the similarity between the N text feature vectors and the benchmark word vectors to obtain L first texts; the benchmark word vectors are obtained by performing first text processing on a preset benchmark text through the text processing model; A second determination module is used to determine third word vectors corresponding to L first texts; the third word vector is an average value of L fourth word vectors, and the L fourth word vectors are obtained by performing first text processing on the L first texts based on the text processing model; The second recognition module is used to perform semantic recognition on the second text according to the L third word vectors corresponding to the first texts. The device according to claim 15, wherein the second identification module is specifically configured to: According to the L third word vectors corresponding to the first texts, a preset word list is updated; the updated word list includes the third word vector used to represent the first keyword, and the updated word list is used to represent the mapping relationship between the first keyword and the text category; In the case where the second keyword included in the second text is not the same keyword as the first keyword, determining the text category corresponding to the second text according to the similarity between the text feature vector corresponding to the second text and the third word vector; the text feature vector corresponding to the second text is obtained by performing second text processing on the second text by the text processing model; In the case that the second keyword and the first keyword are the same keyword, the text category corresponding to the second text is determined according to the mapping relationship between the first keyword represented by the updated vocabulary and the text category. An electronic device, comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the semantic recognition method as described in any one of claims 1 to 8. A readable storage medium, wherein the readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the semantic recognition method as described in any one of claims 1 to 8 are implemented. A chip, wherein the chip comprises a processor and a communication interface, the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the steps of the semantic recognition method as described in any one of claims 1 to 8. A computer program product, wherein the computer program product is stored in a storage medium, and the computer program product is executed by at least one processor to implement the steps of the semantic recognition method according to any one of claims 1 to 8.