WO2021073119A1

WO2021073119A1 - Method and apparatus for entity disambiguation based on intention recognition model, and computer device

Info

Publication number: WO2021073119A1
Application number: PCT/CN2020/093428
Authority: WO
Inventors: 张师琲
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-10-15
Filing date: 2020-05-29
Publication date: 2021-04-22
Also published as: CN111079429A; CN111079429B

Abstract

The present invention relates to the field of artificial intelligence, and discloses a method and an apparatus for entity disambiguation based on an intention recognition model, and a computer device and a storage medium: acquiring a first sentence to be disambiguated, and acquiring an entity word labelled as ambiguous in the first sentence; selecting a specified standard sentence; calculating a first distance between the first sentence and the specified standard sentence; if the first distance is less than a first distance threshold, then acquiring a specified intention recognition model; inputting the first sentence into the specified intention recognition model to obtain a recognition result, the specified intention recognition model being trained using sample data, and the sample data only being composed of sentences labelled as a specified type of intention; and, if the recognition result is successful recognition, then acquiring a specified entity meaning corresponding to the first sentence, and labelling the entity word with the specified entity meaning. A new dimension (intention recognition) is introduced into the process of disambiguation, increasing the accuracy of entity disambiguation.

Description

Entity disambiguation method, device and computer equipment based on intention recognition model

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on October 15, 2019, the application number is 201910978260.9, and the invention title is "Method, Apparatus, and Computer Equipment for Entity Disambiguation Based on Intent Recognition Model", and its entire contents Incorporated in this application by reference.

Technical field

This application relates to the field of artificial intelligence, and in particular to an entity disambiguation method, device, computer equipment and storage medium based on an intention recognition model.

Background technique

Entity disambiguation is a key task in natural language processing. Since entity references (such as nouns) existing in massive data can usually correspond to multiple named entity concepts, this undoubtedly creates a great obstacle to entity disambiguation. The task of entity disambiguation is to match these ambiguous entity references among a large number of candidate entities to match the corresponding target entities. The inventor realizes that the accuracy of the current entity disambiguation scheme is insufficient. For example, using entity links for disambiguation requires linking the named entity to be disambiguated to the corresponding entity in the external knowledge base for disambiguation, so the accuracy depends on The records of external knowledge bases are not accurate enough to accurately distinguish entities in different contexts. Therefore, the accuracy of current entity disambiguation needs to be improved.

technical problem

The main purpose of this application is to provide an entity disambiguation method, device, computer equipment and storage medium based on an intention recognition model, aiming to improve the accuracy of entity disambiguation.

Technical solutions

In order to achieve the above objective, this application proposes an entity disambiguation method based on an intention recognition model, which includes the following steps:

Acquire the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to a preset ambiguity labeling method, so as to obtain the entity words labeled as ambiguous in the first sentence;

According to the preset standard sentence selection method, select the designated standard sentence from the preset standard sentence database;

Calculate the first distance between the first sentence and the specified standard sentence according to a preset distance calculation formula, and determine whether the first distance is less than a preset first distance threshold;

If the first distance is less than the preset first distance threshold, the specified intent recognition model corresponding to the specified standard sentence is obtained according to the corresponding relationship between the preset standard sentence and the intent recognition model, wherein the specified intent recognition The model is trained using sample data, and the sample data is only composed of sentences marked as a specified type of intent;

Inputting the first sentence into the designated intent recognition model to perform operations, so as to obtain a recognition result output by the designated intent recognition model, wherein the recognition result includes recognition success or recognition failure;

Determine whether the recognition result is successful;

If the recognition result is successful, then according to the preset first sentence-standard sentence-intent recognition model-entity meaning corresponding relationship, the designated entity meaning corresponding to the first sentence is acquired, and the first sentence A disambiguation labeling operation is performed in the sentence, so that the entity word that is labeled as ambiguous is labeled with the specified entity meaning.

This application provides an entity disambiguation device based on an intention recognition model, including:

The entity word acquisition unit is used to acquire the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to a preset ambiguity labeling method, so as to obtain the first sentence marked as ambiguous Entity words

The designated standard sentence acquisition unit is used to select the designated standard sentence from the preset standard sentence database according to the preset standard sentence selection method;

The first distance judgment unit is configured to calculate the first distance between the first sentence and the designated standard sentence according to a preset distance calculation formula, and determine whether the first distance is less than the preset first distance Threshold

A designated intent recognition model acquiring unit, configured to, if the first distance is less than a preset first distance threshold, acquire the designated intent corresponding to the designated standard sentence according to the corresponding relationship between the preset standard sentence and the intent recognition model A recognition model, wherein the designated intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as designated types of intents;

A recognition result obtaining unit, configured to input the first sentence into the designated intent recognition model for calculation, thereby obtaining a recognition result output by the designated intent recognition model, wherein the recognition result includes recognition success or recognition failure;

A recognition result judging unit for judging whether the recognition result is a successful recognition;

The designated entity meaning labeling unit is used to obtain the designated entity corresponding to the first sentence according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning Meaning, and perform a disambiguation labeling operation on the first sentence, so that the entity word labeled as ambiguous is labeled with the specified entity meaning.

The present application provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements the steps of any one of the above methods when the computer program is executed.

The present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of any one of the methods described above are implemented.

Beneficial effect

The entity disambiguation method, device, computer equipment and storage medium based on the intention recognition model of the present application obtain the first sentence to be disambiguated, obtain the entity words marked as ambiguous in the first sentence; select the specified criteria Sentence; calculate the first distance between the first sentence and the specified standard sentence; if the first distance is less than the preset first distance threshold, obtain the specified intent recognition model; input the first sentence The designated intent recognition model performs operations to obtain a recognition result, wherein the designated intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as designated types of intents; if the recognition result For successful recognition, the meaning of the designated entity corresponding to the first sentence is obtained, and the disambiguation operation is performed on the first sentence, so that the entity word marked as ambiguous is marked with the designated entity meaning. Thus, in the process of disambiguation, a new dimension (intent recognition) is introduced to improve the accuracy of entity disambiguation.

Description of the drawings

Fig. 1 is a schematic flowchart of an entity disambiguation method based on an intention recognition model according to an embodiment of this application;

2 is a schematic block diagram of the structure of an entity disambiguation apparatus based on an intention recognition model according to an embodiment of the application;

FIG. 3 is a schematic block diagram of the structure of a computer device according to an embodiment of the application.

The best implementation of this application

1, an embodiment of the present application provides an entity disambiguation method based on an intention recognition model, including the following steps:

S1. Obtain the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to a preset ambiguity labeling method, so as to obtain the entity words marked as ambiguous in the first sentence;

S2, according to the preset standard sentence selection method, select the designated standard sentence from the preset standard sentence database;

S3. Calculate the first distance between the first sentence and the specified standard sentence according to a preset distance calculation formula, and determine whether the first distance is less than a preset first distance threshold;

S4. If the first distance is less than the preset first distance threshold, obtain a specified intent recognition model corresponding to the specified standard sentence according to the correspondence between the preset standard sentence and the intent recognition model, wherein the specified The intention recognition model is trained using sample data, and the sample data is only composed of sentences marked as specified types of intentions;

S5. Input the first sentence into the designated intent recognition model to perform operations, thereby obtaining a recognition result output by the designated intent recognition model, where the recognition result includes recognition success or recognition failure;

S6. Determine whether the recognition result is successful;

S7. If the recognition result is that the recognition is successful, obtain the specified entity meaning corresponding to the first sentence according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning, and compare the A disambiguation labeling operation is performed in the first sentence, so that the entity words that are labeled as ambiguous are labeled with the specified entity meaning.

As described in step S1 above, obtain the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to the preset ambiguity labeling method, so as to obtain the first sentence marked as ambiguous Entity words. The purpose of entity disambiguation in this application is to obtain the true meaning of ambiguous entity words, so the ambiguous entity words need to be marked as ambiguous. The preset ambiguity tagging method is, for example, that the first sentence is input into the bidirectional encoder in the preset ambiguity tagging model for processing, so that the first ambiguity corresponding to each word in the first sentence is one-to-one. Annotate the sequence, and obtain the hidden state vector set of the last-level conversion unit of the two-way encoder, wherein the ambiguity annotation model is composed of a two-way encoder and a support vector machine, and the two-way encoder includes a multi-layer conversion unit; The set of hidden state vectors is input to the support vector machine for operation to obtain a second ambiguity tag sequence corresponding to each word in the first sentence one-to-one; the first ambiguity is calculated according to a preset similarity value calculation method Mark the similarity value between the annotation sequence and the second ambiguous annotation sequence, and determine whether the similarity value is greater than a preset similarity threshold; if the similarity value is greater than the preset similarity threshold, obtain the first The entity words marked as ambiguous in the two-ambiguity tagging sequence.

As described in the above step S2, according to the preset standard sentence selection method, the designated standard sentence is selected from the preset standard sentence database. The standard sentence is used to select a suitable intent recognition model, so it is necessary to pick out the designated standard sentence that is similar to the first sentence. The preset standard sentence selection method is, for example, according to the formula:

Calculate the sentence similarity value sim between the first sentence and a standard sentence in the standard sentence database; determine whether the sentence similarity value sim is greater than the preset sentence similarity threshold in the standard sentence database Standard sentence

If it exists, the standard sentence whose sentence similarity value sim is greater than the preset sentence similarity threshold is recorded as the designated standard sentence.

As described in step S3 above, calculate the first distance between the first sentence and the specified standard sentence according to the preset distance calculation formula, and determine whether the first distance is less than the preset first distance threshold . The first distance reflects the degree of similarity between the first sentence and the specified standard sentence. If the value of the first distance is smaller, the more similar is indicated. When the first sentence is exactly the same as the specified standard sentence, the The first distance is equal to zero. The preset distance calculation formula is, for example, by querying a preset word vector library, obtaining a first word vector sequence I corresponding to the first sentence, and obtaining a second word vector sequence R corresponding to the specified standard sentence ; According to the formula:

Calculate the first distance D between the first sentence and the specified standard sentence, where |I| is the number of words in the first word vector sequence; |R| is the number of words in the second word vector sequence The number of words; w is the word vector; α is the amplification factor for adjusting the cosine similarity between the two word vectors; max(α×cosDis(w, R) is the word vector corresponding to all words in the second word vector sequence R The maximum value of the cosine similarity with the word vector w in the first word vector sequence I.

As described in step S4 above, if the first distance is less than the preset first distance threshold, the specified intent recognition model corresponding to the specified standard sentence is obtained according to the corresponding relationship between the preset standard sentence and the intent recognition model , Wherein the designated intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as designated types of intents. If the first distance is less than the preset first distance threshold, it indicates that there is an applicable intention recognition model, and according to the corresponding relationship between the preset standard sentence and the intention recognition model, the designated standard sentence corresponding to the designated standard sentence is obtained. Intent recognition model. In addition, the designated intent recognition model used in this application is trained using sample data, and the sample data is only composed of sentences marked as designated types of intents, so that the size of the designated intent recognition model is smaller and the training data required is smaller. , It is easier to train, and the accuracy of intention recognition is higher for sentences within a limited range (that is, sentences similar to the specified standard sentence, such as the first sentence). Further, the sample data for training the specified intent recognition model consists of only a limited number of words, and the limited number of words is the same or similar to the words in the first sentence, so that the training is faster and the first sentence is more efficient. Recognition is more accurate (because the number of words in the sample data is limited and is the same or similar to the words in the first sentence, so the sample data can find all training sentences by traversal method, so the first sentence must be in the training process Sentences that have appeared, so it is more accurate and faster to recognize the first sentence).

As described in step S5 above, the first sentence is input into the designated intent recognition model to perform operations, so as to obtain a recognition result output by the designated intent recognition model, wherein the recognition result includes recognition success or recognition failure. Since the designated intent recognition model can only recognize one type of intent (namely, the designated intent type), its successful recognition means that the first sentence is the designated intent type. If the recognition fails, other intent recognition models need to be adopted. Identify again.

As described in step S6 above, it is determined whether the recognition result is successful. Because there are only two recognition results: recognition success or recognition failure. When the recognition is successful, it indicates that the first sentence is the designated intent type, otherwise, the intent type of the first sentence cannot be determined.

As described in step S7 above, if the recognition result is successful, then according to the preset first sentence-standard sentence-intent recognition model-entity meaning correspondence relationship, the designated entity meaning corresponding to the first sentence is obtained , And perform a disambiguation labeling operation on the first sentence, so that the entity word labeled as ambiguous is labeled with the specified entity meaning.

Ambiguous entity words have different meanings in different intention contexts, and if the specific intention type can be identified, the exact meaning of ambiguous words can also be determined. Accordingly, this application obtains the specified entity meaning corresponding to the first sentence according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning, and disambiguates the first sentence Marking operation, so that the entity word marked as ambiguous is marked with the specified entity meaning. Therefore, the actual meaning of the entity words marked as ambiguous in the first sentence can be known from the meaning of the specified entity. For example, the first sentence is: My phone is broken, borrow your apple and use it. Use the specified standard sentence like "My phone is dead, can I use your Apple?" to obtain the corresponding intent recognition model (used to recognize the intent to use the phone), and get the recognition result of successful recognition, and then According to the corresponding relationship of the first sentence-standard sentence-intent recognition model-entity meaning (phone), the apple in the first sentence can be marked with the designated entity meaning (phone).

In one embodiment, the step S1 of performing ambiguity labeling processing on the first sentence according to the preset ambiguity labeling method, so as to obtain the entity words marked as ambiguous in the first sentence, includes:

S101. Input the first sentence into a two-way encoder in a preset ambiguity labeling model for processing, so that a first ambiguity label sequence corresponding to each word in the first sentence one-to-one, and obtain the two-way A set of hidden state vectors of the conversion unit of the last layer of the encoder, wherein the ambiguity annotation model is composed of a bidirectional encoder and a support vector machine, and the bidirectional encoder includes a multi-layer conversion unit;

S102. Input the set of hidden state vectors into the support vector machine for operation to obtain a second ambiguity tag sequence corresponding to each word of the first sentence one-to-one, where the function used by the support vector machine for operation is

among them

Is the label value corresponding to the i-th word of the first sentence, y is the independent variable, yi is the label corresponding to the i-th word of the first sentence, w _yi is the parameter vector corresponding to the i-th word, hi Is the hidden state vector corresponding to the i-th word, w _yi and hi have the same number of component vectors;

S103. Calculate the similarity value between the first ambiguous annotation sequence and the second ambiguous annotation sequence according to a preset similarity value calculation method, and determine whether the similarity value is greater than a preset similarity threshold;

S104: If the similarity degree value is greater than a preset similarity degree threshold, obtain an entity word marked as ambiguous in the second ambiguous annotation sequence.

As described above, the ambiguity annotation processing on the first sentence is implemented, so as to obtain the entity words marked as ambiguous in the first sentence. This application uses an ambiguity annotation model with a special structure for ambiguity annotation. The ambiguity labeling model is composed of a bidirectional encoder and a support vector machine, thereby improving the accuracy of ambiguity labeling. The support vector machine is a model that can be used for labeling, but its input features need to be manually set, so the accuracy is low. Therefore, this application uses the hidden state vector set of the last layer of the conversion unit of the two-way encoder as the support vector machine The input improves the accuracy. The bidirectional encoder includes a multi-layer conversion unit, wherein the conversion unit is composed of multiple encoders and decoders, and can output a first ambiguity annotation sequence, which is used as a reference for whether the second ambiguity annotation sequence is accurate. Then calculate the similarity value between the first ambiguous annotation sequence and the second ambiguous annotation sequence, and if the similarity value is greater than the preset similarity threshold, it indicates that the annotation of the ambiguous annotation model is accurate, and then the first ambiguous annotation sequence is obtained. The entity words marked as ambiguous in the two-ambiguity tagging sequence. Wherein, calculating the similarity value between the first ambiguity annotation sequence and the second ambiguity annotation sequence may be any method, for example, a calculation method based on cosine similarity is adopted.

In one embodiment, the step S2 of selecting a specified standard sentence from a preset standard sentence database according to a preset standard sentence selection method includes:

S201. According to the formula:

Calculate the sentence similarity value sim between the first sentence and a standard sentence in the standard sentence database, where A is the word frequency vector of the first sentence, B is the word frequency vector of the standard sentence, and Ai is the first The number of times the i-th word of the sentence appears in the whole sentence; Bi is the number of times the i-th word of the standard sentence appears in the whole sentence;

S202: Determine whether there is a standard sentence with the sentence similarity value sim greater than a preset sentence similarity threshold in the standard sentence database;

S203: If it exists, record the standard sentence with the sentence similarity value sim greater than the preset sentence similarity threshold as the designated standard sentence.

As described above, the specified standard sentence is selected from the preset standard sentence database. The more similar the designated standard sentence is to the first sentence, the better the final disambiguation effect. This application is based on the formula:

Calculate the sentence similarity value sim between the first sentence and a standard sentence in the standard sentence database; determine whether the sentence similarity value sim is greater than the preset sentence similarity threshold in the standard sentence database Standard sentence; if it exists, the standard sentence whose sentence similarity value sim is greater than the preset sentence similarity threshold is recorded as the designated standard sentence. Wherein, the sentence similarity value sim is used to measure the similarity between two sentences, and its maximum value is 1. When the value is 1, it indicates that the two sentences have exactly the same words. Accordingly, a designated standard sentence similar to the first sentence is selected. Among them, the word frequency vector is composed of the number of occurrences of each word as the value of the sub-vector. For example, the sentence is: I say I want a book, then it has four words (I, say, want, book), which constitutes the word frequency The vector is (2,1,1,1).

In one embodiment, the step S3 of calculating the first distance between the first sentence and the designated standard sentence according to a preset distance calculation formula includes:

S301: Obtain a first word vector sequence I corresponding to the first sentence by querying a preset word vector library, and obtain a second word vector sequence R corresponding to the designated standard sentence;

S302. According to the formula:

As described above, the calculation of the first distance between the first sentence and the specified standard sentence is realized. Among them, the word vector library stores word vectors, which are used to convert words into vector forms to facilitate computer understanding. The word vector database can be obtained by using an existing database, or by using the word vector training tool word2vec to train a pre-collected corpus. According to the formula:

Calculate the first distance D between the first sentence and the specified standard sentence. Substituting the first word vector sequence I corresponding to the first sentence and the second word vector sequence R corresponding to the designated standard sentence into the above formula, the first sentence and the designated The first distance D between standard sentences.

In one embodiment, if the first distance is less than a preset first distance threshold, then according to the corresponding relationship between the preset standard sentence and the intent recognition model, the designated intent recognition corresponding to the designated standard sentence is obtained. A model, where the designated intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as designated types of intents before step S4, including:

S31. Obtain multiple pre-collected sample data, and divide the multiple sample data into training data and test data; wherein the sample data is a sentence marked as a specified type of intention;

S32. Input the training data into a preset neural network model for training, where the training adopts a stochastic gradient descent method to obtain an intermediate intention recognition model;

S33. Use the test data to verify the intermediate intention recognition model, and determine whether the verification passes;

S34. If the verification is passed, record the intermediate intent recognition as the designated intent recognition model.

As mentioned above, the training specified intent recognition model is realized. This application uses sample data for training. The sample data is only composed of sentences marked as specified types of intents, thereby reducing the amount of training data, and only one type of intent needs to be recognized, transforming complex multi-classification tasks It becomes a simple two-classification task, which improves the accuracy and speed of recognition. The neural network model is, for example, VGG16 model, ResNet50 model, DPN131 model, InceptionV3 model, etc. The stochastic gradient descent method refers to randomly sampling some training data for training, which can solve the problem of slow training speed caused by a large amount of training data. The test data is then used to verify the intermediate intent recognition model, and if the verification is passed, the intermediate intent recognition is recorded as the designated intent recognition model.

In one embodiment, there are multiple designated standard sentences, and after the step S6 of determining whether the recognition result is a successful recognition, the method includes:

S61. If the recognition result is recognition failure, obtain candidate standard sentences from a plurality of designated standard sentences, wherein the second distance between the candidate standard sentence and the first sentence is greater than the first sentence. The distance threshold is and is smaller than the preset second distance threshold;

S62. Obtain a candidate intent recognition model corresponding to the candidate standard sentence according to the preset corresponding relationship between the standard sentence and the intent recognition model;

S63. Input the first sentence into the candidate intent recognition model to perform operations, so as to obtain a second recognition result output by the candidate intent recognition model, where the second recognition result includes recognition success or recognition failure;

S64: Determine whether the second recognition result is successful in recognition;

S65. If the second recognition result is that the recognition is successful, obtain the candidate entity meaning corresponding to the first sentence according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning, and A disambiguation labeling operation is performed on the first sentence, so that the entity words labeled as ambiguous are labeled with the candidate entity meaning.

As described above, it is possible to recognize the intent again using the alternative intent recognition model. Since the intent recognition model of the present application is a small volume model and can only recognize one type of intent, there are cases where the recognition of the designated intent recognition model fails. At this time, if there is a suitable intent recognition model that can successfully recognize the first sentence, then the intent type can still be recognized. This application adopts the method of adjusting the distance threshold to obtain a suitable model, specifically: obtaining candidate standard sentences from a plurality of designated standard sentences, wherein the second distance between the candidate standard sentence and the first sentence It is greater than the first distance threshold and less than the preset second distance threshold; according to the correspondence between the preset standard sentence and the intent recognition model, obtain the candidate intent recognition model corresponding to the candidate standard sentence. If the candidate intent recognition model can be successfully identified, it can also achieve the purpose of disambiguation. According to this, according to the preset first sentence-standard sentence-intent recognition model-entity meaning correspondence relationship, the first sentence is obtained Corresponding candidate entity meaning, and performing a disambiguation labeling operation on the first sentence, so that the entity word marked as ambiguous is marked with the candidate entity meaning.

In one embodiment, after the step S64 of judging whether the second recognition result is a successful recognition, the method includes:

S641: If the second recognition result is a recognition failure, obtain the number of the specified standard sentences;

S642: Determine whether the number of the designated standard sentences is greater than a preset number threshold;

S643: If the number of the designated standard sentences is not greater than the preset number threshold, perform a tag modification operation, where the tag modification operation is used to modify the tag of an entity word that is tagged as ambiguous to an unambiguous tag.

As mentioned above, annotation feedback is realized. If the second recognition result is recognition failure, and the number of specified standard sentences is not greater than the preset number threshold, it indicates that the first sentence has only one intention, that is, there is no ambiguity in the first sentence, so The aforementioned ambiguous labeling is not accurate, and accordingly, a labeling modification operation is performed, wherein the labeling modification operation is used to modify the label of the entity word that is marked as ambiguous to an unambiguous label. Accordingly, the error of mislabeling of ambiguity can be prevented, and the ambiguity label can be corrected quickly.

2, an embodiment of the present application provides an entity disambiguation device based on an intention recognition model, including:

The entity word acquisition unit 10 is configured to acquire the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to a preset ambiguity labeling method, so as to obtain the first sentence marked as ambiguous Entity words;

The designated standard sentence acquisition unit 20 is configured to select a designated standard sentence from a preset standard sentence database according to a preset standard sentence selection method;

The first distance judgment unit 30 is configured to calculate the first distance between the first sentence and the designated standard sentence according to a preset distance calculation formula, and determine whether the first distance is less than the preset first distance. Distance threshold

The designated intent recognition model acquiring unit 40 is configured to, if the first distance is less than a preset first distance threshold, acquire the designated standard sentence corresponding to the designated standard sentence according to the corresponding relationship between the preset standard sentence and the intent recognition model An intent recognition model, wherein the designated intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as designated types of intents;

The recognition result obtaining unit 50 is configured to input the first sentence into the designated intent recognition model to perform operations, thereby obtaining a recognition result output by the designated intent recognition model, wherein the recognition result includes recognition success or recognition failure;

The recognition result judging unit 60 is used to judge whether the recognition result is a successful recognition;

The designated entity meaning labeling unit 70 is configured to, if the recognition result is successful, obtain the designated entity corresponding to the first sentence according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning Entity meaning, and perform a disambiguation labeling operation on the first sentence, so that the entity word that is labeled as ambiguous is labeled with the specified entity meaning.

In one embodiment, the entity word acquiring unit 10 includes:

The two-way encoder processing subunit is used to input the first sentence into the two-way encoder in the preset ambiguity labeling model for processing, so that the first ambiguity label is one-to-one corresponding to each word in the first sentence Sequence, and obtain the hidden state vector set of the last-level conversion unit of the bidirectional encoder, wherein the ambiguity annotation model is composed of a bidirectional encoder and a support vector machine, and the bidirectional encoder includes a multilayer conversion unit;

The second ambiguity tag sequence acquisition subunit is used to input the hidden state vector set into the support vector machine for operation to obtain a second ambiguity tag sequence corresponding to each word of the first sentence one-to-one, wherein the support The function used by the vector machine for calculation is

among them

The similarity value judgment subunit is used to calculate the similarity value between the first ambiguity annotation sequence and the second ambiguity annotation sequence according to a preset similarity value calculation method, and to determine whether the similarity value is greater than the expected value. Set the similarity threshold;

The entity word acquiring subunit is configured to acquire the entity word marked as ambiguous in the second ambiguous annotation sequence if the similarity value is greater than a preset similarity threshold.

In one embodiment, the designated standard sentence obtaining unit 20 includes:

The sentence similarity value sim calculation subunit is used according to the formula:

The sentence similarity value sim judgment subunit is used to judge whether there is a standard sentence whose sentence similarity value sim is greater than a preset sentence similarity threshold in the standard sentence database;

The designated standard sentence marking subunit is used to record the standard sentence with the sentence similarity value sim greater than the preset sentence similarity threshold as the designated standard sentence if it exists.

In one embodiment, the first distance determining unit 30 includes:

The word vector database query subunit is used to query a preset word vector database to obtain a first word vector sequence I corresponding to the first sentence, and to obtain a second word vector sequence R corresponding to the specified standard sentence ；

The first distance D calculation subunit is used according to the formula:

In one embodiment, the device includes:

The sample data dividing unit is used to obtain a plurality of pre-collected sample data, and divide the plurality of sample data into training data and test data; wherein, the sample data is a sentence marked as a specified type of intention;

The intermediate intent recognition model acquisition unit is used to input training data into the preset neural network model for training, where the training adopts the stochastic gradient descent method to obtain the intermediate intent recognition model;

The verification passing judgment unit is configured to verify the intermediate intention recognition model by using the test data, and judge whether the verification is passed;

The designated intent recognition model marking unit is used to record the intermediate intent recognition as the designated intent recognition model if the verification is passed.

In one embodiment, there are multiple specified standard sentences, and the device includes:

The candidate standard sentence obtaining unit is configured to obtain a candidate standard sentence from a plurality of designated standard sentences if the recognition result is a recognition failure, wherein the first sentence between the candidate standard sentence and the first sentence 2. The distance is greater than the first distance threshold and less than the preset second distance threshold;

The candidate intent recognition model obtaining unit is configured to obtain the candidate intent recognition model corresponding to the candidate standard sentence according to the preset corresponding relationship between the standard sentence and the intent recognition model;

The second recognition result acquisition unit is configured to input the first sentence into the candidate intent recognition model to perform operations, thereby obtaining a second recognition result output by the candidate intent recognition model, wherein the second recognition result Including recognition success or recognition failure;

The second recognition result judging unit is used to judge whether the second recognition result is a successful recognition;

An alternative entity meaning labeling unit, configured to, if the second recognition result is successful, obtain the corresponding relationship to the first sentence according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning The candidate entity meaning of, and a disambiguation labeling operation is performed on the first sentence, so that the entity word that is labeled as ambiguous is labeled with the candidate entity meaning.

In one embodiment, the device includes:

A quantity acquiring unit, configured to acquire the quantity of the designated standard sentence if the second recognition result is a recognition failure;

A quantity threshold judging unit for judging whether the quantity of the specified standard sentences is greater than a preset quantity threshold;

An annotation modification unit, configured to perform an annotation modification operation if the number of the designated standard sentences is not greater than a preset number threshold, wherein the annotation modification operation is used to modify the annotations of the entity words that are marked as ambiguous to be unambiguous Label.

The operations performed by the aforementioned units or sub-units in the aforementioned embodiment respectively correspond to the steps of the entity disambiguation method based on the intention recognition model of the aforementioned embodiment, and will not be repeated here.

3, an embodiment of the present application also provides a computer device. The computer device may be a server, and its internal structure may be as shown in the figure. The computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor designed by the computer is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer equipment is used to store the data used in the entity disambiguation method based on the intention recognition model. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to realize an entity disambiguation method based on the intention recognition model.

The processor executes the above-mentioned entity disambiguation method based on the intention recognition model, wherein the steps included in the method respectively correspond to the steps of executing the entity disambiguation method based on the intention recognition model of the foregoing embodiment in a one-to-one correspondence, and will not be repeated here.

An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, an entity disambiguation method based on an intention recognition model is implemented, and the storage medium is a volatile storage medium. Or a non-volatile storage medium, wherein the steps included in the method respectively correspond to the steps of performing the entity disambiguation method based on the intention recognition model of the foregoing embodiment, and will not be repeated here.

Claims

An entity disambiguation method based on intent recognition model, which includes:

Acquire the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to a preset ambiguity labeling method, so as to obtain the entity words labeled as ambiguous in the first sentence;

According to the preset standard sentence selection method, select the designated standard sentence from the preset standard sentence database;

Calculate the first distance between the first sentence and the specified standard sentence according to a preset distance calculation formula, and determine whether the first distance is less than a preset first distance threshold;

If the first distance is less than the preset first distance threshold, the specified intent recognition model corresponding to the specified standard sentence is obtained according to the corresponding relationship between the preset standard sentence and the intent recognition model, wherein the specified intent recognition The model is trained using sample data, and the sample data is only composed of sentences marked as specified types of intentions;

Inputting the first sentence into the designated intent recognition model to perform operations, so as to obtain a recognition result output by the designated intent recognition model, wherein the recognition result includes recognition success or recognition failure;

Determine whether the recognition result is successful;

If the recognition result is successful, then according to the preset first sentence-standard sentence-intent recognition model-entity meaning corresponding relationship, the designated entity meaning corresponding to the first sentence is acquired, and the first sentence A disambiguation labeling operation is performed in the sentence, so that the entity word that is labeled as ambiguous is labeled with the specified entity meaning.
The entity disambiguation method based on the intent recognition model according to claim 1, wherein the first sentence is subjected to ambiguity labeling processing according to the preset ambiguity labeling method, so as to obtain the object in the first sentence The steps for marking ambiguous entity words include:

The first sentence is input into the bidirectional encoder in the preset ambiguity annotation model for processing, so that the first ambiguity annotation sequence corresponding to each word in the first sentence one-to-one, and the bidirectional encoder is obtained The hidden state vector set of the last-level conversion unit of, wherein the ambiguity annotation model is composed of a bidirectional encoder and a support vector machine, and the bidirectional encoder includes a multilayer conversion unit;

The hidden state vector set is input into the support vector machine for operation, and a second ambiguous tag sequence corresponding to each word of the first sentence one-to-one is obtained, wherein the function used by the support vector machine for the operation is
among them
Is the label value corresponding to the i-th word of the first sentence, y is the independent variable, yi is the label corresponding to the i-th word of the first sentence, w yi is the parameter vector corresponding to the i-th word, hi Is the hidden state vector corresponding to the i-th word, w yi and hi have the same number of component vectors;

Calculate the similarity value between the first ambiguous annotation sequence and the second ambiguous annotation sequence according to a preset similarity value calculation method, and determine whether the similarity value is greater than a preset similarity threshold;

If the similarity degree value is greater than the preset similarity degree threshold, acquiring the entity words marked as ambiguous in the second ambiguous annotation sequence.
The entity disambiguation method based on the intent recognition model according to claim 1, wherein the step of selecting a specified standard sentence from a preset standard sentence database according to a preset standard sentence selection method includes:

According to the formula:
Calculate the sentence similarity value sim between the first sentence and a standard sentence in the standard sentence database, where A is the word frequency vector of the first sentence, B is the word frequency vector of the standard sentence, and Ai is the first The number of times the i-th word of the sentence appears in the whole sentence; Bi is the number of times the i-th word of the standard sentence appears in the whole sentence;

Judging whether there is a standard sentence whose sentence similarity value sim is greater than a preset sentence similarity threshold in the standard sentence database;

If it exists, the standard sentence whose sentence similarity value sim is greater than the preset sentence similarity threshold is recorded as the designated standard sentence.
The entity disambiguation method based on the intent recognition model according to claim 1, wherein the step of calculating the first distance between the first sentence and the specified standard sentence according to a preset distance calculation formula, include:

By querying a preset word vector library, obtaining a first word vector sequence I corresponding to the first sentence, and obtaining a second word vector sequence R corresponding to the designated standard sentence;

According to the formula:

Calculate the first distance D between the first sentence and the specified standard sentence, where |I| is the number of words in the first word vector sequence; |R| is the number of words in the second word vector sequence The number of words; w is the word vector; α is the magnification factor that adjusts the cosine similarity between the two word vectors; max(α×cosDis(w, R) is the word vector corresponding to all words in the second word vector sequence R The maximum value of the cosine similarity with the word vector w in the first word vector sequence I.
The entity disambiguation method based on the intent recognition model according to claim 1, wherein if the first distance is less than a preset first distance threshold, according to the correspondence between the preset standard sentence and the intent recognition model Before the step of obtaining a designated intent recognition model corresponding to the designated standard sentence, wherein the designated intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as designated types of intents, including:

Acquiring a plurality of sample data collected in advance, and dividing the plurality of sample data into training data and test data; wherein, the sample data is a sentence marked as a specified type of intent;

Input the training data into the preset neural network model for training, where the training adopts the stochastic gradient descent method to obtain the intermediate intention recognition model;

Use the test data to verify the intermediate intention recognition model, and determine whether the verification passes;

If the verification is passed, the intermediate intent recognition is recorded as the designated intent recognition model.
The entity disambiguation method based on an intent recognition model according to claim 1, wherein there are multiple specified standard sentences, and after the step of determining whether the recognition result is successful, the method includes:

If the recognition result is recognition failure, obtain candidate standard sentences from a plurality of specified standard sentences, wherein the second distance between the candidate standard sentence and the first sentence is greater than the first distance threshold And is smaller than the preset second distance threshold;

Obtaining the candidate intent recognition model corresponding to the candidate standard sentence according to the preset corresponding relationship between the standard sentence and the intent recognition model;

Inputting the first sentence into the candidate intent recognition model for calculation, thereby obtaining a second recognition result output by the candidate intent recognition model, where the second recognition result includes recognition success or recognition failure;

Judging whether the second recognition result is a successful recognition;

If the second recognition result is successful, then according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning, the candidate entity meaning corresponding to the first sentence is obtained, and the corresponding The disambiguation labeling operation is performed in the first sentence, so that the entity word that is labeled as ambiguous is labeled with the candidate entity meaning.
The entity disambiguation method based on the intent recognition model according to claim 6, wherein after the step of determining whether the second recognition result is a successful recognition, the method comprises:

If the second recognition result is a recognition failure, obtain the number of the specified standard sentences;

Judging whether the number of specified standard sentences is greater than a preset number threshold;

If the number of the designated standard sentences is not greater than the preset number threshold, a label modification operation is performed, wherein the label modification operation is used to modify the label of the entity word that is marked as ambiguous to an unambiguous label.
An entity disambiguation device based on an intention recognition model, which includes:

The entity word acquisition unit is used to acquire the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to a preset ambiguity labeling method, so as to obtain the first sentence marked as ambiguous Entity words

The designated standard sentence acquisition unit is used to select the designated standard sentence from the preset standard sentence database according to the preset standard sentence selection method;

The first distance judgment unit is configured to calculate the first distance between the first sentence and the designated standard sentence according to a preset distance calculation formula, and determine whether the first distance is less than the preset first distance Threshold

A designated intent recognition model acquiring unit, configured to, if the first distance is less than a preset first distance threshold, acquire the designated intent corresponding to the designated standard sentence according to the corresponding relationship between the preset standard sentence and the intent recognition model A recognition model, wherein the designated intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as designated types of intents;

A recognition result obtaining unit, configured to input the first sentence into the designated intent recognition model for calculation, thereby obtaining a recognition result output by the designated intent recognition model, wherein the recognition result includes recognition success or recognition failure;

A recognition result judging unit for judging whether the recognition result is a successful recognition;

The designated entity meaning labeling unit is used to obtain the designated entity corresponding to the first sentence according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning Meaning, and perform a disambiguation labeling operation on the first sentence, so that the entity word labeled as ambiguous is labeled with the specified entity meaning.
A computer device includes a memory and a processor, the memory stores a computer program, wherein the processor implements an entity disambiguation method based on an intent recognition model when the processor executes the computer program, and the method includes:

Acquire the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to a preset ambiguity labeling method, so as to obtain the entity words labeled as ambiguous in the first sentence;

According to the preset standard sentence selection method, select the designated standard sentence from the preset standard sentence database;

Calculate the first distance between the first sentence and the specified standard sentence according to a preset distance calculation formula, and determine whether the first distance is less than a preset first distance threshold;

If the first distance is less than the preset first distance threshold, the specified intent recognition model corresponding to the specified standard sentence is obtained according to the corresponding relationship between the preset standard sentence and the intent recognition model, wherein the specified intent recognition The model is trained using sample data, and the sample data is only composed of sentences marked as specified types of intentions;

Inputting the first sentence into the designated intent recognition model to perform operations, so as to obtain a recognition result output by the designated intent recognition model, wherein the recognition result includes recognition success or recognition failure;

Determine whether the recognition result is successful;

If the recognition result is successful, then according to the preset first sentence-standard sentence-intent recognition model-entity meaning corresponding relationship, the designated entity meaning corresponding to the first sentence is acquired, and the first sentence A disambiguation labeling operation is performed in the sentence, so that the entity word that is labeled as ambiguous is labeled with the specified entity meaning.
9. The computer device according to claim 9, wherein the first sentence is subjected to ambiguity labeling processing according to a preset ambiguity labeling method, so as to obtain the information of the entity words marked as ambiguous in the first sentence The steps include:

The first sentence is input into the bidirectional encoder in the preset ambiguity annotation model for processing, so that the first ambiguity annotation sequence corresponding to each word in the first sentence one-to-one, and the bidirectional encoder is obtained The hidden state vector set of the last-level conversion unit of, wherein the ambiguity annotation model is composed of a bidirectional encoder and a support vector machine, and the bidirectional encoder includes a multilayer conversion unit;

The hidden state vector set is input into the support vector machine for operation, and a second ambiguous tag sequence corresponding to each word of the first sentence one-to-one is obtained, wherein the function used by the support vector machine for the operation is
among them
Is the label value corresponding to the i-th word of the first sentence, y is the independent variable, yi is the label corresponding to the i-th word of the first sentence, w yi is the parameter vector corresponding to the i-th word, hi Is the hidden state vector corresponding to the i-th word, w yi and hi have the same number of component vectors;

Calculate the similarity value between the first ambiguous annotation sequence and the second ambiguous annotation sequence according to a preset similarity value calculation method, and determine whether the similarity value is greater than a preset similarity threshold;

If the similarity degree value is greater than the preset similarity degree threshold, acquiring the entity words marked as ambiguous in the second ambiguous annotation sequence.
9. The computer device according to claim 9, wherein the step of selecting a specified standard sentence from a preset standard sentence database according to a preset standard sentence selection method comprises:

According to the formula:
Calculate the sentence similarity value sim between the first sentence and a standard sentence in the standard sentence database, where A is the word frequency vector of the first sentence, B is the word frequency vector of the standard sentence, and Ai is the first The number of times the i-th word of the sentence appears in the whole sentence; Bi is the number of times the i-th word of the standard sentence appears in the whole sentence;

Judging whether there is a standard sentence whose sentence similarity value sim is greater than a preset sentence similarity threshold in the standard sentence database;

If it exists, the standard sentence whose sentence similarity value sim is greater than the preset sentence similarity threshold is recorded as the designated standard sentence.
9. The computer device according to claim 9, wherein the step of calculating the first distance between the first sentence and the designated standard sentence according to a preset distance calculation formula comprises:

By querying a preset word vector library, obtaining a first word vector sequence I corresponding to the first sentence, and obtaining a second word vector sequence R corresponding to the designated standard sentence;

According to the formula:

Calculate the first distance D between the first sentence and the specified standard sentence, where |I| is the number of words in the first word vector sequence; |R| is the number of words in the second word vector sequence The number of words; w is the word vector; α is the amplification factor for adjusting the cosine similarity between the two word vectors; max(α×cosDis(w, R) is the word vector corresponding to all words in the second word vector sequence R The maximum value of the cosine similarity with the word vector w in the first word vector sequence I.
9. The computer device according to claim 9, wherein if the first distance is less than a preset first distance threshold, the corresponding relationship between the preset standard sentence and the intent recognition model is obtained to obtain the corresponding relationship with the specified standard The specified intent recognition model corresponding to the sentence, wherein the specified intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as the specified type of intent. Before the step, the steps include:

Acquiring a plurality of sample data collected in advance, and dividing the plurality of sample data into training data and test data; wherein, the sample data is a sentence marked as a specified type of intent;

Input the training data into the preset neural network model for training, where the training adopts the stochastic gradient descent method to obtain the intermediate intention recognition model;

Use the test data to verify the intermediate intention recognition model, and determine whether the verification passes;

If the verification is passed, the intermediate intent recognition is recorded as the designated intent recognition model.
9. The computer device according to claim 9, wherein there are multiple specified standard sentences, and after the step of judging whether the recognition result is successful, the method comprises:

If the recognition result is recognition failure, obtain candidate standard sentences from a plurality of specified standard sentences, wherein the second distance between the candidate standard sentence and the first sentence is greater than the first distance threshold And is smaller than the preset second distance threshold;

Obtaining the candidate intent recognition model corresponding to the candidate standard sentence according to the preset corresponding relationship between the standard sentence and the intent recognition model;

Inputting the first sentence into the candidate intent recognition model for calculation, thereby obtaining a second recognition result output by the candidate intent recognition model, where the second recognition result includes recognition success or recognition failure;

Judging whether the second recognition result is a successful recognition;

If the second recognition result is successful, then according to the preset correspondence relationship of the first sentence-standard sentence-intent recognition model-entity meaning, the candidate entity meaning corresponding to the first sentence is obtained, and the corresponding The disambiguation labeling operation is performed in the first sentence, so that the entity word that is labeled as ambiguous is labeled with the candidate entity meaning.
The computer device according to claim 14, wherein after the step of judging whether the second recognition result is a successful recognition, it comprises:

If the second recognition result is a recognition failure, obtain the number of the specified standard sentences;

Judging whether the number of specified standard sentences is greater than a preset number threshold;

If the number of the designated standard sentences is not greater than the preset number threshold, a label modification operation is performed, wherein the label modification operation is used to modify the label of the entity word that is marked as ambiguous to an unambiguous label.
A computer-readable storage medium having a computer program stored thereon, wherein the computer program implements an entity disambiguation method based on an intention recognition model when the computer program is executed by a processor, and the method includes:

Acquire the first sentence to be disambiguated, and perform ambiguity labeling processing on the first sentence according to a preset ambiguity labeling method, so as to obtain the entity words labeled as ambiguous in the first sentence;

According to the preset standard sentence selection method, select the designated standard sentence from the preset standard sentence database;

Calculate the first distance between the first sentence and the specified standard sentence according to a preset distance calculation formula, and determine whether the first distance is less than a preset first distance threshold;

If the first distance is less than the preset first distance threshold, the specified intent recognition model corresponding to the specified standard sentence is obtained according to the corresponding relationship between the preset standard sentence and the intent recognition model, wherein the specified intent recognition The model is trained using sample data, and the sample data is only composed of sentences marked as specified types of intentions;

Inputting the first sentence into the designated intent recognition model to perform operations, so as to obtain a recognition result output by the designated intent recognition model, wherein the recognition result includes recognition success or recognition failure;

Determine whether the recognition result is successful;

If the recognition result is successful, then according to the preset first sentence-standard sentence-intent recognition model-entity meaning corresponding relationship, the designated entity meaning corresponding to the first sentence is acquired, and the first sentence A disambiguation labeling operation is performed in the sentence, so that the entity word that is labeled as ambiguous is labeled with the specified entity meaning.
16. The computer-readable storage medium according to claim 16, wherein the first sentence is subjected to ambiguity labeling processing according to a preset ambiguity labeling method, so as to obtain the ambiguity in the first sentence The steps of entity words include:

The first sentence is input into the bidirectional encoder in the preset ambiguity annotation model for processing, so that the first ambiguity annotation sequence corresponding to each word in the first sentence one-to-one, and the bidirectional encoder is obtained The hidden state vector set of the last-level conversion unit of, wherein the ambiguity annotation model is composed of a bidirectional encoder and a support vector machine, and the bidirectional encoder includes a multilayer conversion unit;

The hidden state vector set is input into the support vector machine for operation, and a second ambiguous tag sequence corresponding to each word of the first sentence one-to-one is obtained, wherein the function used by the support vector machine for the operation is
among them
Is the label value corresponding to the i-th word of the first sentence, y is the independent variable, yi is the label corresponding to the i-th word of the first sentence, w yi is the parameter vector corresponding to the i-th word, hi Is the hidden state vector corresponding to the i-th word, w yi and hi have the same number of component vectors;

Calculate the similarity value between the first ambiguous annotation sequence and the second ambiguous annotation sequence according to a preset similarity value calculation method, and determine whether the similarity value is greater than a preset similarity threshold;

If the similarity degree value is greater than the preset similarity degree threshold, acquiring the entity words marked as ambiguous in the second ambiguous annotation sequence.
16. The computer-readable storage medium according to claim 16, wherein the step of selecting a specified standard sentence from a preset standard sentence database according to a preset standard sentence selection method comprises:

According to the formula:
Calculate the sentence similarity value sim between the first sentence and a standard sentence in the standard sentence database, where A is the word frequency vector of the first sentence, B is the word frequency vector of the standard sentence, and Ai is the first The number of times the i-th word of the sentence appears in the whole sentence; Bi is the number of times the i-th word of the standard sentence appears in the whole sentence;

Judging whether there is a standard sentence whose sentence similarity value sim is greater than a preset sentence similarity threshold in the standard sentence database;

If it exists, the standard sentence whose sentence similarity value sim is greater than the preset sentence similarity threshold is recorded as the designated standard sentence.
The computer-readable storage medium according to claim 16, wherein the step of calculating the first distance between the first sentence and the designated standard sentence according to a preset distance calculation formula comprises:

Obtaining a first word vector sequence I corresponding to the first sentence and obtaining a second word vector sequence R corresponding to the designated standard sentence by querying a preset word vector library;

According to the formula:

Calculate the first distance D between the first sentence and the specified standard sentence, where |I| is the number of words in the first word vector sequence; |R| is the number of words in the second word vector sequence The number of words; w is the word vector; α is the magnification factor that adjusts the cosine similarity between the two word vectors; max(α×cosDis(w, R) is the word vector corresponding to all words in the second word vector sequence R The maximum value of the cosine similarity with the word vector w in the first word vector sequence I.
The computer-readable storage medium according to claim 16, wherein, if the first distance is less than a preset first distance threshold, the corresponding relationship between the preset standard sentence and the intent recognition model is obtained. The specified intent recognition model corresponding to the specified standard sentence, wherein the specified intent recognition model is trained using sample data, and the sample data is only composed of sentences marked as the specified type of intent. Before the step, the steps include:

Acquiring a plurality of sample data collected in advance, and dividing the plurality of sample data into training data and test data; wherein, the sample data is a sentence marked as a specified type of intent;

Input the training data into the preset neural network model for training, where the training adopts the stochastic gradient descent method to obtain the intermediate intention recognition model;

Use the test data to verify the intermediate intention recognition model, and determine whether the verification passes;

If the verification is passed, the intermediate intent recognition is recorded as the designated intent recognition model.