WO2020224213A1

WO2020224213A1 - Sentence intent identification method, device, and computer readable storage medium

Info

Publication number: WO2020224213A1
Application number: PCT/CN2019/117344
Authority: WO
Inventors: 赵婧; 王健宗
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-05-06
Filing date: 2019-11-12
Publication date: 2020-11-12
Also published as: CN110232114A

Abstract

A sentence intent identification method, a device, and a computer readable storage medium, pertaining to the technical field of speech semantics. The method comprises: pre-processing original sentence samples, and obtaining pre-processed samples (S11); extracting sentence feature vectors from the pre-processed samples (S12); and training, on the basis of the sentence feature vectors, a sentence intent identification model by using a cross-entropy cost function method, and obtaining a trained sentence intent identification model (S13). The method further comprises: acquiring a target sentence to be identified (S14); outputting, on the basis of the target sentence, a sentence intent corresponding to the target sentence by using the trained sentence intent identification model (S15); and acquiring, from an answer database, an answer matching the sentence intent corresponding to the target sentence, and showing the answer to a user (S16). The method enables extraction of abstract features from speech data by using a deep neural network, and achieves accurate sentence identification.

Description

Sentence intention recognition method, device and computer readable storage medium

Based on the Paris Convention, this application declares that it enjoys the priority of the Chinese patent application filed on May 6, 2019 with the application number CN201910370432.4 and titled "Sentence Intent Recognition Method, Device and Computer-readable Storage Medium". This Chinese patent application The overall content of is incorporated in this application by reference.

Technical field

This application relates to the technical field of speech semantics, and in particular to a method, device and computer-readable storage medium for sentence intention recognition.

Background technique

How to understand the user's intention according to the context in the multi-round conversation of a chatbot is an important and difficult problem in multi-round interaction. The existing question comprehension methods are mostly for single sentences, and focus on the understanding of certain sentence structure. How to identify the current user's intentions based on the context, not just a single round of analysis, so that the dialogue has a fine-grained understanding ability in a continuous context is an urgent problem to be solved.

Summary of the invention

This application provides a sentence intention recognition method, device, and computer-readable storage medium, the main purpose of which is to realize the recognition of the current user’s intention according to the context, so that when the user uses natural language expressions to ask questions to the chat robot, Get concise and accurate answers from the chatbot.

In order to achieve the above objective, this application also provides a sentence intention recognition method, which includes:

Obtain a sample of the original sentence;

Prepare the original sentence sample to obtain the preprocessed sample;

Extracting sentence feature vectors from the preprocessed samples;

Training a sentence intention recognition model based on the sentence feature vector and using a cross-entropy cost function method to obtain a trained sentence intention recognition model;

Get the target sentence to be recognized;

Based on the target sentence, and using the trained sentence intention recognition model, output the sentence intention corresponding to the target sentence;

An answer that matches the intent of the sentence corresponding to the target sentence is obtained from the answer database and displayed to the user.

In order to achieve the above objective, the present application also provides a sentence intention recognition device, characterized in that the device includes a memory and a processor, the memory stores a sentence intention recognition program that can run on the processor, and When the sentence intention recognition program is executed by the processor, the following steps are implemented:

Obtain a sample of the original sentence;

Prepare the original sentence sample to obtain the preprocessed sample;

Extracting sentence feature vectors from the preprocessed samples;

Get the target sentence to be recognized;

In addition, in order to achieve the above object, the present application also provides a computer-readable storage medium having a sentence intention recognition program stored on the computer-readable storage medium, and the sentence intention recognition program can be executed by one or more processors, To realize the steps of the sentence intention recognition method as described above.

This application obtains original sentence samples; preprocesses the original sentence samples to obtain preprocessed samples; extracts sentence feature vectors from the preprocessed samples; trains sentences based on the sentence feature vectors and uses the cross-entropy cost function method Intent recognition model to obtain the trained sentence intent recognition model; obtain the target sentence to be recognized; based on the target sentence, and use the trained sentence intent recognition model to output the sentence intent corresponding to the target sentence; obtain the sentence intent corresponding to the target sentence from the answer database The sentence intent to match the answer corresponding to the target sentence is displayed to the user. This application realizes the recognition of the current user's intention based on the context, so that when the user asks a question to the chat robot using natural language expressions, the chat robot returns a concise and accurate answer.

Description of the drawings

FIG. 1 is a schematic flowchart of a method for recognizing sentence intentions according to an embodiment of the application;

2 is a schematic diagram of the internal structure of a sentence intention recognition device provided by an embodiment of the application;

3 is a schematic diagram of modules of a sentence intention recognition program in a sentence intention recognition device provided by an embodiment of the application.

The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the application, and are not used to limit the application.

This application provides a method for identifying sentence intentions. Referring to FIG. 1, it is a schematic flowchart of a sentence intention recognition method provided by an embodiment of this application. The method can be executed by a device, and the device can be implemented by software and/or hardware.

In this embodiment, the sentence intention recognition method includes:

S10. Obtain an original sentence sample.

In this embodiment, the web crawler technology is used to obtain questions asked by users in various application environments from the network.

S11. Preprocess the original sentence sample to obtain a preprocessed sample.

Preferably, the preprocessing the original sentence sample to obtain the preprocessed sample includes:

(1) Use natural language processing technology to segment the original sentence samples to get the sentence after the word segmentation.

This link uses the nltk function provided by python to segment the historical dialogue. The specific implementation process is as follows:

Import original sentence samples

Import the nltk module and use the word segmentation function of the nltk module to segment each sentence to get a single word

Import the stop word list to remove the meaningless meaning and functional words such as prepositions and auxiliary verbs to generate a set of words that represent the meaning of the dialogue.

(2) Use coding technology to transcode the sentence after word segmentation to obtain transcoded samples.

Use encoding technology to convert each word in the word set into a numeric value, that is, use One Hot Encoder encoding technology to convert the string type of the word set into a numeric type, and convert each word into an unordered binary number to generate a one-to-one mapping set.

(3) Use the normalization method to normalize the transcoded samples to obtain the preprocessed samples.

In order to meet the data requirements of the subsequent model algorithm, the Normalizer algorithm is used to normalize the values in the mapping set, so that the sum of the values corresponding to each piece of data is 1, and a normalized and planned word mapping set is generated.

S12. Extract a sentence feature vector from the preprocessed sample.

Preferably, the extracting sentence feature vectors from the preprocessed sample includes:

Extracting text features from the preprocessed sample;

Using PCA technology to perform feature reduction on text features to obtain sentence feature vectors.

In an embodiment, preferably, said extracting text features from the preprocessed sample includes:

Extract text words from the preprocessed samples;

Use a clustering algorithm to cluster text words and select the cluster center as a main keyword.

Calculate the distance between other text words and the cluster center, and select the first N words closest to the cluster center as the text feature.

In one embodiment, the PAC technology is used to reduce the dimensionality of text features. The core of the technology is to calculate the appropriateness of the dimensionality using the percentage of variance, that is, to calculate how many dimensions the data set is reduced to.

S13. Train the sentence intention recognition model based on the sentence feature vector and use the cross-entropy cost function method to obtain a trained sentence intention recognition model.

Preferably, training a sentence intention recognition model based on the sentence feature vector and using a cross-entropy cost function method to obtain a trained sentence intention recognition model includes:

Use linear regression classifiers to classify sentence feature vectors and generate classification models for each category;

Integrate the classification models of each category to obtain the integrated classification model;

Use the LSTM deep neural network model to train the integrated classification model to obtain the trained classification model;

Use the cross-entropy cost function algorithm to optimize the trained classification model, and output the sentence intention recognition model.

This case uses the linear regression classifier adaptive enhancement algorithm in the Boosting algorithm. The core of the algorithm is an iterative algorithm. In each round of iteration, a new classifier is generated on the low-dimensional dialogue keyword set, and then the classifier is used for all Samples are classified to assess the importance of each sample (informative). Specifically, the algorithm will assign a weight to different categories of low-dimensional dialogue keyword sets. Each time the trained new classifier is used to label each low-dimensional dialogue keyword set sample, if a sample point has been classified correctly, its weight is reduced; if the sample point is not correctly classified, its weight is increased. The higher the weight of the sample, the greater the proportion of the next training, that is to say, the more difficult to distinguish the sample will become more and more important in the training process. The whole iterative process until the error rate is small enough or reaches a certain number of times.

Preferably, the training of the integrated classification model using the LSTM deep neural network model to obtain the trained classification model includes:

Convert the integrated classification model into a vector;

Based on the vector, forward calculating the output value of each neuron in the LSTM deep neural network model;

Backward calculation of the error term value of each neuron in the LSTM deep neural network model. The back propagation of the LSTM error term includes two directions: one is the back propagation along time, that is, starting from the current time t, calculate each time The error term of; one is to propagate the error term to the upper level;

According to the error term value of each neuron, iteratively calculate the gradient of each weight in the LSTM deep neural network model, until the iteration ends, and output the trained classification model.

When the LSTM deep neural network model is used in this application, when the output of the neuron is close to 1, the learning is slow. To solve this problem, the cross-entropy cost function algorithm is introduced in this case, and a weight update that does not include sigmoid is selected for the output layer.

S14. Obtain the target sentence to be recognized.

In this embodiment, the question sentence posed by the user is acquired as the target sentence.

S15. Based on the target sentence, and using the trained sentence intention recognition model, output the sentence intention corresponding to the target sentence.

In this embodiment, when a new machine question-and-answer dialogue occurs, the constructed sentence intention recognition model is used to quickly match the most suitable model to obtain a more accurate intention answer.

Based on the training and optimization of the deep learning model, the new question and answer sentences are input into the model. The model quickly performs various steps according to the context, and quickly returns appropriate answers to more accurately solve the user’s problem. Users get satisfactory answers quickly, saving users time.

S16. Obtain an answer that matches the intent of the sentence corresponding to the target sentence from the answer database, and display it to the user.

Preferably, obtaining an answer that matches the intent of the sentence corresponding to the target sentence from the answer database and displaying it to the user includes:

Obtaining a plurality of answers matching the intent of the sentence corresponding to the target sentence from the answer database;

Calculate the similarity between each matched answer and the user's intention;

According to the similarity, sort from largest to smallest, and display to users.

The application also provides a sentence intention recognition device. Referring to FIG. 2, it is a schematic diagram of the internal structure of a sentence intention recognition device provided by an embodiment of this application.

In this embodiment, the sentence intention recognition device 1 may be a personal computer (PC), or a terminal device such as a smart phone, a tablet computer, or a portable computer. The sentence intention recognition device 1 at least includes a memory 11, a processor 12, a communication bus 13, and a network interface 14.

Wherein, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc. The memory 11 may be an internal storage unit of the sentence intention recognition device 1 in some embodiments, for example, the hard disk of the sentence intention recognition device 1. In other embodiments, the memory 11 may also be an external storage device of the sentence intention recognition device 1, for example, a plug-in hard disk equipped on the sentence intention recognition device 1, a smart media card (SMC), and a secure digital (Secure Digital). Digital, SD) card, flash card (Flash Card), etc. Further, the memory 11 may also include both an internal storage unit of the sentence intention recognition apparatus 1 and an external storage device. The memory 11 can be used not only to store application software and various data installed in the sentence intention recognition device 1, such as the code of the sentence intention recognition program 01, etc., but also to temporarily store data that has been output or will be output.

The processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip in some embodiments, and is used to run the program code or processing stored in the memory 11 Data, such as execution statement intention recognition program 01, etc.

The communication bus 13 is used to realize the connection and communication between these components.

The network interface 14 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface), and is usually used to establish a communication connection between the device 1 and other electronic devices.

Optionally, the device 1 may also include a user interface. The user interface may include a display (Display) and an input unit such as a keyboard (Keyboard). The optional user interface may also include a standard wired interface and a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, an organic light-emitting diode (OLED) touch device, and the like. Among them, the display can also be appropriately called a display screen or a display unit, which is used to display the information processed in the sentence intention recognition device 1 and to display a visualized user interface.

FIG. 2 only shows the sentence intention recognition device 1 with components 11-14 and the sentence intention recognition program 01. Those skilled in the art will understand that the structure shown in FIG. 1 does not constitute a limitation on the sentence intention recognition device 1 It may include fewer or more components than shown, or a combination of some components, or a different component arrangement.

In the embodiment of the device 1 shown in FIG. 2, the sentence intention recognition program 01 is stored in the memory 11; when the processor 12 executes the sentence intention recognition program 01 stored in the memory 11, the following steps are implemented:

Get a sample of the original sentence.

Prepare the original sentence sample to obtain the preprocessed sample.

Import original sentence samples

Extract sentence feature vectors from the preprocessed samples.

Extracting text features from the preprocessed sample;

Extract text words from the preprocessed samples;

Based on the sentence feature vector and using the cross-entropy cost function method to train the sentence intention recognition model, a trained sentence intention recognition model is obtained.

Convert the integrated classification model into a vector;

Get the target sentence to be recognized.

Based on the target sentence, and using the trained sentence intention recognition model, the sentence intention corresponding to the target sentence is output.

Calculate the similarity between each matched answer and the user's intention;

Optionally, in other embodiments, the sentence intention recognition program can also be divided into one or more modules, and the one or more modules are stored in the memory 11 and run by one or more processors (in this embodiment, The processor 12) is executed to complete the application. The module referred to in the application refers to a series of computer program instruction segments capable of completing specific functions, and is used to describe the execution process of the sentence intention recognition program in the sentence intention recognition device.

For example, referring to FIG. 3, it is a schematic diagram of program modules of the sentence intention recognition program in an embodiment of the sentence intention recognition device of this application. In this embodiment, the sentence intention recognition program can be divided into an acquisition module 10 and a preprocessing module 20 , Extraction module 30, training module 40, output module 50 and display module 60, exemplarily:

The obtaining module 10 obtains an original sentence sample;

The preprocessing module 20 preprocesses original sentence samples to obtain preprocessed samples;

The extraction module 30 extracts sentence feature vectors from the preprocessed samples;

The training module 40 trains the sentence intention recognition model based on the sentence feature vector and uses the cross-entropy cost function method to obtain a trained sentence intention recognition model;

The obtaining module 10 obtains the target sentence to be recognized;

The output module 50 is based on the target sentence and uses the trained sentence intention recognition model to output the sentence intention corresponding to the target sentence;

The display module 60 obtains an answer that matches the intent of the sentence corresponding to the target sentence from the answer database, and displays it to the user.

The functions or operation steps implemented by the program modules such as the acquisition module 10, the preprocessing module 20, the extraction module 30, the training module 40, the output module 50, and the display module 60 are substantially the same as those in the foregoing embodiment, and will not be repeated here. .

In addition, an embodiment of the present application also proposes a computer-readable storage medium having a sentence intention recognition program stored on the computer-readable storage medium, and the sentence intention recognition program can be executed by one or more processors to achieve the following operating:

Obtain a sample of the original sentence;

Prepare the original sentence sample to obtain the preprocessed sample;

Extracting sentence feature vectors from the preprocessed samples;

Training a sentence intention recognition model by using the sentence feature vector to obtain a trained sentence intention recognition model;

Obtain the target sentence data to be recognized;

Based on the target sentence data, and using the trained sentence intention recognition model, output multiple speech texts with different probabilities corresponding to the target sentence data;

According to the multiple speech texts with different probabilities, the speech text with the greatest similarity is determined, and the speech text with the greatest similarity is used as the recognition result corresponding to the target sentence data.

The specific implementation of the computer-readable storage medium of the present application is basically the same as the above-mentioned sentence intention recognition device and method, and will not be repeated here.

It should be noted that the serial numbers of the above embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments. And the terms "include", "include" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements not only includes those elements, but also includes The other elements listed may also include elements inherent to the process, device, article, or method. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, device, article or method that includes the element.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。 Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM) as described above. , Magnetic disk, optical disk), including several instructions to make a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) execute the method described in each embodiment of the present application.

The above are only preferred embodiments of this application, and do not limit the scope of this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of this application, or directly or indirectly used in other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims

A method for recognizing sentence intention, characterized in that the method includes:

Obtain a sample of the original sentence;

Prepare the original sentence sample to obtain the preprocessed sample;

Extracting sentence feature vectors from the preprocessed samples;

Training a sentence intention recognition model based on the sentence feature vector and using a cross-entropy cost function method to obtain a trained sentence intention recognition model;

Get the target sentence to be recognized;

Based on the target sentence, and using the trained sentence intention recognition model, output the sentence intention corresponding to the target sentence;

An answer that matches the intent of the sentence corresponding to the target sentence is obtained from the answer database and displayed to the user.
The sentence intention recognition method according to claim 1, wherein the preprocessing the original sentence sample to obtain the preprocessed sample comprises:

Use natural language processing technology to segment the original sentence samples to obtain the sentence after word segmentation;

Use coding technology to transcode the sentence after word segmentation to obtain transcoded samples;

The normalization method is used to normalize the transcoded samples to obtain pre-processed samples.
The sentence intention recognition method according to claim 1, wherein said extracting sentence feature vectors from said preprocessed sample comprises:

Extracting text features from the preprocessed sample;

Using PCA technology to perform feature reduction on text features to obtain sentence feature vectors.
5. The sentence intention recognition method according to claim 3, wherein said extracting sentence feature vector from said preprocessed sample comprises:

Extracting text features from the preprocessed sample;

Using PCA technology to perform feature reduction on text features to obtain sentence feature vectors.
The sentence intention recognition method according to claim 1, wherein said training sentence intention recognition model based on said sentence feature vector and using a cross-entropy cost function method to obtain a trained sentence intention recognition model comprises:

Use linear regression classifiers to classify sentence feature vectors and generate classification models for each category;

Integrate the classification models of each category to obtain the integrated classification model;

Use the LSTM deep neural network model to train the integrated classification model to obtain the trained classification model;

Use the cross-entropy cost function algorithm to optimize the trained classification model, and output the sentence intention recognition model.
The sentence intention recognition method according to claim 5, wherein the training of the integrated classification model using the LSTM deep neural network model to obtain the trained classification model comprises:

Convert the integrated classification model into a vector;

Based on the vector, forward calculating the output value of each neuron in the LSTM deep neural network model;

Backward calculation of the error term value of each neuron in the LSTM deep neural network model. The back propagation of the LSTM error term includes two directions: one is the back propagation along time, that is, starting from the current time t, calculate each time The error term of; one is to propagate the error term to the upper level;

According to the error term value of each neuron, iteratively calculate the gradient of each weight in the LSTM deep neural network model, until the iteration ends, and output the trained classification model.
The sentence intention recognition method according to any one of claims 1 to 6, wherein obtaining an answer matching the sentence intention corresponding to the target sentence from an answer database and displaying it to the user comprises:

Obtaining a plurality of answers matching the intent of the sentence corresponding to the target sentence from the answer database;

Calculate the similarity between each matched answer and the user's intention;

According to the similarity, sort from largest to smallest, and display to users.
A sentence intention recognition device, characterized in that the device includes a memory and a processor, the memory stores a sentence intention recognition program that can run on the processor, and the sentence intention recognition program is processed by the processor. The following steps are implemented when the device is executed:

Obtaining step: obtaining a preset number of noisy speech and denoising speech corresponding to each noisy speech as training samples, and dividing the training samples into a first data set, a second data set, and a third data set;

Obtain a sample of the original sentence;

Prepare the original sentence sample to obtain the preprocessed sample;

Extracting sentence feature vectors from the preprocessed samples;

Training a sentence intention recognition model based on the sentence feature vector and using a cross-entropy cost function method to obtain a trained sentence intention recognition model;

Get the target sentence to be recognized;

Based on the target sentence, and using the trained sentence intention recognition model, output the sentence intention corresponding to the target sentence;

An answer that matches the intent of the sentence corresponding to the target sentence is obtained from the answer database and displayed to the user.
8. The sentence intention recognition device of claim 8, wherein the preprocessing the original sentence sample to obtain the preprocessed sample comprises:

Use natural language processing technology to segment the original sentence samples to obtain the sentence after word segmentation;

Use coding technology to transcode the sentence after word segmentation to obtain transcoded samples;

The normalization method is used to normalize the transcoded samples to obtain pre-processed samples.
8. The sentence intention recognition device according to claim 8, wherein said extracting sentence feature vectors from said preprocessed sample comprises:

Extracting text features from the preprocessed sample;

Using PCA technology to perform feature reduction on text features to obtain sentence feature vectors.
9. The sentence intention recognition device according to claim 10, wherein said extracting text features from said preprocessed sample comprises:

Extract text words from the preprocessed samples;

Use a clustering algorithm to cluster text words and select the cluster center as a main keyword.

Calculate the distance between other text words and the cluster center, and select the first N words closest to the cluster center as the text feature.
8. The sentence intention recognition device according to claim 8, wherein said training sentence intention recognition model based on said sentence feature vector and using a cross-entropy cost function method to obtain a trained sentence intention recognition model comprises:

Use linear regression classifiers to classify sentence feature vectors and generate classification models for each category;

Integrate the classification models of each category to obtain the integrated classification model;

Use the LSTM deep neural network model to train the integrated classification model to obtain the trained classification model;

Use the cross-entropy cost function algorithm to optimize the trained classification model, and output the sentence intention recognition model.
The sentence intention recognition device according to claim 12, wherein the training of the integrated classification model using the LSTM deep neural network model to obtain the trained classification model comprises:

Convert the integrated classification model into a vector;

Based on the vector, forward calculating the output value of each neuron in the LSTM deep neural network model;

Backward calculation of the error term value of each neuron in the LSTM deep neural network model. The back propagation of the LSTM error term includes two directions: one is the back propagation along time, that is, starting from the current time t, calculate each time The error term of; one is to propagate the error term to the upper level;

According to the error term value of each neuron, iteratively calculate the gradient of each weight in the LSTM deep neural network model, until the iteration ends, and output the trained classification model.
The sentence intention recognition device according to any one of claims 8 to 13, wherein obtaining an answer matching the sentence intention corresponding to the target sentence from an answer database and displaying it to the user comprises:

Obtaining a plurality of answers matching the intent of the sentence corresponding to the target sentence from the answer database;

Calculate the similarity between each matched answer and the user's intention;

According to the similarity, sort from largest to smallest, and display to users.
A computer-readable storage medium, wherein the computer-readable storage medium includes a sentence intention recognition program, and when the sentence intention recognition program is executed by a processor, the following steps are implemented:

Obtain a sample of the original sentence;

Prepare the original sentence sample to obtain the preprocessed sample;

Extracting sentence feature vectors from the preprocessed samples;

Training a sentence intention recognition model based on the sentence feature vector and using a cross-entropy cost function method to obtain a trained sentence intention recognition model;

Get the target sentence to be recognized;

Based on the target sentence, and using the trained sentence intention recognition model, output the sentence intention corresponding to the target sentence;

An answer that matches the intent of the sentence corresponding to the target sentence is obtained from the answer database and displayed to the user.
15. The computer-readable storage medium of claim 15, wherein the preprocessing the original sentence sample to obtain the preprocessed sample comprises:

Use natural language processing technology to segment the original sentence samples to obtain the sentence after word segmentation;

Use coding technology to transcode the sentence after word segmentation to obtain transcoded samples;

The normalization method is used to normalize the transcoded samples to obtain pre-processed samples.
15. The computer-readable storage medium of claim 15, wherein said extracting sentence feature vectors from said preprocessed sample comprises:

Extracting text features from the preprocessed sample;

Using PCA technology to perform feature reduction on text features to obtain sentence feature vectors.
17. The computer-readable storage medium of claim 17, wherein said extracting text features from said preprocessed sample comprises:

Extract text words from the preprocessed samples;

Use a clustering algorithm to cluster text words and select the cluster center as a main keyword.

Calculate the distance between other text words and the cluster center, and select the first N words closest to the cluster center as the text feature.
15. The computer-readable storage medium according to claim 15, wherein said training a sentence intention recognition model based on said sentence feature vector and using a cross-entropy cost function method to obtain a trained sentence intention recognition model comprises:

Use linear regression classifiers to classify sentence feature vectors and generate classification models for each category;

Integrate the classification models of each category to obtain the integrated classification model;

Use the LSTM deep neural network model to train the integrated classification model to obtain the trained classification model;

Use the cross-entropy cost function algorithm to optimize the trained classification model, and output the sentence intention recognition model.
The computer-readable storage medium of claim 19, wherein the training of the integrated classification model by using the LSTM deep neural network model to obtain the trained classification model comprises:

Convert the integrated classification model into a vector;

Based on the vector, forward calculating the output value of each neuron in the LSTM deep neural network model;

Backward calculation of the error term value of each neuron in the LSTM deep neural network model. The back propagation of the LSTM error term includes two directions: one is the back propagation along time, that is, starting from the current time t, calculate each time The error term of; one is to propagate the error term to the upper level;

According to the error term value of each neuron, iteratively calculate the gradient of each weight in the LSTM deep neural network model, until the iteration ends, and output the trained classification model.