WO2021051507A1

WO2021051507A1 - Bot conversation generation method, device, readable storage medium, and bot

Info

Publication number: WO2021051507A1
Application number: PCT/CN2019/116628
Authority: WO
Inventors: 于凤英; 王健宗
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-09-18
Filing date: 2019-11-08
Publication date: 2021-03-25
Also published as: CN110717022A

Abstract

A bot conversation generation method, a device, a non-volatile computer readable storage medium, and a bot, pertaining to the technical field of computers. The method, the device, the non-volatile computer readable storage medium, and the bot comprise: acquiring a first conversation sentence, performing word segmentation on the first conversation sentence, and obtaining words composing the first conversation sentence; searching a word vector database for respective word vectors of the words composing the first conversation sentence, and constructing an input vector sequence using the word vectors of the words composing the first conversation sentence; processing the input vector sequence by using a conversation generation model, and obtaining preferable conversation sentences and corresponding first output probabilities; calculating, according to the first output probabilities, a smoothness level of each of the preferable conversation sentences; and determining a preferable conversation sentence having the highest smoothness level as a second conversation sentence, and responding to the first conversation sentence by using the second conversation sentence. The invention improves clarity and smoothness of a conversation process.

Description

Method, device, readable storage medium and robot for generating robot dialogue

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on September 18, 2019, the application number is 201910880856.5, and the invention title is "a method, device, readable storage medium and robot for generating a robot dialogue", all of which The content is incorporated in this application by reference.

Technical field

This application belongs to the field of computer technology, and in particular relates to a method and device for generating a robot dialogue, a computer non-volatile readable storage medium, and a robot.

Background technique

With the continuous development of science and technology, dialogue robots are applied to more and more fields. These dialogue robots can communicate with users through voice or text, providing a basis for automated and intelligent user services. However, the conversations generated by the current robots often contain some uncomfortable sentences, and the user experience is poor.

technical problem

In view of this, the embodiments of the present application provide a method and device for generating a robot dialog, a computer non-volatile readable storage medium, and a robot, so as to solve the problem that some existing dialogs generated by current robots often contain some Problems with uncomfortable sentences and poor user experience.

Technical solutions

The first aspect of the embodiments of the present application provides a method for generating a robot dialogue, which may include:

Collect a first dialogue sentence, and perform word segmentation processing on the first dialogue sentence to obtain each word that composes the first dialogue sentence;

Query the word vector of each word constituting the first dialogue sentence in a preset word vector database, and construct the word vector of each word constituting the first dialogue sentence as an input vector sequence;

Use a preset dialogue generation model to process the input vector sequence to obtain each preferred dialogue sentence and the corresponding first output probability;

Respectively calculating the fluency of each preferred dialogue sentence according to the first output probability;

The preferred dialogue sentence with the highest fluent degree is determined as the second dialogue sentence, and the second dialogue sentence is used to respond to the first dialogue sentence.

The second aspect of the embodiments of the present application provides an apparatus for generating a robot dialogue, which may include:

The word segmentation processing module is configured to collect a first dialogue sentence, and perform word segmentation processing on the first dialogue sentence respectively to obtain each word that composes the first dialogue sentence;

The input vector sequence construction module is used to query the word vector of each word composing the first dialogue sentence in a preset word vector database, and construct the word vector of each word composing the first dialogue sentence as input Vector sequence

The dialogue generation module is used to process the input vector sequence using a preset dialogue generation model to obtain each preferred dialogue sentence and the corresponding first output probability;

The fluency calculation module is used to calculate the fluency of each preferred dialogue sentence according to the first output probability;

The sentence response module is used to determine the preferred dialogue sentence with the highest smoothness as the second dialogue sentence, and use the second dialogue sentence to respond to the first dialogue sentence.

A third aspect of the embodiments of the present application provides a computer non-volatile readable storage medium, the computer non-volatile readable storage medium stores computer readable instructions, and the computer readable instructions are executed by a processor When implementing the following steps:

The fourth aspect of the embodiments of the present application provides a robot, including a memory, a processor, and computer-readable instructions stored in the memory and running on the processor, and the processor executes the computer The following steps are implemented when reading instructions:

Beneficial effect

Compared with the prior art, the embodiment of the present application has the beneficial effect that: the embodiment of the present application first collects the first dialogue sentence, and performs word segmentation processing on the first dialogue sentence to obtain each of the first dialogue sentences. Then, in the preset word vector database, the word vectors of the words constituting the first dialogue sentence are respectively queried, and the word vectors of the words constituting the first dialogue sentence are constructed as an input vector sequence, and then, Use the preset dialogue generation model to process the input vector sequence to obtain each preferred dialogue sentence and the corresponding first output probability, and calculate the smoothness of each preferred dialogue sentence according to the first output probability. Finally, The preferred dialogue sentence with the highest smoothness is determined as the second dialogue sentence, and the second dialogue sentence is used to respond to the first dialogue sentence. In this way, only the most fluent dialogue sentences are output, and a large number of uncomfortable sentences are filtered out, making the entire dialogue process clearer and smoother, and greatly improving the user experience.

Description of the drawings

In order to more clearly describe the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only of the present application. For some embodiments, those of ordinary skill in the art can obtain other drawings based on these drawings without creative labor.

FIG. 1 is a flowchart of an embodiment of a method for generating a robot dialog in an embodiment of the application;

Figure 2 is a schematic flow chart of dividing the dialogue corpus into DN corpus sub-bases according to the dialogue scene generated by the dialogue sentence;

Figure 3 is a schematic diagram of the correspondence between each corpus and each model;

FIG. 4 is a structural diagram of an embodiment of a device for generating a robot dialog in an embodiment of the application;

Fig. 5 is a schematic block diagram of a robot in an embodiment of the application.

Embodiments of the present invention

In order to make the purposes, features, and advantages of the present application more obvious and understandable, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the following The described embodiments are only a part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

Referring to FIG. 1, an embodiment of a method for generating a robot dialogue in an embodiment of the present application may include:

Step S101: Collect a first dialogue sentence, and perform word segmentation processing on the first dialogue sentence, respectively, to obtain each word composing the first dialogue sentence.

The implementation subject of this application is a robot for dialogue with the user. The first dialogue sentence is a dialogue sentence expressed by the user through text or voice. The robot monitors the user's text input and voice input in real time, and when it detects a new text or voice input, it can collect it to form the first dialogue sentence. It should be noted that when the user enters in the form of text, the robot can directly collect these words to form the first dialogue sentence. When the user enters in the form of voice, the robot can perform the voice first. Convert to text, and then use the converted text to form the first dialogue sentence.

Word segmentation refers to dividing a dialogue sentence into individual words. In this embodiment, the sentence can be segmented according to a general dictionary to ensure that the separated words are all normal words. If the word is not in the dictionary, it will be separated. Single word. When both the forward and backward directions can be formed into words, such as "request for god", it will be divided according to the statistical word frequency. If the word frequency of "requirement" is high, the word "requirement/shen" is divided, and if the word frequency of "quest for god" is high, it is divided into "must" /Pray for God". After the word segmentation processing is performed on the first dialogue sentence, each word that composes the first dialogue sentence can be obtained.

Step S102: Query the word vectors of the words constituting the first dialogue sentence in a preset word vector database, and construct the word vectors of the words constituting the first dialogue sentence as an input vector sequence.

The word vector database is a database that records the correspondence between words and word vectors. The word vector may be a corresponding word vector obtained by training the word according to the word2vec model. That is, the probability of occurrence of the word is expressed according to the context information of the word. The training of word vectors is still based on the idea of word2vec. First, each word is represented as a 0-1 vector (one-hot) form, and then the word2vec model is trained with the word vector, and n-1 words are used to predict the nth word , The intermediate process obtained after the neural network model prediction is used as the word vector. Specifically, for example, the one-hot vector of "celebration" is assumed to be [1,0,0,0,...,0], and the one-hot vector of "meeting" is [0,1,0,0,... …,0], the one-hot vector for "smooth" is [0,0,1,0,……,0], the vector for predicting "closing" [0,0,0,1,……,0], The model is trained to generate the coefficient matrix W of the hidden layer. The product of the one-hot vector of each word and the coefficient matrix is the word vector of the word. The final form will be similar to "Celebrate [-0.28,0.34,-0.02, …...,0.92]" such a multi-dimensional vector. After the word vectors of the words composing the first dialog sentence are obtained by respectively querying, the word vectors of the words composing the first dialog sentence can be constructed in the form of a sequence, that is, the input vector sequence.

Step S103: Use a preset dialogue generation model to process the input vector sequence to obtain each preferred dialogue sentence and the corresponding first output probability.

The dialogue generation model may be selected from a preset model set. The model set includes DN models, and each model corresponds to a dialogue scene. These dialogue scenes include but are not limited to education, financial management, Parenting, news, etc. scenes.

In this embodiment, a dialogue corpus including a large number of dialogue sentences can be established in advance, and then the dialogue corpus is divided into DN corpus sub-bases according to the dialogue scenes generated by the dialogue sentences. All correspond to a dialogue scene. For example, as shown in Figure 2, the dialogue corpus can be divided into an educational corpus, a financial management corpus, a parenting corpus, a news corpus, and so on.

Since the same dialogue sentence may represent different meanings in different dialogue scenarios, if the same model is used to process the dialogue sentences generated in various dialogue scenarios, the accuracy is often low. Therefore, in this embodiment A corresponding model is set for each dialogue scene, and the model is trained using dialogue sentences in the corresponding corpus as a sample. As shown in Figure 3, Model 1 is a model corresponding to the dialogue scene of education, which is trained by using dialogue sentences in the educational corpus as a sample. Model 2 is a model corresponding to the dialogue scene of financial management. Then it is obtained by using the dialogue sentences in the financial management corpus as the sample training, and so on.

Before step S103, the dialogue scene of the first dialogue sentence may be determined first, and then a model corresponding to the dialogue scene of the first dialogue sentence may be selected from a preset model set as the dialogue generation model. The model selected in this way is more targeted, which can greatly improve the accuracy of the dialogue sentence generation.

The dialogue generation model may be an encoder-decoder (Encoder-Decoder) structure model, its input is a sequence, and the output is also a sequence. The encoder (Encoder) transforms a variable-length sequence into For a fixed-length vector expression, the decoder (Decoder) converts this fixed-length vector into a variable-length target sequence.

In this embodiment, multiple candidate dialogue sentences can be constructed in advance, wherein each candidate dialogue sentence corresponds to a permutation and combination of words in the dialogue corpus. Through the steps of word segmentation and word vector query similar to those in step S101 and step S102, the output vector sequence of each candidate dialogue sentence can be constructed respectively.

Here, the input vector sequence is denoted as:

Denote any one of the output vector sequences as:

Where Tx is the length of the input vector sequence, x ₁ is the first vector in the input vector sequence,

Is the last vector in the input vector sequence, and so on. Ty is the length of the output vector sequence, y ₁ is the first vector in the output vector sequence,

Is the last vector in the output vector sequence, and so on.

What the encoder receives at time t is the t-th vector in the input vector sequence and the hidden state of the encoder at time t-1, and the output is the hidden state of the encoder at time t, namely :

h _t ＝RNN _enc (x _t ,h _t-1 )

Where, x _t is the t-th vector in the input vector sequence, h _t is the hidden state of the encoder at time t, and RNN _enc is the RNN network model used by the encoder.

The decoder at time t receives the t-1th vector in the output vector sequence and the hidden state of the decoder at time t-1, and outputs the hidden state of the decoder at time t ,which is:

s _t =RNN _dec (y _t-1 ,s _t-1 )

Wherein, y _t is the t-th vector in the output vector sequence, _st is the hidden state of the decoder at time t, and RNN _dec is the RNN network model used by the decoder.

On this basis, the score between the hidden state of the encoder at each time and the hidden state of the decoder at each time can be calculated according to the following formula:

e _ij = score(s _i ,h _j )

Where e _ij is the score between the hidden state of the encoder at time j and the hidden state of the decoder at time i, and score is a preset score calculation function, including but not limited to the commonly used dot function, general function and concat function.

Then, the weight corresponding _{to e ij} can be calculated according to the following formula:

Among them, exp is the natural exponential function, and α _{ij is} the weight corresponding to e _ij.

Next, calculate the output probability of the t-th vector in the output vector sequence according to the following formula:

Where c _i and

Is the intermediate variable in the calculation process, tanh is the hyperbolic tangent function, softmax is the normalized exponential function, W _c and W _s are the model parameters obtained after training, p(y _t |y _＜t ,x) is p The abbreviation of (y _t |y ₁ ,y ₂ ,…,y _t-1 ,x), that is, the input vector sequence is x, and the first t-1 vectors in the output vector sequence are y ₁ ,y ₂ ,… , _{Under the condition of y t-1} , the probability that the t-th vector in the output vector sequence is y _t.

Finally, the output probability of the output vector sequence can be calculated according to the following formula:

According to the above method, the output probability of each candidate dialogue sentence (that is, the output probability of the corresponding output vector sequence) is calculated, and the first N candidate dialogue sentences with the largest output probability are selected as the preferred dialogue sentences, and N is greater than 2. Integer. The output probability of each preferred dialogue sentence is the first output probability.

Step S104: Calculate the fluency of each preferred dialogue sentence according to the first output probability.

First, the output probability of each preferred dialogue sentence in the preset reference model can be calculated separately, that is, the second output probability. The reference model may be a unigram model. In the unigram model, it is assumed that the words in the sentence are independently exchangeable, and the order information of the words is irrelevant. Under such a premise, the probability of each word in each preferred dialogue sentence appearing in the dialogue corpus can be separately counted.

In a specific implementation of this embodiment, after the dialogue corpus has been divided into DN corpora, a preferred corpus can be selected from the dialogue corpus, and the preferred corpus is the same as the The corpus sub-base corresponding to the dialogue scene of the first dialogue sentence is then separately counted for the probability of each word in each preferred dialogue sentence appearing in the preferred corpus sub-base. That is, the probability of each occurrence in the preferred corpus sub-base is used to replace the probability of respective occurrence in the entire corpus of the corpus. After that, the second output probability of each preferred dialogue sentence can be calculated separately according to the following formula:

Where m is the sequence number of each word, 1≤m≤|S _n |, w _n,m is the mth word in the nth preferred dialogue sentence, and p(w _n,m ) is the nth preferred dialogue sentence The probability of the m-th word in, respectively appearing in the dialogue corpus, P _u (S _n ) is the second output probability of the n-th preferred dialogue sentence.

Then, the smoothness of each preferred dialogue sentence can be calculated according to the following formula:

Among them, n is the serial number of each preferred dialogue sentence, 1≤n≤N, N is the number of preferred dialogue sentences, S _n is the nth preferred dialogue sentence, |S _n | is the length of the nth preferred dialogue sentence, P _m (S _n ) is the first output probability of the nth preferred dialogue sentence, ln is the natural logarithmic function, and SLOR(S _n ) is the smoothness of the nth preferred dialogue sentence.

The core of this formula is to estimate the smoothness through the output probability of the sentence, which is also consistent with the natural law of language use: any kind of dialogue generation model is essentially based on the corpus training used in mass daily life. It is obtained that the greater the output probability of a sentence in the model, the more widely it is used in people’s daily life. Its use conforms to people’s language habits (that is, the higher the smoothness), and a sentence is in the model. The smaller the output probability in, it means that it is hardly used in people's daily life, and its use is not in line with people's language habits (that is, the fluency is lower).

The reason for removing the second output probability of the sentence in the unigram model is to avoid the influence of the probability of a single word in the dialogue corpus on the smoothness of the entire sentence. For example, the following two statements:

Statement 1: "I come from China"

Statement 2: "I am from Turkmenistan"

These two sentences have the same length, and the probability of "China" appearing in the dialogue corpus is greater than that of "Turkmenistan", then sentence 1 has a greater output probability than sentence 2. But in fact, the smoothness of these two sentences is the same, and removing the second output probability can effectively avoid such problems.

Step S105: Determine the preferred dialogue sentence with the highest smoothness as the second dialogue sentence, and use the second dialogue sentence to respond to the first dialogue sentence.

The second dialogue sentence is a dialogue sentence expressed by the robot through text or voice. It should be noted that when the robot responds in the form of text, it can directly use the second dialogue sentence to respond. When the robot responds in the form of voice, the text-to-speech response can be performed first. Convert, and then respond with the converted voice.

In summary, the embodiment of the present application first collects the first dialogue sentence, and performs word segmentation processing on the first dialogue sentence to obtain each word that composes the first dialogue sentence, and then stores it in a preset word vector database Query the word vectors of the words composing the first dialog sentence respectively, and construct the word vectors of the words composing the first dialog sentence as an input vector sequence, and then use a preset dialog generation model to input the input vector. The vector sequence is processed to obtain each preferred dialogue sentence and the corresponding first output probability, and the smoothness of each preferred dialogue sentence is calculated according to the first output probability. Finally, the preferred dialogue sentence with the highest smoothness is determined as the second Dialogue sentence, and use the second dialogue sentence to respond to the first dialogue sentence. In this way, only the most fluent dialogue sentences are output, and a large number of uncomfortable sentences are filtered out, making the entire dialogue process clearer and smoother, and greatly improving the user experience.

It should be understood that the size of the sequence number of each step in the foregoing embodiment does not mean the order of execution. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

Corresponding to the method for generating a robot dialog described in the above embodiment, FIG. 4 shows a structural diagram of an embodiment of a device for generating a robot dialog provided by an embodiment of the present application.

In this embodiment, an apparatus for generating a robot dialog may include:

The word segmentation processing module 401 is configured to collect a first dialogue sentence, and perform word segmentation processing on the first dialogue sentence respectively to obtain each word constituting the first dialogue sentence;

The input vector sequence construction module 402 is used to query the word vector of each word composing the first dialogue sentence in a preset word vector database, and construct the word vector of each word composing the first dialogue sentence as Input vector sequence;

The dialogue generation module 403 is configured to process the input vector sequence using a preset dialogue generation model to obtain each preferred dialogue sentence and the corresponding first output probability;

The fluency calculation module 404 is configured to calculate the fluency of each preferred dialog sentence according to the first output probability;

The sentence response module 405 is configured to determine the preferred dialogue sentence with the highest smoothness as the second dialogue sentence, and use the second dialogue sentence to respond to the first dialogue sentence.

Further, the compliance calculation module may include:

The output probability calculation sub-module is used to calculate the second output probability of each preferred dialogue sentence in the preset reference model;

The fluency calculation sub-module is used to calculate the fluency of each preferred dialogue sentence according to the following formula:

Among them, n is the serial number of each preferred dialogue sentence, 1≤n≤N, N is the number of preferred dialogue sentences, S _n is the nth preferred dialogue sentence, |S _n | is the length of the nth preferred dialogue sentence, P _m (S _n ) is the first output probability of the nth preferred dialogue sentence, P _u (S _n ) is the second output probability of the nth preferred dialogue sentence, ln is the natural logarithmic function, and SLOR(S _n ) is The smoothness of the nth preferred dialogue sentence.

Further, the output probability calculation sub-module may include:

The probability statistics unit is used to separately count the probability of each word in each preferred dialogue sentence appearing in the preset dialogue corpus;

The output probability calculation unit is used to calculate the second output probability of each preferred dialogue sentence according to the following formula:

Where m is the sequence number of each word, 1≤m≤|S _n |, w _n,m is the mth word in the nth preferred dialogue sentence, and p(w _n,m ) is the nth preferred dialogue sentence The probability that the m-th word in, respectively appears in the dialogue corpus.

Further, the device for generating a robot dialogue may further include:

A dialogue scene determination module, configured to determine the dialogue scene of the first dialogue sentence;

The dialogue generation model selection module is used to select a model corresponding to the dialogue scene of the first dialogue sentence from a preset model set as the dialogue generation model. The model set includes DN models, each of which is Corresponds to a dialogue scene.

Further, the probability statistics unit may include:

The corpus division sub-unit is used to divide the preset dialogue corpus into DN corpus sub-bases, where each corpus sub-base corresponds to a dialogue scene;

A preferred corpus selection subunit for selecting a preferred corpus from the dialogue corpus, where the preferred corpus is a corpus corresponding to the dialogue scene of the first dialogue sentence;

The probability statistics subunit is used to separately count the probability of each word in each preferred dialogue sentence appearing in the preferred corpus sub-base.

Those skilled in the art can clearly understand that, for the convenience and conciseness of the description, the specific working processes of the above described devices, modules and units can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

In the above-mentioned embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail or recorded in an embodiment, reference may be made to related descriptions of other embodiments.

Fig. 5 shows a schematic block diagram of a robot provided by an embodiment of the present application. For ease of description, only parts related to the embodiment of the present application are shown.

In this embodiment, the robot 5 may include: a processor 50, a memory 51, and computer-readable instructions 52 stored in the memory 51 and executable on the processor 50, such as executing the aforementioned robot dialogue generation Computer readable instructions for the method. When the processor 50 executes the computer-readable instructions 52, the steps in the above embodiments of the robot dialog generation method are implemented, for example, steps S101 to S105 shown in FIG. 1. Alternatively, when the processor 50 executes the computer-readable instructions 52, the functions of the modules/units in the foregoing device embodiments, such as the functions of the modules 401 to 405 shown in FIG. 4, are implemented.

Exemplarily, the computer-readable instructions 52 may be divided into one or more modules/units, and the one or more modules/units are stored in the memory 51 and executed by the processor 50, To complete this application. The one or more modules/units may be a series of computer-readable instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer-readable instructions 52 in the robot 5.

The processor 50 may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.

The memory 51 may be an internal storage unit of the robot 5, such as a hard disk or a memory of the robot 5. The memory 51 may also be an external storage device of the robot 5, such as a plug-in hard disk equipped on the robot 5, a smart memory card (Smart Media Card, SMC), or a Secure Digital (SD) card, Flash Card, etc. Further, the memory 51 may also include both an internal storage unit of the robot 5 and an external storage device. The memory 51 is used to store the computer-readable instructions and other instructions and data required by the robot 5. The memory 51 can also be used to temporarily store data that has been output or will be output.

A person of ordinary skill in the art can understand that all or part of the processes in the method of the above-mentioned embodiments can be implemented by instructing relevant hardware through computer-readable instructions. The computer-readable instructions can be stored in a non-volatile computer. In a readable storage medium, when the computer-readable instructions are executed, they may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, a person of ordinary skill in the art should understand that it can still implement the foregoing The technical solutions recorded in the examples are modified, or some of the technical features are equivalently replaced; these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

A method for generating a robot dialogue, which is characterized in that it comprises:

Collect a first dialogue sentence, and perform word segmentation processing on the first dialogue sentence to obtain each word that composes the first dialogue sentence;

Query the word vector of each word constituting the first dialogue sentence in a preset word vector database, and construct the word vector of each word constituting the first dialogue sentence as an input vector sequence;

Use a preset dialogue generation model to process the input vector sequence to obtain each preferred dialogue sentence and the corresponding first output probability;

Respectively calculating the fluency of each preferred dialogue sentence according to the first output probability;

The preferred dialogue sentence with the highest fluent degree is determined as the second dialogue sentence, and the second dialogue sentence is used to respond to the first dialogue sentence.
The method for generating a robot dialogue according to claim 1, wherein the calculating the smoothness of each preferred dialogue sentence according to the first output probability comprises:

Respectively calculate the second output probability of each preferred dialogue sentence in the preset benchmark model;

Calculate the fluency of each preferred dialogue sentence according to the following formula:

Among them, n is the serial number of each preferred dialogue sentence, 1≤n≤N, N is the number of preferred dialogue sentences, S n is the nth preferred dialogue sentence, |S n | is the length of the nth preferred dialogue sentence, P m (S n ) is the first output probability of the nth preferred dialogue sentence, P u (S n ) is the second output probability of the nth preferred dialogue sentence, ln is the natural logarithmic function, and SLOR(S n ) is The smoothness of the nth preferred dialogue sentence.
The method for generating a robot dialogue according to claim 2, wherein said calculating the second output probability of each preferred dialogue sentence in a preset reference model respectively comprises:

Respectively count the probability of each word in each preferred dialogue sentence appearing in the preset dialogue database;

Calculate the second output probability of each preferred dialogue sentence according to the following formula:

Where m is the sequence number of each word, 1≤m≤|S n |, w n,m is the mth word in the nth preferred dialogue sentence, and p(w n,m ) is the nth preferred dialogue sentence The probability of the m-th word in, respectively appearing in the dialogue corpus.
The method for generating a robot dialog according to claim 3, characterized in that, before using a preset dialog generation model to process the input vector sequence, the method further comprises:

Determine the dialogue scene of the first dialogue sentence;

A model corresponding to the dialogue scene of the first dialogue sentence is selected from a preset model set as the dialogue generation model. The model set includes DN models, and each model corresponds to a dialogue scene.
The method for generating a robot dialogue according to claim 4, wherein said separately counting the probability of each word in each preferred dialogue sentence appearing in a preset dialogue corpus comprises:

Divide the preset dialogue corpus into DN corpus sub-bases, where each corpus sub-base corresponds to a dialogue scene;

Selecting a preferred corpus sub-base from the dialogue corpus, where the preferred corpus sub-base is a corpus sub-base corresponding to the dialogue scene of the first dialogue sentence;

The probability of each word in each preferred dialogue sentence appearing in the preferred corpus is separately counted.
A device for generating a robot dialogue, which is characterized in that it comprises:

The word segmentation processing module is configured to collect a first dialogue sentence, and perform word segmentation processing on the first dialogue sentence respectively to obtain each word that composes the first dialogue sentence;

The input vector sequence construction module is used to query the word vector of each word composing the first dialogue sentence in a preset word vector database, and construct the word vector of each word composing the first dialogue sentence as input Vector sequence

The dialogue generation module is used to process the input vector sequence using a preset dialogue generation model to obtain each preferred dialogue sentence and the corresponding first output probability;

The fluency calculation module is used to calculate the fluency of each preferred dialogue sentence according to the first output probability;

The sentence response module is used to determine the preferred dialogue sentence with the highest smoothness as the second dialogue sentence, and use the second dialogue sentence to respond to the first dialogue sentence.
8. The robot dialog generating device according to claim 6, wherein the smoothness calculation module comprises:

The output probability calculation sub-module is used to calculate the second output probability of each preferred dialogue sentence in the preset reference model;

The fluency calculation sub-module is used to calculate the fluency of each preferred dialogue sentence according to the following formula:

Among them, n is the serial number of each preferred dialogue sentence, 1≤n≤N, N is the number of preferred dialogue sentences, S n is the nth preferred dialogue sentence, |S n | is the length of the nth preferred dialogue sentence, P m (S n ) is the first output probability of the nth preferred dialogue sentence, P u (S n ) is the second output probability of the nth preferred dialogue sentence, ln is the natural logarithmic function, and SLOR(S n ) is The smoothness of the nth preferred dialogue sentence.
8. The robot dialog generating device according to claim 7, wherein the output probability calculation sub-module comprises:

The probability statistics unit is used to separately count the probability of each word in each preferred dialogue sentence appearing in the preset dialogue corpus;

The output probability calculation unit is used to calculate the second output probability of each preferred dialogue sentence according to the following formula:

Where m is the sequence number of each word, 1≤m≤|S n |, w n,m is the mth word in the nth preferred dialogue sentence, and p(w n,m ) is the nth preferred dialogue sentence The probability that the m-th word in, respectively appears in the dialogue corpus.
8. The robot dialog generating device according to claim 8, wherein the robot dialog generating device further comprises:

A dialogue scene determination module, configured to determine the dialogue scene of the first dialogue sentence;

The dialogue generation model selection module is used to select a model corresponding to the dialogue scene of the first dialogue sentence from a preset model set as the dialogue generation model. The model set includes DN models, each of which is Corresponds to a dialogue scene.
The robot dialogue generating device according to claim 9, wherein the probability statistics unit comprises:

The corpus division sub-unit is used to divide the preset dialogue corpus into DN corpus sub-bases, where each corpus sub-base corresponds to a dialogue scene;

The preferred corpus selection subunit is used to select a preferred corpus from the dialogue corpus, where the preferred corpus is a corpus corresponding to the dialogue scene of the first dialogue sentence;

The probability statistics subunit is used to separately count the probability of each word in each preferred dialogue sentence appearing in the preferred corpus sub-base.
A computer non-volatile readable storage medium, the computer non-volatile readable storage medium storing computer readable instructions, wherein the computer readable instructions are executed by a processor to implement the following steps:

Collecting a first dialogue sentence, and performing word segmentation processing on the first dialogue sentence to obtain each word that composes the first dialogue sentence;

Query the word vector of each word constituting the first dialogue sentence in a preset word vector database, and construct the word vector of each word constituting the first dialogue sentence as an input vector sequence;

Use a preset dialogue generation model to process the input vector sequence to obtain each preferred dialogue sentence and the corresponding first output probability;

Respectively calculating the fluency of each preferred dialogue sentence according to the first output probability;

The preferred dialogue sentence with the highest fluent degree is determined as the second dialogue sentence, and the second dialogue sentence is used to respond to the first dialogue sentence.
11. The computer non-volatile readable storage medium according to claim 11, wherein said calculating the fluency of each preferred dialog sentence according to the first output probability comprises:

Respectively calculate the second output probability of each preferred dialogue sentence in the preset benchmark model;

Calculate the fluency of each preferred dialogue sentence according to the following formula:

Among them, n is the serial number of each preferred dialogue sentence, 1≤n≤N, N is the number of preferred dialogue sentences, S n is the nth preferred dialogue sentence, |S n | is the length of the nth preferred dialogue sentence, P m (S n ) is the first output probability of the nth preferred dialogue sentence, P u (S n ) is the second output probability of the nth preferred dialogue sentence, ln is the natural logarithmic function, and SLOR(S n ) is The smoothness of the nth preferred dialogue sentence.
The computer non-volatile readable storage medium according to claim 12, wherein said separately calculating the second output probability of each preferred dialogue sentence in a preset reference model comprises:

Respectively count the probability of each word in each preferred dialogue sentence appearing in the preset dialogue database;

Calculate the second output probability of each preferred dialogue sentence according to the following formula:

Where m is the sequence number of each word, 1≤m≤|S n |, w n,m is the mth word in the nth preferred dialogue sentence, and p(w n,m ) is the nth preferred dialogue sentence The probability of the m-th word in, respectively appearing in the dialogue corpus.
The computer non-volatile readable storage medium according to claim 13, wherein before using a preset dialogue generation model to process the input vector sequence, the method further comprises:

Determine the dialogue scene of the first dialogue sentence;

A model corresponding to the dialogue scene of the first dialogue sentence is selected from a preset model set as the dialogue generation model. The model set includes DN models, and each model corresponds to a dialogue scene.
The computer non-volatile readable storage medium according to claim 14, wherein said separately counting the probability of each word in each preferred dialogue sentence appearing in a preset dialogue corpus comprises:

Divide the preset dialogue corpus into DN corpus sub-bases, where each corpus sub-base corresponds to a dialogue scene;

Selecting a preferred corpus sub-base from the dialogue corpus, where the preferred corpus sub-base is a corpus sub-base corresponding to the dialogue scene of the first dialogue sentence;

The probability of each word in each preferred dialogue sentence appearing in the preferred corpus is separately counted.
A robot comprising a memory, a processor, and computer readable instructions stored in the memory and capable of running on the processor, wherein the processor executes the computer readable instructions to implement the following steps :

Collect a first dialogue sentence, and perform word segmentation processing on the first dialogue sentence to obtain each word that composes the first dialogue sentence;

Query the word vector of each word constituting the first dialogue sentence in a preset word vector database, and construct the word vector of each word constituting the first dialogue sentence as an input vector sequence;

Use a preset dialogue generation model to process the input vector sequence to obtain each preferred dialogue sentence and the corresponding first output probability;

Respectively calculating the fluency of each preferred dialogue sentence according to the first output probability;

The preferred dialogue sentence with the highest fluent degree is determined as the second dialogue sentence, and the second dialogue sentence is used to respond to the first dialogue sentence.
16. The robot according to claim 16, wherein the calculation of the smoothness of each preferred dialogue sentence according to the first output probability comprises:

Respectively calculate the second output probability of each preferred dialogue sentence in the preset benchmark model;

Calculate the fluency of each preferred dialogue sentence according to the following formula:

Among them, n is the serial number of each preferred dialogue sentence, 1≤n≤N, N is the number of preferred dialogue sentences, S n is the nth preferred dialogue sentence, |S n | is the length of the nth preferred dialogue sentence, P m (S n ) is the first output probability of the nth preferred dialogue sentence, P u (S n ) is the second output probability of the nth preferred dialogue sentence, ln is the natural logarithmic function, and SLOR(S n ) is The smoothness of the nth preferred dialogue sentence.
18. The robot according to claim 17, wherein said separately calculating the second output probability of each preferred dialogue sentence in a preset reference model comprises:

Respectively count the probability of each word in each preferred dialogue sentence appearing in the preset dialogue database;

Calculate the second output probability of each preferred dialogue sentence according to the following formula:

Where m is the sequence number of each word, 1≤m≤|S n |, w n,m is the mth word in the nth preferred dialogue sentence, and p(w n,m ) is the nth preferred dialogue sentence The probability of the m-th word in, respectively appearing in the dialogue corpus.
The robot according to claim 18, characterized in that, before using a preset dialogue generation model to process the input vector sequence, it further comprises:

Determine the dialogue scene of the first dialogue sentence;

A model corresponding to the dialogue scene of the first dialogue sentence is selected from a preset model set as the dialogue generation model. The model set includes DN models, and each model corresponds to a dialogue scene.
The robot according to claim 19, wherein said separately counting the probability of each word in each preferred dialogue sentence appearing in a preset dialogue corpus comprises:

Divide the preset dialogue corpus into DN corpus sub-bases, where each corpus sub-base corresponds to a dialogue scene;

Selecting a preferred corpus sub-base from the dialogue corpus, where the preferred corpus sub-base is a corpus sub-base corresponding to the dialogue scene of the first dialogue sentence;

The probability of each word in each preferred dialogue sentence appearing in the preferred corpus is separately counted.