JP2023078411A - Information processing method, model training method, apparatus, appliance, medium and program product - Google Patents

Abstract

To provide an information processing method, model training method, apparatus, appliance, medium and program product adapted to acquire a target response sentence having high interaction quality.SOLUTION: An information processing method acquires an interaction model having high interaction accuracy and a target response sentence having high interaction quality, in which the interaction model is acquired by performing training based on a corrected response-sample sentence, a second candidate response-sample sentence and a recall response-sample sentence and a plurality of candidate response-sample sentences are acquired by inputting an initial interaction-sample sentence to an initial interaction model. The second candidate response-sample sentence is any of the plurality of candidate response-sample sentences, the corrected response-sample sentence is a sentence having high interaction quality acquired by correcting a first response-sample sentence out of the candidate response-sample sentences, and the recall response-sample sentence is another sample sentence except for the initial interaction-sample sentence and the plurality of candidate response-sample sentences out of the training sample sentences.SELECTED DRAWING: Figure 2

Description

The present disclosure relates to the field of computer technology, in particular to the field of artificial intelligence and speech technology, and specifically to information processing methods, model training methods, devices, equipment, media and program products.

With the development of natural language processing technology, the machine learning model can be used in the field of smart dialogue, and the dialogue model responds based on the sentences input by the user to achieve the effect of interacting with the user.

Currently, the dialogue model has low dialogue accuracy and poor dialogue quality.

The present disclosure provides information processing methods, model training methods, devices, apparatus, media and program products.

According to one aspect of the present disclosure, an information processing method is provided, the method comprising:
obtaining an initial dialogue;
inputting the initial dialogue sentence into a trained dialogue model to obtain a target response sentence;
The dialogue model is a model obtained by training based on a modified reply sample sentence, a second candidate reply sample sentence and a recall reply sample sentence, and an initial dialogue sample sentence is input to the initial dialogue model to generate a plurality of candidates. obtaining a reply sample sentence, wherein the second candidate reply sample sentence is one of the plurality of candidate reply sample sentences, and the modified reply sample sentence is a first reply sample among the candidate reply sample sentences; It is a sentence obtained by correcting a sentence, and the recall response sample sentence is a sample sentence other than the initial dialogue sample sentence and the plurality of candidate response sample sentences among the training sample sentences.

According to another aspect of the disclosure, there is provided a model training method, the method comprising:
obtaining an initial dialog sample sentence;
inputting the initial dialogue sample sentences into an initial dialogue model to obtain a plurality of candidate reply sample sentences;
modifying a first candidate response sample sentence among the plurality of candidate response sample sentences to obtain a modified response sample sentence;
training the initial dialogue model based on the modified sample response sentence, a second candidate response sample sentence of the plurality of candidate response sample sentences, and a recall sample response sentence to obtain a dialogue model;
The recall response sample sentences are sample sentences other than the initial dialogue sample sentences and the plurality of candidate response sample sentences among the training sample sentences.

According to another aspect of the present disclosure, an information processing device is provided, the device comprising:
a retrieving module for retrieving an initial dialogue statement;
an input module for inputting the initial dialogue sentence into a trained dialogue model to obtain a target response sentence;
The dialogue model is a model obtained by training based on a modified reply sample sentence, a second candidate reply sample sentence and a recall reply sample sentence, and an initial dialogue sample sentence is input to the initial dialogue model to generate a plurality of candidates. obtaining a reply sample sentence, wherein the second candidate reply sample sentence is one of the plurality of candidate reply sample sentences, and the modified reply sample sentence is a first reply sample among the candidate reply sample sentences; It is a sentence obtained by correcting a sentence, and the recall response sample sentence is a sample sentence other than the initial dialogue sample sentence and the plurality of candidate response sample sentences among the training sample sentences.

According to another aspect of the disclosure, there is provided a model training device, the device comprising:
a sentence acquisition module for acquiring initial dialogue sample sentences;
a sentence input module for inputting the initial dialogue sample sentences into an initial dialogue model to obtain a plurality of candidate reply sample sentences;
a correction module for correcting a first candidate reply sample sentence among the plurality of candidate reply sample sentences to obtain a corrected reply sample sentence;
a training module for training the initial dialogue model based on the modified sample response sentence, a second candidate response sample sentence among the plurality of candidate response sample sentences, and the recall sample response sentence to obtain a dialogue model; including
The recall response sample sentences are sample sentences other than the initial dialogue sample sentences and the plurality of candidate response sample sentences among the training sample sentences.

According to another aspect of the present disclosure, an electronic device is provided, the electronic device comprising:
at least one processor;
a memory communicatively coupled to the at least one processor;
The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor such that the at least one processor can perform the above method.

According to another aspect of the present disclosure, a non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions causing a computer to perform the above method.

According to another aspect of the present disclosure, a computer program, said computer program realizing the steps of the above method when executed by a processor.

Some embodiments of the present disclosure perform training based on the modified response sample sentences, the second candidate response sample sentences, and the recall response sample sentences to obtain an interaction model, and input the initial interaction sample sentences into the initial interaction model. to obtain a plurality of candidate reply sample sentences, the second candidate reply sample sentence is one of the plurality of candidate reply sample sentences, and the modified reply sample sentence is the first reply among the candidate reply sample sentences The recall response sample sentences are the sample sentences of the training sample sentences other than the initial dialogue sample sentences and the multiple candidate response sample sentences. , modified sample response sentences, second candidate response sample sentences, and recall response sample sentences, an initial dialogue model is continuously trained to obtain a highly accurate dialogue model, and the initial dialogue sentence is used as the dialogue model. Type to get a dialogue-quality target response.

It should be understood that the description in this section is not intended to identify key or critical features of embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. . Other features of the present disclosure will be readily understood from the following description.

The drawings are used for better understanding of the present technical solution and do not limit the present disclosure.
1 is a schematic flow chart of an information processing method provided by Embodiment 1 of the present disclosure; 2 is a schematic flow chart of a model training method provided by Embodiment 2 of the present disclosure; 3 is a flow chart of an information processing method provided by Embodiment 3 of the present disclosure; 1 is a schematic configuration diagram of an information processing device provided by an exemplary embodiment of the present disclosure; FIG. 1 is a schematic block diagram of a model training device provided by an exemplary embodiment of the present disclosure; FIG. 1 is a schematic block diagram of an exemplary electronic device for implementing embodiments of the present disclosure; FIG.

Illustrative embodiments of the present disclosure will now be described in conjunction with the drawings, in which various details of the embodiments of the present disclosure are included for ease of understanding and are merely exemplary. should be regarded as Accordingly, those skilled in the art should appreciate that various changes and modifications can be made to the examples described herein without departing from the scope and spirit of this disclosure. Similarly, for the sake of clarity and brevity, the following description omits descriptions of well-known functions and constructions.

In addition, in the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, disclosure, and other processing of the relevant user personal information shall comply with the provisions of relevant laws and regulations, and shall not violate public order and morals. .

Artificial intelligence is a field that studies how computers simulate certain human thought processes and intelligent actions (learning, reasoning, thinking, planning, etc.), and includes both hardware-level technology and software-level technology. Artificial intelligence hardware technology generally includes sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech recognition technology, natural language processing technology and machine learning/deep learning, big data processing technology, knowledge graph technology and other directions.

With the development of natural language processing technology, the machine learning model can be used in the field of smart dialogue, and the dialogue model responds based on the sentences input by the user to achieve the effect of interacting with the user.

In the field of dialogue systems, large-scale dialogue models trained on social media commentary data continue to emerge. However, due to the gap between the social media commentary scene and the actual human interaction scene, the model's generative ability is not good.

Generative interaction models generate multiple candidate responses during reasoning and then use the generated scores to evaluate and sort the responses. However, sorting methods based on generated scores do not effectively put high-quality responses in the foreground.

Currently, the dialogue model has low dialogue accuracy and poor dialogue quality.

In view of the above technical problems, some embodiments of the present disclosure obtain a dialogue model by training based on the modified reply sample sentence, the second candidate reply sample sentence and the recall reply sample sentence, and An initial dialogue sample sentence is input to the initial dialogue model to obtain a plurality of candidate reply sample sentences, a second candidate reply sample sentence is any one of the plurality of candidate reply sample sentences, and a modified reply sample sentence is a candidate A recall sample sentence is a high-quality dialogue sentence obtained by correcting the first response sample sentence among the response sample sentences. By continuing to train the initial dialogue model for the sample sentences other than the sentences, the modified response sample sentences, the second candidate response sample sentences, and the recall response sample sentences, a dialogue model with high accuracy of the dialogue is obtained. and input the initial dialogue sentences into the dialogue model to obtain target response sentences with high dialogue quality.

The technical solutions provided by each embodiment of the present disclosure will be described in detail below in conjunction with the drawings.

FIG. 1 is a schematic flowchart of an information processing method provided by Embodiment 1 of the present disclosure. As shown in FIG. 1, the method includes the following steps S101-102.

S101, obtaining an initial dialogue sentence.
S102, Input the initial dialogue sentence into the trained dialogue model to obtain the target reply sentence.
The dialogue model is a model obtained by training based on the modified reply sample sentence, the second candidate reply sample sentence, and the recall reply sample sentence, and the initial dialogue sample sentence is input to the initial dialogue model to generate a plurality of candidate answers. Obtaining sample sentences, the second candidate response sample sentence is one of a plurality of candidate response sample sentences, and the modified response sample sentence is obtained by modifying the first response sample sentence among the candidate response sample sentences. The recall response sample sentences, which are retrieved sentences, are sample sentences other than the initial dialogue sample sentences and the plurality of candidate response sample sentences among the training sample sentences.

In this embodiment, the execution subject of the above method may be a server or a terminal device.

When the execution subject of the above method is a server, the implementation form of the server is not limited. For example, the server may be a general-purpose server, a cloud server, a cloud host, a server device such as a virtual center. The configuration of the server mainly includes processor, hard disk, memory, system bus, etc., and general computer architecture type.

When the execution subject of the above method is a terminal device, the implementation form of the terminal device is not limited. Terminal devices include, but are not limited to, personal computers, tablet computers, smart phones, and smart wearable devices.

In the present embodiment, training is performed based on the modified reply sample sentence, the second candidate reply sample sentence, and the recall reply sample sentence to obtain a dialogue model, and the initial dialogue sample sentence is input to the initial dialogue model to generate a plurality of candidates. Obtaining reply sample sentences, the second candidate reply sample sentence being one of a plurality of candidate reply sample sentences, and the modified reply sample sentence modifying the first reply sample sentence among the candidate reply sample sentences. recall sample sentences are sample sentences other than initial dialogue sample sentences and a plurality of candidate response sample sentences among training sample sentences; , the second candidate response sample sentence and the recall response sample sentence are continuously trained to obtain a dialogue model with high accuracy of dialogue, obtain an initial dialogue sentence, and convert the initial dialogue sentence to a dialogue Input into the model to get high-quality target responses for dialogue.

The technical proposal of the present disclosure will be described below according to the application scene.

Application scene 1: The smartphone responds to the initial dialogue sentence "How is the weather today?" voiced by the user, the smartphone uploads the initial dialogue sentence to the server, and the server transforms the initial dialogue sentence into a trained dialogue model. By inputting, the target reply sentence "Today is sunny" is acquired, the server down-transmits the target reply sentence to the smartphone, and the smartphone reproduces the target reply sentence "Today is sunny" by voice.

Application scene 2: The smartphone responds to the initial dialogue sentence "How is the weather today?" It's sunny today" is obtained, and the smartphone reproduces the target response sentence "It's sunny today" by voice.

Before using an interaction model, you need to train an initial interaction model to obtain an interaction model. The process of training the dialogue model is described below.

FIG. 2 is a schematic flowchart of a model training method provided by Example 2 of the present disclosure. As shown in FIG. 2, the method includes the following steps S201-204.

S201, obtain an initial dialogue sample sentence.

S202, inputting the initial dialogue sample sentences into the initial dialogue model to obtain a plurality of candidate reply sample sentences;

S203, correcting the first candidate reply sample sentence among the plurality of candidate reply sample sentences to obtain a corrected reply sample sentence;

S204, training an initial dialogue model based on the modified reply sample sentence, the second candidate reply sample sentence among the plurality of candidate reply sample sentences, and the recall reply sample sentence to obtain a dialogue model;
The recall response sample sentences are sample sentences other than the initial dialogue sample sentences and the plurality of candidate response sample sentences among the training sample sentences.

The training device for training the interaction model described above may be any type of computer device, and embodiments of the present disclosure are not limited thereto.

Note that the initial dialogue model may be a trained model, the accuracy of the initial dialogue model is low, and the quality of the dialogue using the initial dialogue model is poor.

An initial dialogue sample sentence is obtained, the initial dialogue sample sentence is input to the initial dialogue model, and a modified reply sample sentence is obtained. modifying a first candidate response sample sentence among the plurality of candidate response sample sentences to obtain a modified response sample sentence; and randomly selecting a second candidate response sample sentence from among the plurality of candidate response sample sentences. , select recall response sample sentences from sample sentences other than the initial dialogue sample sentence and a plurality of candidate response sample sentences among the training sample sentences. The modified reply sample sentences, the second candidate reply sample sentences and the recall reply sample sentences constitute one training data set. Repeat the above steps to get a training dataset for model training.

For the initial dialogue sample sentences, datasets in different fields such as news, social media, literature, and live-action dialogue are used as much as possible in order to increase the coverage of the dataset.

In the above embodiment, the first candidate reply sample sentence among the plurality of candidate reply sample sentences is corrected to obtain the corrected reply sample sentence. For example, the first candidate reply sample sentence is copied, corrected, or created to obtain a modified reply sample sentence.

For example, in response to the operation of inputting an initial dialogue sample sentence in the labeling interface, the initial dialogue sample sentence "It rained every day and I feel sick" is obtained, and the initial dialogue sample sentence is input into the initial dialogue model, Multiple candidate response sample sentences "On rainy days, music and chocolate look great", "Rainy days are perfect for sleeping", "I feel sick too, no one wants to hang out with me", " I like rainy days,” “Me too!

The first candidate response sample sentence among the multiple candidate response sample sentences, "On rainy days, music and chocolate go well together" is corrected, and the corrected response sample sentence is changed to "On rainy days, music and chocolate look good together." Randomly select the second candidate response sample sentence "It's perfect for sleeping on rainy days" from multiple candidate response sample sentences, and select the initial dialogue sample sentence from the training sample sentences. and a plurality of candidate reply sample sentences, the recall reply sample sentence "Today is sunny" is selected. The corrected sample response sentence “On rainy days, music and chocolate look great”, the second candidate sample response sentence “Rainy days are perfect for sleeping”, and the recall sample sentence “Today is sunny” are 1. Configure one training dataset.

In the above embodiment, the initial dialogue model is trained based on the modified sample response sentence, the second candidate response sample sentence of the plurality of candidate response sample sentences, and the recall sample response sentence to obtain the interaction model. As one possible form, the corrected response sample sentences, the second candidate response sample sentences, and the recalled response sample sentences are input to the sentence generation model of the initial dialogue model, and the probabilities of the actual response sentences, the corrected response sample sentences, the second 2 candidate response sample probabilities and recall response sample probabilities are obtained, and an initial An initial sentence generation model and an initial sentence decision model of a dialogue model are jointly trained to obtain a dialogue model.

In one embodiment, an initial sentence generation model and an initial sentence decision model of the initial dialogue model are generated based on the probabilities of the actual response, the modified response sample, the second candidate response sample, and the recall response sample. Collaboratively train to get an interaction model. determining a loss function based on the actual response text and the modified response sample text, and based on the loss function, the probability of the modified response sample text is greater than the probability of the second candidate response sample text, and the probability of the modified response sample text is jointly training an initial sentence generation model and an initial sentence decision model, with the training target being that is greater than the probability of the recall response sample sentence and the probability of the second candidate response sample sentence is greater than the probability of the recall response sample sentence, Get the interaction model.

Together with the description of each embodiment above, FIG. 3 is a flowchart of an information processing method provided by Embodiment 3 of the present disclosure. As shown in FIG. 3, the method includes the following steps S301-S304.

S301, the terminal device responds to a voice input operation and obtains an initial dialogue sentence.

S302, the terminal device sends an initial dialogue message to the server;

S303, the server receives the initial dialogue text, inputs the initial dialogue text into the dialogue model, obtains the target response text, and sends the target response text down to the terminal device.

S304, the terminal device receives the target reply text and reproduces the target reply text by voice.

In this embodiment, the implementation form of the server is not limited. For example, the server may be a general-purpose server, a cloud server, a cloud host, a server device such as a virtual center. The configuration of the server mainly includes processor, hard disk, memory, system bus, etc., and general computer architecture type.

In this embodiment, the implementation form of the terminal device is not limited. Terminal devices include, but are not limited to, personal computers, tablet computers, smart phones, and smart wearable devices.

The implementation of each step in this embodiment can refer to the description in the above embodiments, and the descriptions in this embodiment are omitted. effect can be obtained.

FIG. 4 is a schematic block diagram of an information processing device 40 provided by an exemplary embodiment of the present disclosure. This information processing device 40 includes an acquisition module 41 and an input module 42 .

Acquisition module 41 acquires an initial dialogue sentence.

The input module 42 inputs the initial dialogue sentences into the trained dialogue model to obtain target response sentences.
The dialogue model is a model obtained by training based on the modified reply sample sentence, the second candidate reply sample sentence, and the recall reply sample sentence, and the initial dialogue sample sentence is input to the initial dialogue model to generate a plurality of candidate answers. Obtaining sample sentences, the second candidate response sample sentence is one of a plurality of candidate response sample sentences, and the modified response sample sentence is obtained by modifying the first response sample sentence among the candidate response sample sentences. The recall response sample sentences, which are retrieved sentences, are sample sentences other than the initial dialogue sample sentences and the plurality of candidate response sample sentences among the training sample sentences.

Optionally, when input module 42 inputs initial dialogue sentences into a trained dialogue model to obtain target response sentences:
inside the dialogue model, inputting the initial dialogue sentences into the sentence generation model of the dialogue model to obtain a plurality of candidate response sentences and the probability of each candidate response sentence;
A target reply sentence is obtained by inputting a plurality of candidate reply sentences and the probability of each candidate reply sentence into the sentence decision model of the dialogue model.

Optionally, if the input module 42 inputs a plurality of candidate response sentences and the probability of each candidate response sentence into the sentence decision model of the interaction model to obtain a target response sentence:
A plurality of candidate reply sentences and the probability of each candidate reply sentence are input to a sentence decision model, and a target reply sentence with the highest probability is selected from among the plurality of candidate reply sentences.

FIG. 5 is a schematic block diagram of a model training device 50 provided by an exemplary embodiment of the present disclosure. This model training device 50 includes a sentence acquisition module 51, a sentence input module 52, a correction module 53 and a training module 54,
A sentence acquisition module 51 acquires an initial dialogue sample sentence,
The sentence input module 52 inputs the initial dialogue sample sentences into the initial dialogue model to obtain a plurality of candidate reply sample sentences;
The correction module 53 corrects a first candidate reply sample sentence among the plurality of candidate reply sample sentences to obtain a corrected reply sample sentence,
training module 54 to obtain an interaction model by training an initial dialogue model based on the modified sample response sentence, a second candidate response sample sentence of the plurality of candidate response sample sentences, and the recall sample response sentence;
The recall reply sample sentences are sample sentences other than the initial dialogue sample sentences and the plurality of candidate reply sample sentences among the training sample sentences.

Optionally, training module 54 trains the initial dialogue model based on the modified sample response sentence, the second candidate response sample sentence of the plurality of candidate response sample sentences, and the recall sample response sentence to obtain the interaction model. If you do
The modified response sample sentence, the second candidate response sample sentence, and the recall response sample sentence are input to the sentence generation model of the initial dialogue model, and the actual response sentence, the probability of the modified response sample sentence, and the probability of the second candidate response sample sentence are calculated. and get the probability of the recall response sample sentence,
jointly training an initial sentence generation model and an initial sentence decision model of an initial dialogue model based on the probabilities of actual response sentences, modified response sample sentences, second candidate response sample sentence probabilities, and recall response sample sentence probabilities; Get the interaction model.

Optionally, training module 54 trains the initial sentence generation model and the initial sentence of the initial dialogue model based on the probability of the actual response, the modified response sample sentence, the second candidate response sample sentence probability, and the recall response sample sentence probability. When co-training a decision model to obtain an interaction model,
determining a loss function based on the actual response text and the modified response sample text;
Based on the loss function, the probability of the modified response sample sentence is greater than the probability of the second candidate response sample sentence, the probability of the modified response sample sentence is greater than the probability of the recall response sample sentence, and the probability of the second candidate response sample sentence is greater than the recall response sample sentence probability, the initial sentence generation model and the initial sentence decision model are jointly trained to obtain the dialogue model.

The specific manner of performing the operation of each module of the apparatus of the above embodiment has already been described in detail in the embodiment of the method, but will not be described in detail here.

According to embodiments of the disclosure, the disclosure further provides an electronic device and a readable storage medium.
According to an embodiment of the disclosure, the disclosure further provides a computer program, which, when executed by a processor, implements the information processing method or model training method provided by the disclosure.

FIG. 6 is a schematic block diagram of an exemplary electronic device 600 for implementing embodiments of the present disclosure. Electronic equipment is intended to represent various forms of digital computers such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronics can also represent various forms of mobile devices such as personal digital assistants, cell phones, smart phones, wearable devices, and other similar computing devices. The parts, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the description and/or the required implementation of the disclosure herein.

As shown in FIG. 6, electronic device 600 performs various suitable operations according to computer programs stored in read only memory (ROM) 602 or loaded into random access memory (RAM) 603 from storage unit 608 . and a computing unit 601 that can perform processing. Various programs and data necessary for the operation of the electronic device 600 may also be stored in the RAM 603 . Computing unit 601 , ROM 602 and RAM 603 are connected to each other via bus 604 . An input/output (I/O) interface 605 is also connected to bus 604 .

A plurality of components of the electronic device 600 are connected to an I/O interface 605, including an input unit 606 such as a keyboard, mouse, etc., an output unit 607 such as each type of display, speakers, etc., a storage unit 608 such as a magnetic disk, an optical disk, etc., and a network It includes a communication unit 609 such as a card, modem, wireless communication transceiver. Communication unit 609 enables electronic device 600 to exchange information/data with other devices via computer networks such as the Internet and/or various telegraph networks.

Computing unit 601 may be various general purpose and/or special purpose processing components having processing and computing capabilities. Some examples of computational unit 601 are a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computational chips, various computational units that run machine learning model algorithms, digital signal including, but not limited to, a processor (DSP), and any suitable processor, controller, microcontroller, or the like. The computing unit 601 performs each of the methods and processes described above, eg, information processing methods and model training methods. For example, in some embodiments information processing methods and model training methods may be implemented as computer software programs tangibly embodied in a machine-readable medium, such as storage unit 608 . In some examples, part or all of the computer program may be loaded and/or installed in electronic device 600 via ROM 602 and/or communication unit 609 . When the computer program is loaded into RAM 603 and executed by computing unit 601, one or more steps of the information processing method and model training method described above may be performed. Alternatively, in other embodiments, computing unit 601 may be configured in any other suitable manner (eg, via firmware) to perform information processing methods and model training methods.

Various embodiments of the systems and techniques described herein above include digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), ), system-on-chip (SOC), complex programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being embodied in one or more computer programs, which are executed and executed on a programmable system including at least one programmable processor. /or may be interpreted, the programmable processor may be an application-specific or general-purpose programmable processor, receives data and instructions from a storage system, at least one input device, and at least one output device; Data and instructions can be transmitted to the storage system, the at least one input device, and the at least one output device.

Program code to implement the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be stored in a general purpose computer, special purpose computer, or other programmable data processing apparatus to perform the functions/operations specified in the flowcharts and/or block diagrams when executed by a processor or controller. It may be provided in a processor or controller. Program code may be executed entirely on a machine, partially on a machine, or as a stand-alone software package, partially on a machine, partially on a remote machine, or completely may be executed on a remote machine or server.

In the context of this disclosure, a machine-readable medium contains a program for use by or in conjunction with an instruction execution system, apparatus, or device, or It may be a tangible medium capable of being stored. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus or devices, or any suitable combination of the above content. . More specific examples of machine-readable storage media are electrical connections based on one or more lines, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable reads. Including memory only (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the above content.

In order to provide interaction with a user, the systems and techniques described herein can be implemented on a computer, which includes a display device (e.g., CRT (Cathode Ray Tube)) for displaying information to the user. or LCD (liquid crystal display) monitor), and a keyboard and pointing device (eg, mouse or trackball) through which a user can provide input to the computer. Other types of devices can also provide interaction with a user, e.g., the feedback provided to the user can be any form of sensing feedback (e.g., vision feedback, auditory feedback, or tactile feedback). may receive input from the user in any form (including acoustic, voice, or tactile input).

The systems and techniques described herein may be computing systems that include back-end components (e.g., data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include front-end components. A system (e.g., a user computer having a graphical user interface or web browser, through which a user can interact with embodiments of the systems and techniques described herein), or such a back-end component , middleware components, and front-end components in any combination. The components of the system can be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), the Internet, and blockchain networks.

The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server is created by computer programs running on corresponding computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or cloud host, is one host product in the cloud computing service system, and is the same as the traditional physical host and VPS service ("Virtual Private Server", or " VPS”) has solved the problems of difficulty in management and weak business expandability. The server may be a server of a distributed system, or a server incorporating a blockchain.

It should be appreciated that steps may be reordered, added, or deleted using the various forms of flow shown above. For example, each step described in the present disclosure may be executed in parallel, sequentially, or in a different order. There is no limitation here as long as the results can be achieved.

The above specific embodiments do not limit the protection scope of the present disclosure. It should be understood that those skilled in the art can make various modifications, combinations, subcombinations and substitutions depending on design requirements and other factors. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this disclosure should all fall within the protection scope of this disclosure.

Claims (15)

An information processing method,
obtaining an initial dialogue;
inputting the initial dialogue sentences into a trained dialogue model to obtain target response sentences;
The dialogue model is a model obtained by training based on a modified reply sample sentence, a second candidate reply sample sentence and a recall reply sample sentence, and an initial dialogue sample sentence is input to the initial dialogue model to generate a plurality of candidates. obtaining a reply sample sentence, wherein the second candidate reply sample sentence is one of the plurality of candidate reply sample sentences, and the modified reply sample sentence is a first reply sample among the candidate reply sample sentences; is a sentence obtained by modifying a sentence, and the recall response sample sentence is a sample sentence other than the initial dialogue sample sentence and the plurality of candidate response sample sentences among the training sample sentences.
An information processing method characterized by: inputting the initial dialogue sentence into a trained dialogue model to obtain a target response sentence;
within the dialogue model, inputting the initial dialogue sentences into a sentence generation model of the dialogue model to obtain a plurality of candidate response sentences and a probability of each of the candidate response sentences;
inputting the plurality of candidate response sentences and the probability of each candidate response sentence into a sentence decision model of the interaction model to obtain a target response sentence;
2. The information processing method according to claim 1, wherein: inputting the plurality of candidate response sentences and the probability of each candidate response sentence into a sentence decision model of the dialogue model to obtain a target response sentence;
inputting the plurality of candidate response sentences and the probability of each of the candidate response sentences into the sentence decision model, and selecting the target response sentence with the highest probability from among the plurality of candidate response sentences;
3. The information processing method according to claim 2, characterized by: A model training method comprising:
obtaining an initial dialog sample sentence;
inputting the initial dialogue sample sentences into an initial dialogue model to obtain a plurality of candidate reply sample sentences;
modifying a first candidate response sample sentence among the plurality of candidate response sample sentences to obtain a modified response sample sentence;
training the initial dialogue model based on the modified sample response sentence, a second candidate response sample sentence of the plurality of candidate response sample sentences, and the recall sample response sentence to obtain a dialogue model. ,
The recall response sample sentences are sample sentences other than the initial dialogue sample sentences and the plurality of candidate response sample sentences among the training sample sentences.
A model training method characterized by: training the initial dialogue model based on the modified sample response sentence, a second candidate response sample sentence of the plurality of candidate response sample sentences, and the recall sample response sentence to obtain a dialogue model;
Inputting the modified reply sample sentence, the second candidate reply sample sentence and the recall reply sample sentence into the sentence generation model of the initial dialogue model, the probability of the actual reply sentence, the probability of the modified reply sample sentence and the second candidate reply sentence obtaining a probability of a sample sentence and a probability of a recall response sample sentence;
an initial sentence generation model and an initial sentence decision model of the initial dialogue model based on the probability of the actual response sentence, the probability of the modified sample response sentence, the probability of the second candidate response sample sentence, and the probability of the recalled response sample sentence; to obtain the interaction model by jointly training the
The model training method according to claim 4, characterized in that: the initial sentence generation model and the initial sentence decision model of the initial dialogue model based on the probability of the actual response sentence, the probability of the modified sample response sentence, the probability of the second candidate response sample sentence, and the probability of the recalled response sample sentence; The step of jointly training with to obtain the interaction model includes:
determining a loss function based on the actual response text and the modified sample response text;
Based on the loss function, the probability of the modified response sample sentence is greater than the probability of the second candidate response sample sentence, the probability of the modified response sample sentence is greater than the probability of the recall response sample sentence, and the second jointly training the initial sentence generation model and the initial sentence decision model to obtain the dialogue model, with a training target that the probability of the candidate response sample sentence is greater than the probability of the recall response sample sentence. include,
The model training method according to claim 5, characterized in that: An information processing device,
a retrieving module for retrieving an initial dialogue statement;
an input module for inputting the initial dialogue sentence into a trained dialogue model to obtain a target response sentence;
The dialogue model is a model obtained by training based on a modified reply sample sentence, a second candidate reply sample sentence and a recall reply sample sentence, and an initial dialogue sample sentence is input to the initial dialogue model to generate a plurality of candidates. obtaining a reply sample sentence, wherein the second candidate reply sample sentence is one of the plurality of candidate reply sample sentences, and the modified reply sample sentence is a first reply sample among the candidate reply sample sentences; is a sentence obtained by modifying a sentence, and the recall response sample sentence is a sample sentence other than the initial dialogue sample sentence and the plurality of candidate response sample sentences among the training sample sentences.
An information processing device characterized by: When the input module inputs the initial dialogue sentence to a trained dialogue model to obtain a target response sentence,
inside the dialogue model, inputting the initial dialogue sentences into a sentence generation model of the dialogue model to obtain a plurality of candidate response sentences and a probability of each of the candidate response sentences;
inputting the plurality of candidate response sentences and the probability of each of the candidate response sentences into a sentence decision model of the dialogue model to obtain a target response sentence;
8. The information processing apparatus according to claim 7, characterized by: When the input module inputs the plurality of candidate reply sentences and the probability of each candidate reply sentence to the sentence decision model of the dialogue model to acquire a target reply sentence,
inputting the plurality of candidate response sentences and the probability of each of the candidate response sentences into the sentence determination model, and selecting a target response sentence with the highest probability from among the plurality of candidate response sentences;
9. The information processing apparatus according to claim 8, characterized by: A model training device,
a sentence acquisition module for acquiring initial dialogue sample sentences;
a sentence input module for inputting the initial dialogue sample sentences into an initial dialogue model to obtain a plurality of candidate reply sample sentences;
a correction module for correcting a first candidate reply sample sentence among the plurality of candidate reply sample sentences to obtain a corrected reply sample sentence;
a training module for training the initial dialogue model based on the modified sample response sentence, a second candidate response sample sentence among the plurality of candidate response sample sentences, and the recall sample response sentence to obtain a dialogue model; including
The recall response sample sentences are sample sentences other than the initial dialogue sample sentences and the plurality of candidate response sample sentences among the training sample sentences.
A model training device characterized by: The training module trains the initial dialogue model based on the modified sample response sentence, a second candidate response sample sentence of the plurality of candidate response sample sentences, and a recall response sample sentence to obtain a dialogue model. case,
Inputting the modified reply sample sentence, the second candidate reply sample sentence and the recall reply sample sentence into the sentence generation model of the initial dialogue model, the probability of the actual reply sentence, the probability of the modified reply sample sentence and the second candidate reply sentence Obtain the probability of sample sentences and the probability of recall response sample sentences,
an initial sentence generation model and an initial sentence decision model of the initial dialogue model based on the probability of the actual response sentence, the probability of the modified sample response sentence, the probability of the second candidate response sample sentence, and the probability of the recalled response sample sentence; to obtain said interaction model;
11. The model training device according to claim 10, wherein: The training module trains the initial sentence generation model of the initial dialogue model based on the actual response sentence probabilities, the modified response sample sentence probabilities, the second candidate response sample sentence probabilities, and the recall response sample sentence probabilities. and an initial sentence decision model to obtain said dialogue model,
determining a loss function based on the actual response text and the modified sample response text;
Based on the loss function, the probability of the modified response sample sentence is greater than the probability of the second candidate response sample sentence, the probability of the modified response sample sentence is greater than the probability of the recall response sample sentence, and the second jointly training the initial sentence generation model and the initial sentence decision model to obtain the dialogue model, with the probability of the candidate reply sample sentence being greater than the probability of the recall reply sample sentence as a training target;
12. The model training device according to claim 11, characterized by: an electronic device,
at least one processor;
a memory communicatively coupled to the at least one processor;
Instructions executable by the at least one processor are stored in the memory, and the instructions are stored so as to enable the at least one processor to perform the method according to any one of claims 1 to 3 or 4 to 6. executed by the at least one processor;
An electronic device characterized by: A non-transitory computer-readable storage medium having computer instructions stored thereon,
The computer instructions cause a computer to perform the method of any of claims 1-3 or 4-6,
A non-transitory computer-readable storage medium characterized by: A computer program,
The computer program, when executed by a processor, implements the steps of the method according to any of claims 1-3 or 4-6,
A computer program characterized by:

Images