CN116681060A - Dialogue data processing method, man-machine interaction method, equipment and storage medium - Google Patents

Abstract

An embodiment of the present invention provides a dialogue data processing method, a human-computer interaction method, a device, and a storage medium. The method includes: the processing device acquires statistical features of the dialogue data to be processed including at least one round of question and answer, and the statistical features are related to the dialogue data to be processed. The dialogue content of the data is irrelevant, and further, it can be determined whether there is a question-answer mismatch in the dialogue data to be processed according to this statistical feature. In the above method, since the statistical characteristics of the dialogue data to be processed are irrelevant to the content of the dialogue and are also correlated with whether there is a question-answer mismatch in the dialogue data to be processed, the dialogue data generated in different scenarios and with different contents are different. The above method can be used to detect whether there is a question-answer mismatch, thereby improving the generalization of question-answer mismatch detection.

Description

Dialog data processing method, human-computer interaction method, device and storage medium

technical field

The present invention relates to the technical field of artificial intelligence, in particular to a dialog data processing method, a human-computer interaction method, a device and a storage medium.

Background technique

Using artificial intelligence technology has been able to achieve man-machine dialogue, and man-machine dialogue can also be applied to various scenarios. For example, a hardware interaction device integrated with a dialogue system may be used to conduct a dialogue with the user. Wherein, the interactive device may specifically be a service robot, a smart speaker, a mobile terminal device, and the like. For example, when the service robot is a guidance robot deployed in shopping malls and restaurants, the service robot can guide the user through multiple rounds of dialogue with the user. For another example, the dialogue system deployed in the cloud can also be used to communicate with users, such as online intelligent customer service, etc., to meet the user's query requirements for relevant information.

In different scenarios, the content of man-machine dialogue is very rich. And in any scenario, a common situation in the process of man-machine dialogue is that the question and answer do not match, that is, the answer is not what was asked, which will seriously affect the user experience. Therefore, how to accurately detect the question-answer mismatch in the process of man-machine dialogue in different scenarios has become an urgent problem to be solved.

Contents of the invention

In view of this, the embodiments of the present invention provide a dialogue data processing method, a human-computer interaction method, a device, and a storage medium, so as to accurately detect whether there is a question-answer mismatch in dialogue data in different scenarios.

In a first aspect, an embodiment of the present invention provides a dialog data processing method, including:

Obtaining statistical features of the dialogue data to be processed including at least one round of question and answer, the statistical features having nothing to do with the dialogue content of the dialogue data to be processed;

Determine whether there is a question-answer mismatch in the dialogue data to be processed according to the statistical features.

In a second aspect, an embodiment of the present invention provides a human-computer interaction method, including:

Displaying a prompt message, the prompt message is generated when the amount of abnormal dialogue data is greater than a preset number, the abnormal dialogue data includes dialogue data that does not match the question and answer, and the dialogue data is generated by the user and the dialogue system during the man-machine dialogue Generate, whether there is a question-answer mismatch in the dialogue data is determined by the statistical characteristics of the dialogue data;

Respond to the processing operation on the prompt message.

In a third aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory, the memory is used to store one or more computer instructions, wherein, when the one or more computer instructions are executed by the processor Realize the dialog data processing method in the first aspect above, or the human-computer interaction method in the second aspect above. The electronic device may also include a communication interface for communicating with other devices or communication systems.

In a fourth aspect, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable code is stored on the non-transitory machine-readable storage medium, and when the executable code is executed by a processor of an electronic device During execution, the processor can at least implement the dialog data processing method in the first aspect above, or the human-computer interaction method in the second aspect above.

In the dialogue data processing method provided by the embodiment of the present invention, the processing device obtains the statistical characteristics of the dialogue data to be processed including at least one round of question and answer, and the statistical characteristics have nothing to do with the dialogue content of the dialogue data to be processed. Further, it can be determined according to the statistical characteristics Whether there is a question-answer mismatch in the pending conversation data. In the above method, since the statistical characteristics of the dialogue data to be processed have nothing to do with the dialogue content, therefore, the dialogue data generated in different scenarios and with different contents can use the above method to detect whether there is a question-answer mismatch, thereby improving the question-answer gap. Generalization of match detection.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

FIG. 1 is a flowchart of a dialog data processing method provided by an embodiment of the present invention;

Fig. 2 is a schematic diagram of an operation interface provided by an embodiment of the present invention;

FIG. 3 is a flow chart of a detection model training method provided by an embodiment of the present invention;

FIG. 4 is a flow chart of another dialogue data processing method provided by an embodiment of the present invention;

FIG. 5 is a flowchart of another detection model training method provided by an embodiment of the present invention;

FIG. 6 is a flowchart of another detection model training method provided by an embodiment of the present invention;

FIG. 7 is a schematic diagram of the application of the dialog data processing method provided by the embodiment of the present invention;

FIG. 8 is a flowchart of a human-computer interaction method provided by an embodiment of the present invention;

FIG. 9 is a flowchart of another human-computer interaction method provided by an embodiment of the present invention;

FIG. 10 is a schematic structural diagram of a dialogue data processing device provided by an embodiment of the present invention;

FIG. 11 is a schematic structural diagram of a human-computer interaction device provided by an embodiment of the present invention;

FIG. 12 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention;

FIG. 13 is a schematic structural diagram of another electronic device provided by an embodiment of the present invention;

Fig. 14 is a schematic structural diagram of another human-computer interaction device provided by an embodiment of the present invention;

FIG. 15 is a schematic structural diagram of another electronic device provided by an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates otherwise, "multiple" Generally, at least two kinds are included, but the case of including at least one kind is not excluded.

It should be understood that the term "and/or" used herein is only an association relationship describing associated objects, which means that there may be three relationships, for example, A and/or B, which may mean that A exists alone, and A and B exist simultaneously. B, there are three situations of B alone. In addition, the character "/" in this article generally indicates that the contextual objects are an "or" relationship.

Depending on the context, the words "if", "if" as used herein may be interpreted as "at" or "when" or "in response to determining" or "in response to identifying". Similarly, depending on the context, the phrases "if determined" or "if identified (the stated condition or event)" could be interpreted as "when determined" or "in response to the determined" or "when identified (the stated condition or event) )” or “in response to recognition of (a stated condition or event)”.

It should be noted that the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a good or system comprising a set of elements includes not only those elements but also items not expressly listed other elements, or also include elements inherent in the commodity or system. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the article or system comprising said element.

It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all It is information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose authorization or reject.

It should be noted that, similar to the description in the background art, intelligent robots such as service robots, welcome robots, and self-moving vending robots can be integrated with a dialogue system. In addition, smart terminals such as mobile terminals, smart home appliances, and smart wearable devices can also be integrated with dialogue systems. Each of the above hardware devices can be considered as an interactive device. In addition, the dialogue system can also be deployed in the cloud to provide users with online shopping, online consultation and other services. Generally speaking, any dialogue system that supports man-machine dialogue with users in voice or text and is deployed on hardware devices or in the cloud can generate dialogue data with users. In addition, the following embodiments of the present invention can determine the task-type dialogue generated by the user and the dialogue system during the man-machine dialogue process as dialogue data to be processed.

Before describing in detail the dialog data processing method and the human-computer interaction method provided by the following embodiments of the present invention, the related concepts involved in the following embodiments can also be explained first:

Task-based dialogue: At least one round of dialogue, that is, at least one round of question and answer, generated by the dialogue system for a certain demand of the user. The dialogue system can determine the user's intention through understanding, clarification, etc., and then complete the user's demand by replying.

Question and answer mismatch: When the dialogue system answers the question given by the user, it does not directly give an answer, or does not have the actual meaning of answering the question. The question and answer mismatch actually means that the answer is not what was asked.

For example, in an offline service scenario such as an offline bank consultation scenario, the following dialogue data can be generated between the user and the service robot set up in the bank lobby:

Service robot: Hello, what do you need to do?

User: I want to change some dollars.

Service Robot: Do you have your ID card and bank card with you?

User: Bring your ID card and bank card.

Service robot: Please put your ID card in the induction area, select foreign exchange exchange and get the number.

In the above dialogue, the service robot finally successfully guides the user to complete the number taking of the foreign exchange dialogue, so there is no question-answer mismatch in the above dialogue data.

For example, in an online service scenario such as an online government information consultation scenario, the following dialogue data can be generated between a user and an online intelligent customer service:

Online intelligent customer service: Hello, I can provide you with information consultation on human resources and social security. You can talk about pension adjustment, salary declaration, medical insurance, maternity allowance, unemployment insurance and so on.

User: human service.

Online intelligent customer service: You can talk about the questions you want to consult, if you cannot answer, you will be transferred to manual customer service.

User: I called a few days ago about maternity allowance.

Online intelligent customer service: May I ask whether you are giving birth or having a miscarriage.

User: I did not write the name of the branch when I applied for maternity allowance last time.

Online intelligent customer service: Sorry, I didn't hear clearly, please tell me.

User: human service.

In the above conversation, the online intelligent customer service did not understand the user's intention in the end, so there is a question-answer mismatch in the above conversation data.

In practice, the mismatch of questions and answers will seriously affect the user experience. At this time, the methods mentioned in the following embodiments of the present invention can be used to detect whether there is a mismatch of questions and answers in the dialogue data.

Based on the above description, some embodiments of the present invention may be described in detail below in conjunction with the accompanying drawings. Under the condition that there is no conflict between the various embodiments, the following embodiments and the features in the embodiments can be combined with each other. In addition, the sequence of steps in the following method embodiments is only an example, rather than a strict limitation.

FIG. 1 is a flow chart of a dialog data processing method provided by an embodiment of the present invention. The dialog data processing method provided by the embodiment of the present invention may be executed by a processing device. Processing devices can be either physical or virtual. As shown in Figure 1, the method may include the following steps:

S101. Obtain statistical features of the dialogue data to be processed including at least one round of question and answer, where the statistical features are not related to the dialogue content of the dialogue data to be processed.

During the man-machine dialogue process between the user and the dialogue system, the dialogue data jointly generated by the two is the dialogue data to be processed. And in response to the opening of the man-machine dialogue, the dialogue system can record the pending dialogue data generated between itself and the user. Wherein, a piece of dialogue data to be processed may include at least one round of questions and answers generated during a dialogue between the user and the dialogue system. For example, the above-mentioned dialogue data generated in the offline service bank consultation scenario or the online government information consultation scenario.

Optionally, both the user and the dialog system can actively start the man-machine dialog. For the case where the user actively triggers the dialogue opening operation, following the above-mentioned offline banking consultation scenario, optionally, the user can actively trigger the dialogue opening operation on the operation interface provided by the service robot to actively open the man-machine dialogue. The dialogue data to be processed generated at this time can usually be in the form of speech or text. To undertake the above-mentioned online government information consultation scenario, users can actively dial the government information consultation phone to initiate a man-machine dialogue. At this point, the generated dialogue data to be processed can usually be in the form of speech. The user can also actively trigger the dialogue opening operation on the operation interface provided by the government website to actively open the man-machine dialogue. At this point, the dialog data to be processed can generally be represented as text data.

For the case where the dialogue system actively triggers the dialogue opening operation, in the online customer service return visit scenario, optionally, the dialogue system can actively call the user to start the man-machine dialogue. The dialogue data to be processed in this scenario can be expressed in phonetic form.

Optionally, for the acquisition of statistical features, the dialogue system may also record the generation time of each statement in the dialogue data to be processed while recording the statement. Based on this, the dialogue system or other devices or systems can make statistics on the dialogue data to be processed according to the generation time, so as to obtain the statistical characteristics of the dialogue data to be processed. The processing device can obtain this statistical feature from the dialogue system.

Optionally, the statistical features of the dialog data to be processed may include features in the dialog data that are not related to the dialog content. Optionally, it may specifically include at least one of the following: the number of dialogue rounds, the number of times the dialogue system did not give an answer, whether the dialogue generated by the user is too long, the number of times the user did not produce a dialogue within the preset duration, and the number of times the dialogue was switched to a manual , Whether the dialogue generated by the user contains negative emotions, the number of references in the dialogue, whether the dialogue system does not give an answer and the user hangs up in the last round of dialogue, and whether the user is within the preset duration of the last round of dialogue No dialogue etc. was produced. According to the statistical characteristics given above, it can be seen that the statistical characteristics are actually numerical characteristics.

The above statistical characteristics can also be explained as follows:

The failure of the dialogue system to give an answer can be manifested as the dialogue system giving a response indicating that it does not understand the user's intention, such as "Sorry, I didn't hear you clearly, please say" or "Sorry, please say it again".

Containing negative emotions in the dialog generated by the user may be represented by including preset emotional words in the dialog generated by the user, and the emotional words represent negative emotions.

If the user does not generate a dialog within the preset time period, the user generates a dialog timeout, and the number of times the user does not generate a dialog within the preset time period is the number of times the user generates a dialog timeout. Similarly, whether the user does not generate a dialogue within the preset time period in the last round of dialogue is the timeout of the user's last round of dialogue.

Among them, the statistical features of "whether the conversation generated by the user is too long", "whether the conversation generated by the user contains negative emotions" and "whether the user did not generate a conversation within the preset time period in the last round of dialogue" can also be used in different ways. The mark indicates whether the above-mentioned situation occurs, for example, "1" can be used to indicate that the above-mentioned situation occurs, and "0" can be used to indicate that the above-mentioned situation does not occur.

Then the statistical characteristics of any dialogue data to be processed can be schematically shown in the following table:

The content in the above table indicates that the dialog data to be processed has 3 dialog turns, 1 transfer to manual dialog, and 1 reference in the dialog.

And it should be noted that each feature in the statistical features is specifically related to whether there is a question-answer mismatch in the dialogue data. Optionally, the level of correlation is specifically measured by any one of Pearson correlation coefficient, Spearman (Spearman Rank Correlation Coefficient, Src for short) rank correlation coefficient, Kendall (Kendall) correlation coefficient and the like.

Specifically, the statistical features of "number of dialogue rounds", "number of times the dialogue system did not give an answer", "whether the dialogue generated by the user is too long", and "the number of times the user did not produce a dialogue within the preset duration" are related to the dialogue Data with a question-answer mismatch is highly positively correlated. That is, the more dialogue rounds, the longer the dialogue generated by the user, and the more times the user did not produce a dialogue within the preset time period, the higher the possibility of a question-answer mismatch in the dialogue data.

"Whether the dialogue system did not give an answer and the user hung up in the last round of dialogue", "Whether the user did not generate a dialogue within the preset time period during the last round of dialogue", "Whether the dialogue generated by the user contains negative emotions " and "Number of Referrals in Dialogues" are positively correlated with question-answer mismatches in dialogue data.

The statistical feature "Number of conversations transferred to a human" is negatively correlated with the question-answer mismatch in the conversation data, that is, the fewer the number of conversations transferred to a human, the lower the possibility of a question-answer mismatch in the conversation data.

S102. Determine whether there is a question-answer mismatch in the dialogue data to be processed according to statistical features.

Next, the processing device may directly determine whether there is a question-answer mismatch in the pending dialog according to the statistical characteristics of the pending dialog data.

Optionally, for numerical statistical features, according to the degree of correlation between different statistical features and whether there is a question-answer mismatch in the dialogue data, a weight coefficient can be set for each statistical feature, and the statistical feature and its corresponding The weighting system performs weighted summation, and the summation result can be compared with a preset value to determine whether there is a question-answer mismatch in the dialogue data to be processed.

In this embodiment, the processing device obtains the statistical features of the dialogue data to be processed including at least one round of question and answer, the statistical features have nothing to do with the dialogue content of the dialogue data to be processed, and further, can determine whether the dialogue data to be processed exists according to the statistical features Question and answer do not match. In the above method, since the statistical characteristics of the dialogue data to be processed are features that have nothing to do with the content of the dialogue, therefore, dialogue data with different contents generated in different scenarios can be detected whether there is a question-answer mismatch using the above method, thereby improving Generalizability for question-answer mismatch detection. In addition, compared with the features of the feature vector type, the statistical features of the numerical type contain a smaller amount of data. Therefore, using the above method can also reduce the amount of calculation in the process of detecting whether there is a question-answer mismatch.

It is easy to understand that the greater the amount of dialogue data that does not match the question and answer, the worse the performance of the dialogue system. Therefore, the processing device can use all dialogue data generated in the historical time period as dialogue data to be processed and implement it as shown in Figure 1. According to the detection results, we can know the number of abnormal dialogue data with mismatched questions and answers in the historical time period. Optionally, the quantity can also be displayed on a statistical interface provided by the processing device. If the amount of abnormal dialogue data is greater than the preset amount, a prompt message may also be displayed on the statistics interface. At this time, the operation and maintenance personnel of the dialogue system can choose whether to optimize the dialogue system according to the prompt message displayed on the statistics interface.

Optionally, the statistical interface can also display the quantity change curve of the dialogue data in the historical time period, and other information related to the abnormal dialogue data can also be displayed on the statistical interface, such as the total number of dialogue data in the current historical time period, the current The proportion of abnormal dialogue data in the historical time period, compared with the number of abnormal dialogue data in the previous historical period, the change rate of the abnormal dialogue data in the current historical period, etc. Optionally, the historical time end may be one week, one month, or several months.

The statistical interface provided by the processing device may be as shown in FIG. 2 .

For the process of determining whether there is a question-answer mismatch in the dialogue data to be processed according to statistical features, in addition to the weighted summation method in the embodiment shown in FIG. 1 , another optional method can also be implemented by means of a detection model. Specifically, the statistical features of the dialogue data to be processed can be input into the detection model, so that the detection model can directly output whether there is a question-answer mismatch in the dialogue data to be processed. Wherein, the detection model is essentially a classification model, then optionally, the detection model can be any model that can realize the classification function, such as various neural networks (Neural Networks, NN for short), logistic regression (Logistic regression) Model, Support Vector Machine (Support Vector Machine, SVM) and so on.

In an optional manner, an initial model may be trained in a supervised training manner, so as to obtain the detection model used in the foregoing embodiments. In another optional manner, multiple initial models may also be trained to obtain the above-mentioned detection model. Fig. 3 is a flow chart of a detection model training method provided by an embodiment of the present invention. The dialog data processing method provided by the embodiment of the present invention may be executed by a training device. Optionally, the training device may be the processing device mentioned in the foregoing embodiments, or other devices other than the processing device. As shown in Figure 3, the following steps may be included:

S201. Obtain statistical features of training dialogue data including at least one round of question-and-answer and labeling results reflecting whether there is a question-answer mismatch in the training dialogue data.

The training device may first obtain training dialogue data including at least one round of question and answer. Optionally, the training dialogue data may be collected through the network, or may be dialogue data generated by users within a historical time period. At the same time, the training device can also obtain an annotation result of the training dialogue data, which is used to reflect whether there is a question-answer mismatch in the training dialogue data. Optionally, the annotation result can be manually annotated.

S202. Using the statistical features of the training dialogue data as training data, and using the labeling results as supervision information, respectively train a preset number of initial models to obtain candidate models with different model parameters.

Then, the training device can also acquire a preset number of initial models. Afterwards, in an optional manner, the preset number of initial models may have the same model structure and different initial model parameters, then the statistical features of the training dialogue data in step S201 may be used as training data, and the training dialogue data's The annotation results are used as supervisory information to train a preset number of initial models respectively. Since the initial model parameters of the initial model are different, alternative models with different model parameters can be obtained even with the same training data.

In another optional manner, the training dialog data in step S201 may be divided according to a preset number. Then, the respective statistical features of the plurality of training dialogue data are used as training data of a preset number of initial models, and the labeling results of the training dialogue data are used as supervision information to train a preset number of initial models respectively. In this manner, the preset number of initial models may have the same model structure and the same initial model parameters. Since the training dialogue data used to train the initial model is different, the model parameters of the obtained candidate models are different.

Optionally, the loss calculation can be performed according to the following formula, and the initial model training can be realized by gradient descent:

Among them, p is the confidence degree of the detection result output by the detection model indicating that there is a question-answer mismatch in the training dialogue sample, σ is the loss function used to calculate the loss value between the detection result and the standard result, x=W1X1+W2X2+...+ WkXk+b, X1, X2...Xk are the statistical features of the training dialogue data, W1, W2...Wk are the weight values of the detection model, b is the bias value of the detection model, and the weight value and the bias value together constitute Model parameters for detection models.

S203. Determine the model parameters of the detection model according to the model parameters of the candidate model to obtain a detection model.

Finally, the training device can integrate the respective model parameters of the preset number of candidate models to determine the model parameters of the detection model. Optionally, the model parameters of the preset number of candidate models can be averaged or weighted and summed, and the calculation results can be used as the model parameters of the detection model, thereby obtaining the detection model. Among them, the detection model has the same model structure as the initial model and the candidate model.

In this embodiment, the detection accuracy of the detection model is improved by comprehensively considering model parameters of multiple candidate models, that is, parameter integration. Specifically, the training device first trains a plurality of candidate models, and then obtains the model parameters of the detection model by comprehensively considering the model parameters of the candidate models, thereby improving the accuracy of the model parameters of the detection model, that is, improving the accuracy of the detection model. The detection accuracy of the model.

In order to improve the detection accuracy of the question-answer mismatch, another optional way, after obtaining a preset number of candidate models according to the embodiment shown in Figure 3, can also be combined with an ensemble learning (Ensemble Learning) mechanism, that is Multiple detection results can be considered comprehensively to obtain more accurate detection results.

Specifically, each of the preset number of candidate models can output the detection result of the dialogue data to be processed and the confidence of the detection result, and then integrate multiple detection results such as averaging the confidence or performing a weighted summation of the confidence , to finally obtain the detection result of the dialogue data to be processed.

On the basis of the embodiment shown in FIG. 1 , in order to further improve the accuracy of question-answer mismatch detection, FIG. 4 is a flowchart of another dialogue data processing method provided by an embodiment of the present invention. As shown in Figure 4, the following steps may be included:

S301. Obtain statistical features of the dialogue data to be processed including at least one round of dialogue, where the statistical features are not related to the dialogue content of the dialogue data to be processed.

For the specific implementation process of step S301, reference may be made to the specific description of relevant steps in the embodiment shown in FIG. 1 , which will not be repeated here.

S302. Perform feature combination on statistical features to obtain combined features of the dialogue data to be processed.

S303. Determine whether there is a question-answer mismatch in the dialogue data to be processed according to the combination feature of the dialogue data to be processed.

The processing device may perform feature combination, that is, feature intersection, on the statistical features of the dialogue data to be processed, so as to obtain combined features of the dialogue data to be processed. Among them, the combination of features can also be understood as follows: just like the various statistical features given in the above table, any at least two of the features can be combined. It should be noted that the two features to be combined may be the same or different two features. And the same as the statistical feature, the combined feature has nothing to do with the dialogue content of the dialogue data to be processed.

Optionally, binomial or higher order polynomials can be used for feature combination. For example, for statistical features [a,b,c], the combined features obtained after binomial feature crossover can include [a,b,c,ab,bc,ac,a ² ,b ² ,c ² ]; the combined features after trinomial feature crossover can include [a, b, c, ab, bc, ac, a ³ , b ³ , c ³ , a ² b, ab ² , a ² c, ac ² ,b ² c,bc ² ,abc].

Wherein, a, b, and c may represent any three of the multiple statistical features mentioned in the embodiment shown in FIG. 1 . In the digital sense, a, b, c are expressed as numerical values, and then ab, bc, a ² , c ² , etc. are the products of numerical values. In the physical sense, for example, a represents whether the conversation generated by the user is too long, b represents whether the conversation generated by the user contains negative emotions, c represents the number of times the conversation is transferred to a manual conversation, ab represents whether the conversation generated by the user is too long, and the conversation generated by the user Whether the dialogue contains negative emotions, ac represents whether the dialogue generated by the user is too long, and whether there are manual dialogues and the number of manual dialogues.

It can be seen that compared with the original statistical features, the number of statistical features can be expanded through feature combination, and the expanded combined features can also more comprehensively describe the characteristics of the dialogue data to be processed, so it can be more accurately determined. Whether there is a question-answer mismatch in the conversation data.

In this embodiment, by combining the statistical features of the dialogue data to be processed, combined features with richer quantities and meanings can be obtained, which can more comprehensively describe the characteristics of the dialogue data to be processed, so as to finally more accurately determine the Process dialog data for question-answer mismatches. In addition, for the content not described in detail in this embodiment and the technical effects that can be achieved, refer to the relevant description in the embodiment shown in FIG. 1 , and details are not repeated here.

Optionally, for the method provided by the embodiment shown in FIG. 4 , optionally, the detection model can also perform feature combination of statistical features, and further input the combined features into the detection model, so that the detection model can output the dialog data to be processed. Whether there is a question-answer mismatch detection result.

That is to say, compared with the detection model mentioned in the embodiment shown in Figure 1 and Figure 3, the detection model corresponding to the embodiment shown in Figure 4 also has a feature combination function, then for the detection model that performs feature combination, Fig. 5 is a flow chart of another model training provided by the embodiment of the present invention. This embodiment can also be executed by the above-mentioned training device. As shown in Figure 5, the steps may include the following steps:

S401. Obtain statistical features of training dialogue data including at least one round of dialogue and labeling results reflecting whether there is a question-answer mismatch in the training dialogue data.

For the specific implementation process of the above step S401, reference may be made to the specific description of the relevant steps in the embodiment shown in FIG. 3 , which will not be repeated here.

S402. Perform feature combination on statistical features of the training dialogue data to obtain combined features of the training dialogue data.

The feature combination process of the statistical features of the training dialogue data is similar to the process of feature combination of the statistical features of the dialogue data to be processed. The specific implementation process of the above step 402 can refer to the specific description of the relevant steps in the embodiment shown in FIG. 4 , I won't repeat them here.

S403, using the combined features of the training dialogue data as training data, and using the labeling results as supervision information to perform model training to obtain a detection model.

Finally, the combined features of the training dialogue data can be used as training data, and the labeling results of the training dialogue data can be used as supervision information for supervised training. Optionally, the loss calculation can be performed according to the following formula, and the detection model training can be realized by gradient descent:

Among them, p' is the confidence degree of the detection result output by the detection model, indicating that there is a question-answer mismatch in the training dialogue sample, σ is the loss function used to calculate the loss value between the detection result and the standard result, x=W1'X1'+ W2'X2'+...+Wk'Xk'+b', X1', X2'...Xk' are the combined features of the training dialogue data, W1', W2'...Wk' are the weights of the detection model value, b' is the bias value of the detection model, and the weight value and bias value together constitute the model parameters of the detection model.

In this embodiment, the statistical features of the training dialogue data may be combined first to obtain combined features. Model training is then performed based on a larger number of combined features with richer meanings to improve the detection accuracy of the detection model.

In order to further improve the accuracy of the detection model, FIG. 6 is a flowchart of another model training provided by an embodiment of the present invention. This embodiment can also be executed by the above-mentioned training device. As shown in Figure 6, the steps may include the following steps:

S501. Obtain statistical features of training dialog data including at least one round of question-answering and an annotation result reflecting whether there is a question-answer mismatch in the training dialog data.

S502. Perform feature combination on statistical features of the training dialogue data to obtain combined features of the training dialogue data.

For the specific implementation process of the above step S501 to step S502, refer to the specific description of the relevant steps in the embodiment shown in FIG. 5 , which will not be repeated here.

S503, using the combined features of the training dialogue data as training data, and using the labeling results as supervision information to train a preset number of initial models respectively, so as to obtain candidate models with different model parameters.

Next, the training device can also obtain a preset number of initial models, and in an optional manner, the preset number of initial models can have the same model structure and different initial model parameters, then the training dialogue in step S502 can be The combined features of the data are used as training data, and the annotation results of the training dialogue data are used as supervision information to train a preset number of initial models respectively. Since the initial model parameters of the initial model are different, alternative models with different model parameters can be obtained even with the same training data.

In another optional manner, the training dialogue data in step S501 may be divided according to a preset number. Then, the combined features of the plurality of training dialog data are respectively used as training data of a preset number of initial models, and the labeling results of the training dialog data are used as supervision information to train a preset number of initial models respectively. Wherein, the preset number of initial models may have the same model structure and the same initial model parameters. Since the training dialogue data for training the initial model is different, the combined features of the training dialogue data are also different, and the model parameters of the final candidate models are different.

S504. Determine the parameters of the detection model according to the parameters of the candidate model to obtain a detection model.

Finally, the training device can determine the model parameters of the detection model according to the respective model parameters of the preset number of candidate models. Optionally, the model parameters of the preset number of candidate models can be averaged or weighted and summed, and the calculation results of the model parameters can be used as the model parameters of the detection model, thereby obtaining the detection model. Among them, the detection model has the same model structure as the initial model and the candidate model.

In this embodiment, the detection accuracy of the detection model is improved by feature combination and comprehensive consideration of model parameters, that is, parameter integration. Specifically, the training device first uses the combined features to train multiple candidate models, and then obtains the model parameters of the detection model by comprehensively considering the model parameters of the candidate models, thereby improving the accuracy of the model parameters of the detection model, that is, It is to improve the detection accuracy of the detection model.

In order to further improve the detection accuracy of the question-answer mismatch, in addition to the method shown in the embodiment shown in FIG. 6 , another optional method may be to comprehensively consider multiple detection results to obtain a more accurate detection result. Specifically, after training a preset number of candidate models according to the embodiment shown in FIG. 6 , each of the preset number of candidate models can also output the detection results of the dialogue data to be processed and the confidence of the detection results, Then, multiple detection results are integrated, such as averaging the confidence degrees or performing weighted summation on the confidence degrees, so as to finally obtain the detection result of the dialogue data to be processed.

For ease of understanding, the specific implementation process of the dialog data processing method provided by the above-mentioned embodiments will be described below as an example. The following process can also be understood in conjunction with FIG. 7 .

The dialogue system can be deployed on the cloud or on hardware devices to provide users with consulting services in different scenarios. The user and the dialogue system can generate the multi-segment dialogues given in the above-mentioned embodiments during the man-machine dialogue process. During the process of man-machine dialogue, the dialogue system can record the dialogue produced by both parties in real time, and can also make statistics on the dialogue data after the dialogue is over, so as to obtain the statistical characteristics of the dialogue data.

Then the processing device may obtain the statistical feature. Optionally, the processing device may directly use the statistical feature that is irrelevant to the dialogue content and is expressed as a value to determine whether there is a question-answer mismatch in the dialogue data. Optionally, the processing device may also perform feature combination on the statistical features to obtain a combined feature. The processing device can further utilize this combined feature to determine whether there is a question-answer mismatch in the dialog data. The detection result can also be displayed on the operation interface shown in FIG. 2 .

Optionally, the foregoing processing procedure executed by the processing device may specifically be executed by a detection model deployed in the processing device. As for the specific training process of the detection model, reference may be made to the relevant descriptions in the foregoing embodiments, which will not be repeated here.

The embodiments shown in FIG. 1 to FIG. 7 describe in detail the processing process of the dialog data by the processing device. On this basis, a human-computer interaction method can also be proposed starting from the user's interaction process. Then FIG. 8 is a flowchart of a human-computer interaction method provided by an embodiment of the present invention. The execution body of the method may specifically be a processing device providing an operation interface. As shown in Figure 8, the method may include the following steps:

S601, displaying a prompt message, the prompt message is generated when the amount of abnormal dialog data is greater than the preset amount, the abnormal dialog data includes dialog data that does not match the question and answer, the dialog data is generated by the user and the dialog system during the man-machine dialog, the dialog data Whether there is a question-answer mismatch is determined by the statistical characteristics of the dialogue data, and the statistical characteristics have nothing to do with the dialogue content of the dialogue data.

The processing device may detect whether there is a question-answer mismatch in the dialogue data to be processed generated in the historical time period according to the manners provided in the foregoing embodiments. Among them, the dialogue data to be processed with a question-answer mismatch may be referred to as abnormal dialogue data. If the amount of abnormal dialog data is greater than the preset threshold, a prompt message may be displayed on the operation interface provided by the processing device.

As shown in the operation interface in FIG. 2 , the prompt message may include a text prompt for optimizing the dialogue system. Optionally, other information related to the abnormal dialogue data can also be displayed on the statistical interface, such as the total number of dialogue data in the current historical time period, the proportion of abnormal dialogue data in the current historical time period, and the total number of abnormal dialogue data in the previous historical time period. Compared with the number of abnormal conversations, the rate of change of the number of abnormal conversations in the current historical time period, etc.

S602. Respond to the processing operation on the prompt message.

Then, the operation and maintenance personnel of the dialog system can trigger a processing operation on the prompt message, and the processing device can respond to the operation and maintenance personnel triggering an operation on the prompt message to determine whether to optimize the dialog system.

In this embodiment, by means of the operation interface provided by the processing device, the operation and maintenance personnel of the dialog system can interact with the processing device to determine whether to process the dialog system. At the same time, the operation and maintenance personnel can also accurately understand the performance of the dialogue system with the help of various information displayed on the operation interface. In addition, the content that is not described in detail in this embodiment and the technical effects that can be achieved can be referred to in the relevant descriptions in the above embodiments, and will not be repeated here.

Optionally, the dialog system can be deployed on the online customer service platform, and the prompt message can be displayed on the statistical interface provided by the online customer service platform.

The question-and-answer mismatch solution provided by the above embodiments can also be provided as a software operation service (Software as a Service, SaaS for short) to the operation and maintenance personnel of the online customer service platform deployed with the dialogue system or other online platforms. The service can also be deployed on a cloud-based service platform. Then FIG. 9 is a flowchart of another human-computer interaction method provided by an embodiment of the present invention. The execution subject of the method may be a service platform. As shown in Figure 9, the method may include the following steps:

S701. In response to the triggering of the detection operation, acquire the statistical features of the dialogue data to be processed including at least one round of question and answer, where the statistical features are not related to the dialogue content of the dialogue data to be processed.

S702. Detect whether there is a question-answer mismatch in the dialogue data to be processed according to statistical features.

S703. Display the detection result of the dialogue data to be processed.

Optionally, the operation and maintenance personnel can trigger the detection operation on the cloud operating platform, then the service platform can obtain the pending dialogue data including at least one round of question and answer, and detect whether the pending dialogue data is There is a question-answer mismatch. The final detection result can also be displayed on the statistics interface mentioned by the service platform.

The content not described in detail in this embodiment and the technical effects that can be achieved can be referred to in the relevant descriptions in the above embodiments, and will not be repeated here.

The dialogue data processing apparatus of one or more embodiments of the present invention will be described in detail below. Those skilled in the art can understand that these dialogue data processing devices can be configured by using commercially available hardware components through the steps taught in this solution.

FIG. 10 is a schematic structural diagram of a dialogue data processing device provided by an embodiment of the present invention. As shown in FIG. 10 , the device includes:

The feature acquisition module 11 is configured to acquire the statistical features of the dialogue data to be processed including at least one round of question and answer, the statistical features have nothing to do with the dialogue content of the dialogue data to be processed.

The determination module 12 is configured to determine whether there is a question-answer mismatch in the dialogue data to be processed according to the statistical features.

Optionally, the determining module 12 is configured to input the statistical features into a detection model, so that the detection model outputs whether there is a question-answer mismatch in the dialogue data to be processed.

Optionally, the device further includes: a training module 13, configured to obtain statistical features of training dialogue data including at least one round of question-and-answer and annotate results reflecting whether there is a question-answer mismatch in the training dialogue data; The statistical characteristics of the data are used as training data, and the labeling results are used as supervision information to train a preset number of initial models respectively to obtain candidate models with different model parameters; determine the detection according to the model parameters of the candidate models The model parameters of the model to get the detection model.

Optionally, the determining module 12 is configured to perform feature combination on any at least two of various statistical features to obtain the combined features of the dialogue data to be processed; determine according to the combined features of the dialogue data to be processed Whether there is a question-answer mismatch in the dialogue data to be processed.

Optionally, the determination module 12 is configured to input the statistical features into a detection model, so that the statistical features are combined by the detection model; input the combined features into the detection model, so that the statistical features can be combined by the detection model. The detection model outputs a detection result reflecting whether there is a question-answer mismatch in the dialogue data to be processed.

Optionally, the training module 13 is configured to acquire statistical features of training dialogue data including at least one round of question and answer and annotate results reflecting whether there is a question-answer mismatch in the training dialogue data; Combining features to obtain the combined features of the training dialogue data; using the combined features of the training dialogue data as training data, and using the labeling results as supervision information to perform model training to obtain the detection model.

Optionally, the training module 13 is configured to use the combined features of the training dialogue data as training data, and use the labeling results as supervision information to train a preset number of initial models respectively, so as to obtain alternatives of different model parameters. A model: determining parameters of the detection model according to parameters of the candidate model to obtain the detection model.

The acquiring module 11 is configured to acquire the statistical features output by the dialogue system, and the statistical features are output by the dialogue system after the dialogue data to be processed is counted after the man-machine dialogue ends. Wherein, the dialogue data to be processed is generated by the user and the dialogue system during the man-machine dialogue.

The device also includes: a statistical module 14 and an information display module 15 .

The statistical module 14 is used to count the number of abnormal dialog data with question-answer mismatch in the historical time period.

The information display module 15 is used to display the quantity of the abnormal dialogue data on the statistical interface; if the quantity of the abnormal dialogue data is greater than a preset quantity, it is displayed on the statistical interface to optimize the dialogue system prompt message.

Wherein, the dialog data to be processed includes text or voice generated during the man-machine dialog between the user and the dialog system; the statistical features of the dialog data to be processed include numeric features.

The device shown in FIG. 10 can execute the method of the embodiment shown in FIG. 1 to FIG. 6 . For parts not described in detail in this embodiment, reference can be made to relevant descriptions of the embodiment shown in FIG. 1 to FIG. 6 . For the execution process and technical effect of this technical solution, refer to the description in the embodiment shown in FIG. 1 to FIG. 6 , which will not be repeated here.

Fig. 11 is a schematic structural diagram of a human-computer interaction device provided by an embodiment of the present invention. As shown in Fig. 11, the device includes:

The display module 21 is used to display a prompt message, the prompt message is generated when the amount of abnormal dialogue data is greater than a preset number, the abnormal dialogue data includes dialogue data that does not match the question and answer, and the dialogue data is generated by the user and the dialogue system in person Generated during a computer-to-computer dialogue, whether there is a question-answer mismatch in the dialogue data is determined by the statistical characteristics of the dialogue data, and the statistical characteristics have nothing to do with the dialogue content of the dialogue data.

The response module 22 is configured to respond to the processing operation on the prompt message.

The device shown in FIG. 11 can execute the method of the embodiment shown in FIG. 8 . For parts not described in detail in this embodiment, refer to the relevant description of the embodiment shown in FIG. 8 . For the execution process and technical effect of this technical solution, refer to the description in the embodiment shown in FIG. 8 , and details are not repeated here.

In a possible design, the dialog data processing methods provided in the foregoing embodiments may be applied in an electronic device, and as shown in FIG. 12 , the electronic device may include: a first processor 31 and a first memory 32 . Wherein, the first memory 32 is used to store the program that supports the electronic device to execute the dialog data processing method provided in the above-mentioned embodiments shown in FIGS. 1 to 6 , and the first processor 31 is configured to execute the stored program.

The program includes one or more computer instructions, wherein, when one or more computer instructions are executed by the first processor 31, the following steps can be realized:

Obtaining statistical features of the dialogue data to be processed including at least one round of question and answer, the statistical features having nothing to do with the dialogue content of the dialogue data to be processed;

Determine whether there is a question-answer mismatch in the dialogue data to be processed according to the statistical features.

Optionally, the first processor 31 is further configured to execute all or part of the steps in the foregoing embodiments shown in FIGS. 1 to 6 .

Wherein, the structure of the electronic device may further include a first communication interface 33, which is used for the electronic device to communicate with other devices or communication systems.

In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions used by the above-mentioned electronic device, which includes a program for executing the dialog data processing method shown in FIGS. 1 to 6 above.

In a possible design, the human-computer interaction methods provided in the foregoing embodiments may be applied in another electronic device, and as shown in FIG. 13 , the electronic device may include: a second processor 41 and a second memory 42 . Wherein, the second memory 42 is used to store a program that supports the electronic device to execute the human-computer interaction method provided in the embodiment shown in FIG. 8 , and the second processor 41 is configured to execute the program stored in the second memory 42 .

The program includes one or more computer instructions, wherein, when the one or more computer instructions are executed by the second processor 41, the following steps can be realized:

Displaying a prompt message, the prompt message is generated when the amount of abnormal dialogue data is greater than a preset number, the abnormal dialogue data includes dialogue data that does not match the question and answer, and the dialogue data is generated by the user and the dialogue system during the man-machine dialogue Generate, whether there is a question-answer mismatch in the dialogue data is determined by the statistical characteristics of the dialogue data, and the statistical characteristics have nothing to do with the dialogue content of the dialogue data;

Respond to the processing operation on the prompt message.

Optionally, the second processor 41 is further configured to execute all or part of the steps in the foregoing embodiment shown in FIG. 8 .

Wherein, the structure of the electronic device may further include a second communication interface 43, which is used for the electronic device to communicate with other devices or communication systems.

In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions used by the above-mentioned electronic device, which includes a program for executing the above-mentioned human-computer interaction method shown in FIG. 8 .

Fig. 14 is a schematic structural diagram of another human-computer interaction device provided by an embodiment of the present invention. As shown in Fig. 14, the device includes:

The acquiring module 51 is configured to acquire, in response to the trigger of the detection operation, statistical features of the dialogue data to be processed including at least one round of question and answer, the statistical features being irrelevant to the dialogue content of the dialogue data to be processed.

The detection module 52 is configured to detect whether there is a question-answer mismatch in the dialogue data to be processed according to the statistical characteristics.

The result display module 53 is configured to display the detection result of the dialogue data to be processed.

The device shown in FIG. 14 can execute the method of the embodiment shown in FIG. 9 . For parts not described in detail in this embodiment, reference can be made to relevant descriptions of the embodiment shown in FIG. 9 . For the execution process and technical effect of this technical solution, refer to the description in the embodiment shown in FIG. 9 , and details are not repeated here.

In a possible design, the human-computer interaction methods provided in the foregoing embodiments may be applied in yet another electronic device. As shown in FIG. 15 , the electronic device may include: a third processor 61 and a third memory 62 . Wherein, the third memory 62 is used to store a program that supports the electronic device to execute the human-computer interaction method provided in the embodiment shown in FIG. 9 above, and the third processor 61 is configured to execute the program stored in the third memory 62 .

The program includes one or more computer instructions, wherein, when one or more computer instructions are executed by the third processor 61, the following steps can be realized:

Responsive to the triggering of the detection operation, acquiring statistical features of the dialogue data to be processed including at least one round of question and answer, the statistical features having nothing to do with the dialogue content of the dialogue data to be processed;

Detecting whether there is a question-answer mismatch in the dialogue data to be processed according to the statistical characteristics;

Display the detection result of the dialogue data to be processed.

Optionally, the third processor 61 is further configured to execute all or part of the steps in the foregoing embodiment shown in FIG. 9 .

Wherein, the structure of the electronic device may further include a third communication interface 63, which is used for the electronic device to communicate with other devices or communication systems.

In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions used by the above-mentioned electronic device, which includes a program for executing the above-mentioned human-computer interaction method shown in FIG. 9 .

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (14)

1. A dialogue data processing method, characterized in that, comprising: Obtaining statistical features of the dialogue data to be processed including at least one round of question and answer, the statistical features having nothing to do with the dialogue content of the dialogue data to be processed; Determine whether there is a question-answer mismatch in the dialogue data to be processed according to the statistical features. 2. The method according to claim 1, wherein the determining whether there is a question-answer mismatch in the dialogue data to be processed according to the statistical characteristics comprises: The statistical features are input into a detection model, so that the detection model outputs whether there is a question-answer mismatch in the dialog data to be processed. 3. The method according to claim 2, wherein the method further comprises: Obtaining the statistical features of the training dialog data including at least one round of question-and-answer and annotating results reflecting whether there is a question-answer mismatch in the training dialog data; Using the statistical features of the training dialogue data as training data, and using the labeling results as supervision information to train a preset number of initial models, respectively, to obtain alternative models of different model parameters; Determining model parameters of the detection model according to model parameters of the candidate model to obtain the detection model. 4. The method according to claim 1, wherein the statistical features include a plurality; determining whether there is a question-and-answer mismatch in the dialogue data to be processed according to the statistical features includes: Combining any at least two of the various statistical features to obtain the combined features of the dialogue data to be processed; Determine whether there is a question-answer mismatch in the dialogue data to be processed according to the combination feature of the dialogue data to be processed. 5. The method according to claim 4, wherein said combination of the statistical features to obtain the combined features of the dialogue data to be processed comprises: inputting the statistical features into a detection model to perform feature combination on the statistical features by the detection model; The determining whether there is a question-answer mismatch in the dialogue data to be processed according to the combination feature includes: The combined feature is input into the detection model, so that the detection model outputs a detection result reflecting whether there is a question-answer mismatch in the dialogue data to be processed. 6. The method according to claim 5, further comprising: Obtaining the statistical features of the training dialog data including at least one round of question-and-answer and annotating results reflecting whether there is a question-answer mismatch in the training dialog data; performing feature combination on statistical features of the training dialogue data to obtain combined features of the training dialogue data; The combined features of the training dialogue data are used as training data, and the labeling results are used as supervision information to perform model training to obtain the detection model. 7. The method according to claim 6, wherein the combined features of the training dialogue data are used as training data, and the labeling results are used as supervision information to carry out model training to obtain the detection model, including : Using the combined features of the training dialogue data as training data, and using the labeling results as supervisory information to train a preset number of initial models respectively to obtain alternative models of different model parameters; The parameters of the detection model are determined according to the parameters of the candidate model to obtain the detection model. 8. The method according to claim 1, wherein the dialogue data to be processed is generated by the user and the dialogue system during the man-machine dialogue; The acquisition of the statistical characteristics of the dialogue data to be processed includes: The statistical features output by the dialog system are acquired, and the statistical features are output by the dialog system after the dialog data to be processed is counted after the man-machine dialog ends. 9. The method of claim 8, further comprising: Count the number of abnormal dialogue data with question-answer mismatch in the historical time period; Displaying the quantity of the abnormal dialogue data on the statistics interface; If the quantity of the abnormal dialogue data is greater than a preset quantity, a prompt message for optimizing the dialogue system is displayed on the statistics interface. 10. The method according to claim 1, wherein the dialogue data to be processed includes text or voice generated by the user and the dialogue system during the man-machine dialogue; the statistical characteristics of the dialogue data to be processed include numerical types feature. 11. A human-computer interaction method, comprising: Displaying a prompt message, the prompt message is generated when the amount of abnormal dialogue data is greater than a preset number, the abnormal dialogue data includes dialogue data that does not match the question and answer, and the dialogue data is generated by the user and the dialogue system during the man-machine dialogue, Whether there is a question-answer mismatch in the dialogue data is determined by the statistical characteristics of the dialogue data; Respond to the processing operation on the prompt message. 12. The method according to claim 11, wherein the dialogue system is deployed in an online customer service platform; and the display prompt message includes: The prompt message is displayed on the statistical interface provided by the online customer service platform. 13. An electronic device, characterized by comprising: a memory and a processor; wherein, executable code is stored on the memory, and when the executable code is executed by the processor, the processor is made to execute The dialog data processing method according to any one of claims 1-10, or causing the processor to execute the human-computer interaction method according to claim 11 or 12. 14. A non-transitory machine-readable storage medium, wherein executable code is stored on the non-transitory machine-readable storage medium, and when the executable code is executed by a processor of an electronic device, the The processor executes the dialog data processing method according to any one of claims 1-10, or causes the processor to execute the human-computer interaction method according to claim 11 or 12.