A method and system for optimizing scene interaction of virtual digital human

Abstract

The present invention discloses a scene interaction optimization method and system for a virtual digital human, belonging to the field of artificial intelligence technology. The method includes: collecting the interruption reason categories and interruption durations of the virtual digital human in different states, and classifying the interruption durations into a set of corresponding states to obtain a set of interruption durations corresponding to each interruption reason category; constructing a normal distribution density function of the corresponding category, and establishing a judgment model for the interruption reason category; obtaining the interruption duration of a new interaction process, and predicting the interruption reason category through the judgment model for the interruption reason category; and providing corresponding state transition prompt words and actions for the new interaction process according to the interruption reason category. The present invention can predict the interruption reason through the judgment model for the interruption reason category according to the interruption duration of the interaction process, and provide appropriate prompt words and actions for the state transition of the virtual digital human, thereby improving the interaction performance of the virtual digital human.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a scene interaction optimization method and system for a virtual digital human. Background Art

[0002] Virtual humans refer to digital characters or humanoid images generated by computer technology. They can be used in virtual reality, augmented reality, video games, social media and other scenarios. Virtual humans usually have human characteristics, such as facial expressions, movements, and voices, and can interact with users through machine question-answering and natural language processing. They have a wide range of applications, including online customer service, entertainment industry, education and training, marketing and promotion, etc.

[0003] Users input commands or information through voice, text or gestures to interact with virtual digital humans in a question-and-answer manner. During the interaction with virtual digital humans, users often interrupt them because the information provided by them may not meet their requirements. In order to improve the user experience, virtual digital humans will provide prompt words, actions and even animations to interact with users. However, these prompt words or actions are often fixed and do not take into account the specific reasons for user interruptions, resulting in inaccurate prompt words and actions provided by virtual digital humans, and the interaction process is often not smooth.

[0004] Therefore, a scene interaction optimization method for virtual digital humans is needed to improve the interaction ability of virtual digital humans in scenes where they are interrupted by users. Summary of the invention

[0005] In view of this, the present invention proposes a scene interaction optimization method and system for a virtual digital human, which is used to solve the problem that the virtual digital human provides inaccurate prompt words and actions after being interrupted by a user.

[0006] In a first aspect, the present invention discloses a method for optimizing scene interaction of a virtual digital human, the method comprising:

[0007] Collect the reasons and duration of interruptions of virtual digital humans in different states;

[0008] According to the interruption reason categories in the different states, the interruption durations are classified into sets of corresponding states to obtain an interruption duration set corresponding to each interruption reason category;

[0009] According to the interruption duration set corresponding to each interruption reason category, construct a normal distribution density function of the corresponding category;

[0010] Establishing a judgment model of the interruption reason category according to the normal distribution density function of the corresponding category;

[0011] Obtaining the interruption duration of the new interaction process, and predicting the interruption reason category through the judgment model of the interruption reason category;

[0012] According to the interruption reason category, corresponding state transition prompt words and actions are provided for the new interaction process.

[0013] On the basis of the above technical solution, preferably, the states of the virtual digital human are divided into four states: sleeping state, waiting state, thinking state and answering state, wherein being interrupted by the user occurs in the thinking state and the answering state;

[0014] The categories of reasons why the virtual digital human is interrupted in different states specifically include:

[0015] When in the thinking state, the reasons for being interrupted include the user wanting to change the question, the user thinking that the thinking time is too long, and other reasons;

[0016] When answering, the reasons for interruption include: the user wants to change the question, the user thinks the understanding is wrong, the user wants to refine the answer, and other reasons;

[0017] The interruption duration is the duration from when the virtual digital human changes to the current state to when it is interrupted in the current state.

[0018] On the basis of the above technical solution, preferably, the collection of the categories of interruption reasons and interruption durations of the virtual digital human in different states specifically includes:

[0019] In the thinking state or answering state, record the interruption duration and determine the type of interruption reason;

[0020] In the above-mentioned thinking state, the method for determining the type of interruption reason is as follows:

[0021] After the virtual digital human is interrupted and returns to the waiting state, the question inputted by the user is obtained again. If the similarity between the questions inputted by the user before and after the interruption is lower than the first preset threshold, the interruption reason category is that the user wants to change the question; if the question inputted by the user again contains a specific urging keyword, the interruption reason category is that the user thinks that the thinking time is too long; those that do not belong to the above two categories are uniformly classified as other reasons;

[0022] In the answering state, the method for determining the type of interruption reason is as follows:

[0023] After the virtual digital human is interrupted and returns to the waiting state, the question input by the user is obtained again. If the similarity between the questions input by the user before and after the interruption is lower than the second preset threshold, the category of the interruption reason is that the user wants to change the question; if the similarity between the question input by the user after the interruption and the answer output by the virtual digital human is greater than or equal to the first preset threshold, and the question input by the user again contains specific negative keywords, the category of the interruption reason is that the user believes that the understanding is wrong; if the similarity between the question input by the user after the interruption and the answer output by the virtual digital human is greater than or equal to the second preset threshold, and the question input by the user again contains specific detailed keywords, the category of the interruption reason is that the user wants to refine the answer; those that do not belong to the above two categories are uniformly classified as other reasons.

[0024] On the basis of the above technical solution, preferably, constructing a normal distribution density function of a corresponding category according to the interruption duration set corresponding to each interruption reason category specifically includes:

[0025] Calculate the mean and standard deviation of the interruption duration set corresponding to each interruption reason category;

[0026] The normal distribution density function of the corresponding category is constructed according to the mean and standard deviation, and the formula is:

[0027]

[0028] Among them, i is the serial number of the interruption reason category, f i (t) is the normal distribution density function of the interruption reason category i, μ i is the mean of the interruption duration of interruption reason category i, σ i is the standard deviation of the interruption duration set of interruption reason category i, t is the interruption duration, and exp is the exponential function.

[0029] On the basis of the above technical solution, preferably, the formula of the judgment model of the interruption reason category is:

[0030]

[0031] Wherein, i=1,2,…,n, n is the total number of interruption reason categories, t is the interruption duration, and P(i|t) is the probability that the interruption reason category belongs to the i-th category under the condition that the interruption duration is t.

[0032] On the basis of the above technical solution, preferably, the predicting of the interruption cause category by the judgment model of the interruption cause category specifically includes:

[0033] After obtaining the interruption duration t0 of the new interaction process, the normal distribution density function value f corresponding to each category at t0 is calculated respectively. i (t0);

[0034] The normal distribution density function value f i (t0) is substituted into the judgment model of the interruption reason category, and the probability of belonging to each interruption reason category under the condition of interruption duration t0 is calculated respectively;

[0035] The category with the highest probability is selected as the category of the interruption reason.

[0036] On the basis of the above technical solution, preferably, the method further comprises:

[0037] Obtain the question input again by the user after the interruption of the new interaction process, and verify whether the judgment of the judgment model of the interruption reason category is correct. If it is incorrect, add the interruption duration of the new interaction process to the interruption duration set of the corresponding category, and update the corresponding mean and standard deviation.

[0038] In a second aspect, the present invention discloses a scene interaction system for a virtual digital human, the system comprising:

[0039] Data collection module: used to collect the interruption reason categories and interruption durations of the virtual digital human in different states; according to the interruption reason categories in the different states, the interruption durations are classified into sets of corresponding states to obtain a set of interruption durations corresponding to each interruption reason category;

[0040] Model construction module: used to construct a normal distribution density function of the corresponding category according to the interruption duration set corresponding to each interruption reason category; and establish a judgment model of the interruption reason category according to the normal distribution density function of the corresponding category;

[0041] Interaction optimization module: used to obtain the interruption duration of the new interaction process, predict the interruption reason category through the judgment model of the interruption reason category; provide corresponding state change prompt words and actions for the new interaction process according to the interruption reason category.

[0042] In a third aspect of the present invention, an electronic device is disclosed, comprising: at least one processor, at least one memory, a communication interface and a bus;

[0043] Wherein, the processor, memory, and communication interface communicate with each other via the bus;

[0044] The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to the first aspect of the present invention.

[0045] According to a fourth aspect of the present invention, a computer-readable storage medium is disclosed, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method described in the first aspect of the present invention.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1) The present invention constructs a normal distribution density function of the corresponding category through the interruption duration corresponding to each interruption reason category, and establishes a judgment model of the interruption reason category based on this. After the virtual digital human is interrupted in different states, the interruption reason can be predicted according to the interruption duration of the interaction process through the judgment model of the interruption reason category, and appropriate prompt words and actions are provided for the state transition of the virtual digital human, so as to quickly switch to the state desired by the user, simplifying the existing interaction process of the virtual digital human, in which the interaction process of the state change needs to be changed through multiple prompts after the user interrupts, thereby enhancing the interactivity between the user and the virtual digital human.

[0048] 2) The present invention records the duration of interruptions by the user in the thinking state or the answering state, obtains the question input again by the user after the virtual digital human returns to the waiting state after being interrupted, and determines the category of the interruption reason according to the similarity between the questions input before and after the interruption and the keywords in the questions, thereby quickly identifying the user's intention and facilitating the subsequent rapid prediction of the user's interruption intention.

[0049] 3) The present invention calculates the mean and standard deviation of the interruption duration set corresponding to each interruption reason category respectively, constructs the normal distribution density function of the corresponding category according to the mean and standard deviation, thereby establishing a mapping relationship between the interruption duration and the probability distribution, and finally constructs a judgment model of the interruption reason category according to the normal distribution density function, so as to calculate the probability of the interruption reason category belonging to each category under the condition of known interruption duration, and take the one with the largest probability as the interruption reason category judgment result, so as to realize the early prediction of the user interruption reason during the digital human interaction process, timely provide the user with prompt words and actions corresponding to the user's intention, and improve the interaction efficiency of the virtual digital human. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A schematic diagram of the state transition of a virtual digital human of the present invention;

[0052] Figure 2 It is a flow chart of the scene interaction optimization method of the virtual digital human of the present invention;

[0053] Figure 3 This is another flow chart of a scene interaction optimization method for a virtual digital human of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] According to the interaction process of the virtual digital human, the present invention divides the states of the virtual digital human into four types: sleep state, waiting state, thinking state and answering state. The virtual digital human will not receive any questions in the sleep state, and can only enter the waiting state through a specific wake-up word. After starting the virtual digital human program, the delicate digital human enters the waiting state, at which time it can receive information input by the user. After receiving, the virtual digital human enters the thinking state and transmits the data input by the user to the background server for calculation of the large language model. After the calculation is completed, the virtual digital human enters the answering state and outputs the answer result through language. After the answer is completed, the virtual digital human returns to the waiting state again. If the waiting time exceeds the preset threshold, the digital human will re-enter the sleep state. In addition, in the thinking state or the answering state, the virtual digital human can be interrupted by the user and can return to the waiting state again, which is convenient for the user to re-enter new questions.

[0056] Figure 1 The figure shows a schematic diagram of the state transition of a virtual digital human. During these state transitions, the virtual digital human uses different prompt words and actions to indicate the state changes. Among them, the thinking state and the answering state are most easily interrupted by the user, and common interruption keywords include "stop", "interrupt" and "restart". Once interrupted, the virtual digital human needs to return to the waiting state and use new prompt words and actions to indicate the state change. In order to enhance the interactivity between the user and the virtual digital human, the present invention analyzes and predicts the reasons for the user's interruption, so that before returning to the waiting state, corresponding prompt words and actions are provided according to the reasons for the interruption to indicate the state change, so as to improve the interaction efficiency.

[0057] See also Figure 2 The present invention discloses a scene interaction optimization method for a virtual digital human, characterized in that the method comprises:

[0058] S1. Collect the categories of interruption reasons and interruption duration of virtual digital humans in different states.

[0059] In the embodiment of the present invention, the states of the virtual digital human are divided into four states: sleeping state, waiting state, thinking state and answering state, among which, being interrupted by the user occurs in the thinking state and the answering state;

[0060] In the thinking state or answering state, the duration from when the virtual digital human changes to the current state to when it is interrupted in the current state is recorded, i.e., the interruption duration. Specifically, in the embodiment of the present invention, the interruption duration refers to the duration from when the digital human enters the thinking or answering state, starting from 0 seconds, until it is interrupted, in seconds, accurate to one decimal place. If the digital human enters the thinking or answering state multiple times, the timing is restarted each time.

[0061] Then determine the categories of reasons why the virtual digital person is interrupted in different states.

[0062] Specifically, when in the thinking state, the reasons for interruption include three categories: the user wants to change the question, the user thinks the thinking time is too long, and other reasons; when in the answering state, the reasons for interruption include four categories: the user wants to change the question, the user thinks there is a misunderstanding, the user wants to refine the answer, and other reasons.

[0063] In the thinking state, the way to determine the type of interruption reason is:

[0064] After the virtual digital human is interrupted and returns to the waiting state, the question input by the user is obtained again. If the similarity between the questions input by the user before and after the interruption is lower than a preset threshold, the interruption reason category is that the user wants to change the question; if the question input by the user again contains specific urging keywords, such as "as soon as possible" or "brief", the interruption reason category is that the user thinks that the thinking time is too long; those that do not belong to the above two categories are uniformly classified as other reasons;

[0065] In the answering state, the method for determining the type of interruption reason is as follows:

[0066] After the virtual digital human is interrupted and returns to the waiting state, the question input by the user is obtained again. If the similarity between the questions input by the user before and after the interruption is lower than the second preset threshold, the category of the interruption reason is that the user wants to change the question; if the similarity between the question input by the user after the interruption and the answer output by the virtual digital human is greater than or equal to the first preset threshold, and the question input by the user again contains specific negative keywords, such as "error", "incorrect", etc., the category of the interruption reason is that the user believes that the understanding is wrong; if the similarity between the questions input by the user before and after the interruption is greater than or equal to the second preset threshold, and the question input by the user again contains specific detailed keywords, such as "refinement", "detailed explanation", etc., the category of the interruption reason is that the user wants to refine the answer; those that do not belong to the above two categories are uniformly classified as other reasons.

[0067] The present invention records the duration of interruptions by the user in a thinking state or an answering state, obtains the question input again by the user after the virtual digital person returns to a waiting state after being interrupted, and determines the category of the interruption reason according to the similarity between the questions input before and after the interruption and the keywords in the questions, thereby quickly identifying the user's intention and facilitating the subsequent rapid prediction of the user's interruption intention.

[0068] S2. Classify the interruption durations into sets of corresponding states according to the interruption reason categories in the different states, and obtain an interruption duration set corresponding to each interruption reason category.

[0069] Classify the interruption durations corresponding to the interruption reason category i in the thinking state and the answering state into the corresponding categories respectively, and obtain the interruption duration set C corresponding to each interruption reason category i i .

[0070] S3. According to the interruption duration set corresponding to each interruption reason category, construct a normal distribution density function of the corresponding category.

[0071] In the embodiment of the present invention, the mean value μ of the interruption duration set corresponding to each interruption reason category i is calculated respectively. i and standard deviation σ i , and then construct the normal distribution density function of the corresponding category according to the mean and standard deviation. The formula is:

[0072]

[0073] Among them, i is the serial number of the interruption reason category, f i (t) is the normal distribution density function of the interruption reason category i, μ i is the mean of the interruption duration of interruption reason category i, σ iis the standard deviation of the interruption duration set of interruption reason category i, t is the interruption duration, and exp is the exponential function.

[0074] S4. Establishing a judgment model for the interruption reason category according to the normal distribution density function of the corresponding category.

[0075] In an embodiment of the present invention, the formula of the judgment model of the interruption reason category is:

[0076]

[0077] Wherein, i=1,2,…,n, n is the total number of interruption reason categories, t is the interruption duration, and P(i|t) is the probability that the interruption reason category belongs to the i-th category under the condition that the interruption duration is t.

[0078] S5. Setting prompt words and actions for the state transition of the virtual digital human.

[0079] Specifically, in the thinking state and answering state of the virtual digital human, the present invention dynamically sets transition prompt words and actions according to the category of the interruption reason after the user interrupts the digital human and before the digital human enters the waiting state, so that corresponding transition prompt words and actions are given according to the category of the interruption reason later.

[0080] In the embodiment of the present invention, the format of the transition prompt word setting is: the current state of the virtual digital human, the category of the interruption reason, the prompt word and the action number, and different action numbers correspond to different actions of the virtual digital human. The format of the action setting of the virtual digital human in state transition is: the current state of the virtual digital human, the next state of the virtual digital human, the prompt word and the action number.

[0081] S6. Obtain the interruption duration of the new interaction process, and predict the interruption reason category through the judgment model of the interruption reason category.

[0082] In the embodiment of the present invention, the virtual digital human monitors the user interruption event in the new interaction process, obtains the interruption duration t0 of the new interaction process, and calculates the normal distribution density function value f corresponding to each category at t0. i (t0); then, the normal distribution density function value f i (t0) is substituted into the judgment model of the interruption reason category, and the probability of belonging to each interruption reason category under the condition of interruption duration t0 is calculated respectively; finally, the category with the largest probability is selected as the interruption reason category.

[0083] S7. Providing corresponding state transition prompt words and actions for the new interaction process according to the interruption reason category.

[0084] Since the virtual human needs to quickly perform the transition and say the prompt word after the user interrupts, the reason for the interruption must be predicted in a very short time. Generally, the time from the user interrupting to the start of the transition by the virtual human should not exceed 0.3 seconds, otherwise the user will feel a lag.

[0085] The present invention calculates the mean and standard deviation of the interruption duration set corresponding to each interruption reason category respectively, constructs the normal distribution density function of the corresponding category according to the mean and standard deviation, thereby establishing a mapping relationship between the interruption duration and the probability distribution, and finally constructs a judgment model of the interruption reason category according to the normal distribution density function, so as to calculate the probability that the interruption reason category belongs to each category under the condition of known interruption duration, and take the one with the largest probability as the interruption reason category judgment result, so as to realize the early prediction of the user interruption reason in the digital human interaction process, timely provide the user with prompt words and actions corresponding to the user's intention, and improve the interaction efficiency of the virtual digital human.

[0086] S7: obtaining the question input again by the user after the interruption of the new interaction process, and verifying whether the judgment model of the interruption reason category is correct.

[0087] Specifically, if correct, restart monitoring the user interruption event; if incorrect, add the interruption duration of the new interaction process to the interruption duration set of the corresponding category, and update the corresponding mean and standard deviation to improve the accuracy of the next prediction.

[0088] The present invention constructs a normal distribution density function of the corresponding category through the interruption duration corresponding to each interruption reason category, and establishes a judgment model of the interruption reason category based on this. After the virtual digital human is interrupted in different states, the interruption reason can be predicted according to the interruption duration of the interaction process through the judgment model of the interruption reason category, and appropriate prompt words and actions are provided for the state change of the virtual digital human, so as to quickly switch to the state desired by the user, simplifying the existing virtual digital human interaction process, in which the user needs to change the state through multiple prompts after the interruption, thereby improving the interaction performance between the user and the virtual digital human.

[0089] The specific implementation steps of a scene interaction optimization method for a virtual digital human of the present invention are further described below with reference to specific examples. Figure 3 Shown is a flow chart of another method for optimizing scene interaction of a virtual digital human according to the present invention.

[0090] Step 1, collect historical interruption data: collect data on the virtual digital person being interrupted by the user in the thinking state and the answering state, including the current state of the virtual digital person, the category of the interruption reason and the interruption duration. After completion, proceed to step 2.

[0091] Step 2, data classification: Classify all data according to the current state of the virtual digital person and the reason for interruption. In the thinking state, there are three types of interruption reasons; in the answering state, there are five types of interruption reasons. Although some interruption reasons may be the same in the two states, they need to be classified separately. For example, "the user wants to change the question" in the thinking state and the same reason in the answering state should be regarded as different categories. After classification, only the interruption duration is retained in the interruption duration set corresponding to each category. Generally speaking, the interruption duration set corresponding to each category has at least 20 records. After completion, proceed to step 3.

[0092] Step 3, calculate the mean and standard deviation of each interruption duration set: For each interruption duration in each category, calculate its mean and standard deviation. In this way, each category of records will get the corresponding mean and standard deviation. Finally, all categories of data will form a new set, and each data includes the current status of the virtual digital human, the reason for the interruption, the mean and standard deviation. After completion, proceed to step 4.

[0093] Step 4, set the transition prompt words and actions: In this step, set the prompt words and actions of the virtual digital human when being interrupted. The format of the prompt word setting is: the current state of the virtual digital human, the category of the interruption reason, the prompt word and the action number. Different action numbers correspond to different virtual digital human actions. In addition, it is necessary to set the actions of the virtual digital human in state transition, and the format is: the current state of the virtual digital human, the next state of the virtual digital human, the prompt word and the action number. Through this setting, after determining the category of the interruption reason, you can get the corresponding prompt word and action number, thereby realizing dynamic interaction. After completion, proceed to step 5.

[0094] Step 5, start the virtual digital human: start the virtual digital human and load various setting data to make it enter the start state. After the start is complete, the virtual digital human will be in a waiting state, waiting for the user to enter questions. After completion, proceed to step 6.

[0095] Step 6, monitor user interruption events: When the virtual digital person is in the thinking or answering state, start a high-priority monitoring thread to determine whether the user interrupts the virtual digital person by using the interruption keyword. If a user interruption is detected, record the current state of the virtual digital person and the interruption duration. The interruption duration starts from the time the virtual digital person enters the thinking or answering state, starting at 0 seconds, and the time when the interruption statement is received is the interruption duration. Each time an interruption event is monitored, the event is immediately emitted and waits for the program that receives the interruption message to process. After completion, proceed to step 7.

[0096] Step 7, receiving user interruption messages: start a high-priority process to specifically receive and transmit user interruption event messages. The process receives the interruption event and obtains the transmitted data, including the current state of the virtual digital human, the interruption duration, and the attached parameters. If the virtual digital human is in a thinking state, the attached parameters are the last user's question; if the virtual digital human is in an answering state, the attached parameters include the last user's question and the answer that the virtual digital human has output, excluding the answer that has not been output. After completion, proceed to step 8.

[0097] Step 8, obtain the mean and variance of the interruption duration set of each category in the current state: according to the current state of the virtual digital human, obtain the mean and variance of the interruption duration corresponding to all interruption reason categories in this state. After completion, proceed to step 9.

[0098] Step 9, calculate the probability of each category: first use formula (1) to calculate the value of each category of reasons under the interruption duration, and then use formula (2) to calculate the probability of belonging to each category of interruption reasons under the condition of the interruption duration. After completion, proceed to step 10.

[0099] Step 10, determine the category of the interruption reason: select the maximum value from each probability to determine the category of the interruption reason. According to the current state of the virtual digital human and the category of the interruption reason, find the corresponding prompt word and action number. After completion, proceed to step 11.

[0100] Step 11, the virtual digital human performs the transition action: the virtual digital human outputs the prompt word and performs the corresponding action according to the prompt word and action number. After completion, proceed to step 12.

[0101] Step 12, obtaining the question input by the user again: At this time, the virtual digital human is in a waiting state, and the user interrupts and re-enters the information, and obtains the user's new input. After completion, proceed to step 13.

[0102] Step 13, determine the actual reason for the user interruption: determine based on the attached parameters of the interruption message transmission:

[0103] If the state passed is the thinking state:

[0104] (1) Determine whether the user's re-entry contains keywords such as "as soon as possible" and "brief". If so, it is considered that the interruption reason category is that the user believes that the thinking time is too long.

[0105] (2) If the similarity between the word vector input by the user again and the word vector input last time (hidden in the auxiliary parameters) is less than 0.8, it is considered that the user wants to change the question.

[0106] (3) If the above two conditions are not met, it will be classified as other reasons.

[0107] If the status passed is the answer status:

[0108] (1) If the word vector similarity of the question input by the user before and after the interruption is less than 0.8, it is considered that the user wants to change the question.

[0109] (2) Compare the word vectors of the user input and the answer output by the virtual digital human (also hidden in the auxiliary parameters). If the similarity is higher than 0.8 and the user input contains keywords such as "error" and "incorrect", it is considered that the user feels that the understanding is wrong.

[0110] (3) If the word vector similarity of the questions input by the user before and after the interruption is higher than 0.8, and the question input by the user again contains keywords such as "refine" and "detailed explanation", it is considered that the user wants to refine the answer.

[0111] (4) If the above conditions are not met, it will be classified as other reasons.

[0112] After the judgment is completed, go to step 14.

[0113] Step 14, verify the interruption reason: compare whether the predicted interruption reason category is consistent with the actual reason determined in step 13. If they are consistent, it means that the prediction result is correct, and return to step 6 to continue monitoring the interruption event; if they are inconsistent, it means that the prediction result is inaccurate, and go to step 15.

[0114] Step 15, update the mean and variance: Since the interruption reason category predicted by the interruption duration is incorrect, add the interruption duration to the interruption duration set corresponding to the category, and recalculate the mean and variance of the category to make future predictions more accurate. Return to step 6 after completion.

[0115] Based on the above method embodiment, the present invention further discloses a scene interaction system for a virtual digital human, the system comprising:

[0116] Data collection module: used to collect the interruption reason categories and interruption durations of the virtual digital human in different states; according to the interruption reason categories in the different states, the interruption durations are classified into sets of corresponding states to obtain a set of interruption durations corresponding to each interruption reason category;

[0117] Model construction module: used to construct a normal distribution density function of the corresponding category according to the interruption duration set corresponding to each interruption reason category; and establish a judgment model of the interruption reason category according to the normal distribution density function of the corresponding category;

[0118] Interaction optimization module: used to obtain the interruption duration of the new interaction process, predict the interruption reason category through the judgment model of the interruption reason category; provide corresponding state change prompt words and actions for the new interaction process according to the interruption reason category.

[0119] The above system embodiments and method embodiments correspond one to one, and for a brief description of the system embodiments, please refer to the method embodiments.

[0120] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus; wherein the processor, memory, and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the aforementioned method of the present invention.

[0121] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to implement all or part of the steps of the method described in the embodiment of the present invention. The storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk, and other media that can store program codes.

[0122] The system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be distributed to multiple network units. A person skilled in the art may select some or all of the modules according to actual needs to achieve the purpose of the solution of this embodiment without creative effort.

[0123] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing scene interaction of virtual digital humans, characterized in that: The method comprises: Collect the reasons and duration of interruptions of virtual digital humans in different states; The interruption duration is the duration from when the virtual digital human changes to the current state to when it is interrupted in the current state; The collection of interruption reasons and interruption durations of virtual digital humans in different states specifically includes: In the thinking state or answering state, record the interruption duration and determine the type of interruption reason; In the above-mentioned thinking state, the method for determining the type of interruption reason is as follows: After the virtual digital human is interrupted and returns to the waiting state, the question inputted by the user is obtained again. If the similarity between the questions inputted by the user before and after the interruption is lower than the first preset threshold, the interruption reason category is that the user wants to change the question; if the question inputted by the user again contains a specific urging keyword, the interruption reason category is that the user thinks that the thinking time is too long; those that do not belong to the above two categories are uniformly classified as other reasons; In the answering state, the method for determining the type of interruption reason is as follows: After the virtual digital human is interrupted and returns to the waiting state, the question inputted by the user is obtained again. If the similarity between the questions inputted by the user before and after the interruption is lower than the second preset threshold, the interruption reason category is that the user wants to change the question; if the similarity between the question inputted by the user after the interruption and the answer outputted by the virtual digital human is greater than or equal to the first preset threshold, and the question inputted by the user again contains specific negative keywords, the interruption reason category is that the user believes that the understanding is wrong; if the similarity between the question inputted by the user after the interruption and the answer outputted by the virtual digital human is greater than or equal to the second preset threshold, and the question inputted by the user again contains specific detailed keywords, the interruption reason category is that the user wants to refine the answer; those that do not belong to the above three categories are uniformly classified as other reasons; According to the interruption reason categories in the different states, the interruption durations are classified into sets of corresponding states to obtain an interruption duration set corresponding to each interruption reason category; According to the interruption duration set corresponding to each interruption reason category, construct a normal distribution density function of the corresponding category; Establishing a judgment model of the interruption reason category according to the normal distribution density function of the corresponding category; Obtaining the interruption duration of the new interaction process, and predicting the interruption reason category through the judgment model of the interruption reason category; According to the interruption reason category, corresponding state transition prompt words and actions are provided for the new interaction process.

2. The scene interaction optimization method of virtual digital human according to claim 1 is characterized in that: The constructing of a normal distribution density function of a corresponding category according to the interruption duration set corresponding to each interruption reason category specifically includes: Calculate the mean and standard deviation of the interruption duration set corresponding to each interruption reason category; The normal distribution density function of the corresponding category is constructed according to the mean and standard deviation, and the formula is: Among them, i is the serial number of the interruption reason category, f i (t) is the normal distribution density function of the interruption reason category i, μ i is the mean of the interruption duration of interruption reason category i, σ i is the standard deviation of the interruption duration set of interruption reason category i, t is the interruption duration, and exp is the exponential function.

3. The scene interaction optimization method of virtual digital human according to claim 2 is characterized in that: The formula of the judgment model of the interruption reason category is: Wherein, i=1,2,…,n, n is the total number of interruption reason categories, t is the interruption duration, and P(i|t) is the probability that the interruption reason category belongs to the i-th category under the condition that the interruption duration is t.

4. The scene interaction optimization method of virtual digital human according to claim 3 is characterized in that: The predicting of the interruption reason category by the judgment model of the interruption reason category specifically includes: After obtaining the interruption duration t0 of the new interaction process, the normal distribution density function value f corresponding to each category at t0 is calculated respectively. i (t0); The normal distribution density function value f i (t0) is substituted into the judgment model of the interruption reason category, and the probability of belonging to each interruption reason category under the condition of interruption duration t0 is calculated respectively; The category with the highest probability is selected as the category of the interruption reason.

5. The scene interaction optimization method of virtual digital human according to claim 1 is characterized in that: The method further comprises: Obtain the question input again by the user after the interruption of the new interaction process, and verify whether the judgment of the judgment model of the interruption reason category is correct. If it is incorrect, add the interruption duration of the new interaction process to the interruption duration set of the corresponding category, and update the corresponding mean and standard deviation.

6. A scene interaction system for a virtual digital human, characterized in that: The system comprises: Data collection module: used to collect the interruption reason categories and interruption durations of the virtual digital human in different states; according to the interruption reason categories in the different states, the interruption durations are classified into sets of corresponding states to obtain a set of interruption durations corresponding to each interruption reason category; The interruption duration is the duration from when the virtual digital human changes to the current state to when it is interrupted in the current state; The collection of interruption reasons and interruption durations of virtual digital humans in different states specifically includes: In the thinking state or answering state, record the interruption duration and determine the type of interruption reason; In the above-mentioned thinking state, the method for determining the type of interruption reason is as follows: After the virtual digital human is interrupted and returns to the waiting state, the question inputted by the user is obtained again. If the similarity between the questions inputted by the user before and after the interruption is lower than the first preset threshold, the interruption reason category is that the user wants to change the question; if the question inputted by the user again contains a specific urging keyword, the interruption reason category is that the user thinks that the thinking time is too long; those that do not belong to the above two categories are uniformly classified as other reasons; In the answering state, the method for determining the type of interruption reason is as follows: After the virtual digital human is interrupted and returns to the waiting state, the question inputted by the user is obtained again. If the similarity between the questions inputted by the user before and after the interruption is lower than the second preset threshold, the interruption reason category is that the user wants to change the question; if the similarity between the question inputted by the user after the interruption and the answer outputted by the virtual digital human is greater than or equal to the first preset threshold, and the question inputted by the user again contains specific negative keywords, the interruption reason category is that the user believes that the understanding is wrong; if the similarity between the question inputted by the user after the interruption and the answer outputted by the virtual digital human is greater than or equal to the second preset threshold, and the question inputted by the user again contains specific detailed keywords, the interruption reason category is that the user wants to refine the answer; those that do not belong to the above three categories are uniformly classified as other reasons; Model construction module: used to construct a normal distribution density function of the corresponding category according to the interruption duration set corresponding to each interruption reason category; and establish a judgment model of the interruption reason category according to the normal distribution density function of the corresponding category; Interaction optimization module: used to obtain the interruption duration of the new interaction process, predict the interruption reason category through the judgment model of the interruption reason category; provide corresponding state change prompt words and actions for the new interaction process according to the interruption reason category.

7. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; Wherein, the processor, memory, and communication interface communicate with each other via the bus; The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are executed by a processor to enable a computer to implement the method according to any one of claims 1 to 5.

Images