Self-learning digital human live broadcast method

Abstract

The invention relates to the field of digital human live broadcast, in particular to a self-learning digital human live broadcast method, which comprises the following steps of: extracting keywords of each live broadcast topic by calling a plurality of live broadcast topics stored in a database and live broadcast contents corresponding to each topic; extracting a hotspot trend amplitude of a keyword corresponding to each live broadcast topic collected by a data monitoring unit and a topic popularity characterization coefficient calculated by aiming at a search quantity of the keyword in a platform, and setting a corresponding popularity label for each live broadcast topic and performing grade division; according to the invention, live broadcast is carried out through a novel digital human image so as to stimulate the watching interest of audiences, topic live broadcast is carried out according to the topic popularity, and on the basis of ensuring the popularity of a live broadcast room, the live broadcast content is adjusted in real time. The live broadcast duration of the current hot content is increased, the interaction with audiences is enhanced, and the popularity and exposure of the live broadcast room are improved.

Description

Technical Field

[0001] The present invention relates to the field of digital human live broadcasting, and in particular to a self-learning digital human live broadcasting method. Background Art

[0002] With the younger generation's interest in technology and virtual content, digital human live broadcasts have become very popular in the field of culture and entertainment. As people's attention to digital humans increases, their preference for the digital human live broadcast model is also increasing, which has also promoted the development of digital humans as a new form of live broadcast. At the same time, the introduction of artificial intelligence and machine learning technologies has enabled digital humans to have the ability of autonomous learning and intelligent interaction. They can adjust the live broadcast content in real time based on audience feedback and changes in real-time hot topics, thereby improving audience participation and satisfaction.

[0003] Chinese patent publication number: CN118250509A, discloses a digital human live broadcast system, the system includes: a multimedia information module for collecting and processing multimedia information of a digital human live broadcast interface; a task management module connected to the multimedia information module, the task management module for generating and managing tasks to be performed by the digital human, the task management module including a language model processing submodule and a context interaction submodule, the context interaction submodule for recording the interaction context of the digital human, the language model processing submodule for processing the tasks to be performed based on the interaction context, and obtaining feedback results of the tasks to be performed; a digital human generation module connected to the task management module, the digital human generation module for generating and / or rendering a digital human based on the feedback results of the tasks to be performed, and obtaining digital human rendering results; a live broadcast push stream module connected to the digital human generation module, the live broadcast push stream module for pushing the digital human rendering results to the digital human live broadcast interface.

[0004] However, there are still the following problems in the prior art: The content of the live broadcast is generally finalized in advance and will be broadcast according to the finalized content. No content changes will be made during the live broadcast. Therefore, failure to adjust the live broadcast content in a timely manner according to real-time hot topics may lead to the loss of viewers in the live broadcast room and reduce the popularity and exposure of the live broadcast room. Summary of the invention

[0005] To this end, the present invention provides a self-learning digital human live broadcast method to overcome the problem in the prior art that the content of the live broadcast is generally predetermined and the live broadcast is performed according to the predetermined content, and no content changes are made during the live broadcast. Therefore, the live broadcast content cannot be adjusted in time according to real-time hot topics, which may lead to the loss of viewers in the live broadcast room and reduce the popularity and exposure of the live broadcast room.

[0006] To achieve the above object, the present invention provides a self-learning digital human live broadcast method, which comprises: Calling a number of live broadcast topics and live broadcast contents corresponding to the live broadcast topics stored in the database, extracting keywords of the live broadcast topics, the live broadcast contents including controlling the digital human to make predetermined actions and emit predetermined voices; Extracting the hot trend amplitude of the keyword corresponding to each of the live broadcast topics collected by the data monitoring unit and the search volume for the keyword in the platform to calculate the topic heat characterization coefficient, and setting a corresponding heat label for each of the live broadcast topics according to the topic heat characterization coefficient; According to the order of the hot tags, the corresponding live broadcast topics are obtained and broadcasted through the digital human, and the live broadcast content is adjusted in real time, including: Obtain the audience retention characteristics and the number of bullet comments during the live broadcast of the digital human to analyze the live broadcast exposure representation coefficient, so as to determine whether the live broadcast has entered the exposure stage, and adjust the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet comment keywords according to the topic heat representation coefficient; Or, randomly selecting each of the live broadcast topics stored in the database, calling the live broadcast content corresponding to the randomly selected live broadcast topic for live broadcast, and adjusting the live broadcast content according to the topic heat representation coefficient; The audience retention features include the audience retention amount within a predetermined time and the average retention viewing time.

[0007] Furthermore, the process of calculating the topic heat representation coefficient includes: Obtain the hot trend amplitude of the keywords of each live broadcast topic; Calculating the ratio of the hotspot trend amplitude to a preset hotspot trend threshold and assigning a corresponding weight coefficient as a first heat feature; Calculating the ratio of the search volume for the keyword in the platform to the search volume threshold and assigning a corresponding weight coefficient as the second heat feature; The sum of the first heat feature and the second heat feature is determined as the topic heat characterization coefficient.

[0008] Furthermore, the process of setting a corresponding heat label for each of the live broadcast topics according to the topic heat representation coefficient includes: Presetting the correspondence between the heat label and the predetermined topic heat representation coefficient interval; Determine the topic heat representation coefficient interval to which the topic heat representation coefficient corresponding to the keyword belongs; Setting the heat label corresponding to the topic heat characterization coefficient interval as the heat label of the keyword; Among them, the heat labels correspond one to one to the topic heat representation coefficient intervals.

[0009] Furthermore, each of the heat tags includes a heat tag serial number, and the heat tag serial number is positively correlated with the upper limit of the topic heat representation coefficient interval.

[0010] Further, the corresponding live broadcast topic is obtained in the order of the hot tags and the live broadcast is performed through the digital human, including: Live broadcast topics corresponding to higher popularity tag numbers will be broadcasted first through digital humans.

[0011] Furthermore, the process of obtaining the audience retention characteristics and the number of comments during the live broadcast of the digital human to analyze the live broadcast exposure representation coefficient includes: Determine the sum of the ratio of the audience retention amount to the retention amount threshold and the ratio of the average retention viewing time to the retention viewing time threshold as the first exposure feature; Determining the ratio of the number of bullet comments to the number of bullet comments as a second exposure feature; The sum of the first exposure feature and the second exposure feature is determined as a live broadcast exposure characterization coefficient.

[0012] Further, determining whether the live broadcast has entered the exposure stage includes: If the live broadcast exposure characterization coefficient is greater than or equal to the live broadcast exposure characterization coefficient threshold, it is determined that the live broadcast enters the exposure stage.

[0013] Furthermore, the live broadcast time of the live broadcast content corresponding to the current live broadcast and the determination process of the response frequency of the digital human to the bullet screen keywords are adjusted according to the topic heat representation coefficient, including: If the live broadcast enters the exposure stage, the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the barrage keywords are adjusted according to the topic heat representation coefficient.

[0014] Furthermore, the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet screen keywords are adjusted according to the topic heat representation coefficient, including: Extending the live broadcast time of the live broadcast content corresponding to the current live broadcast, where the extension amount of the live broadcast time is positively correlated with the topic heat representation coefficient; The response frequency of the digital human to the barrage keywords is increased, and the increase in the response frequency is positively correlated with the topic heat representation coefficient.

[0015] Further, adjusting the live broadcast content according to the topic heat representation coefficient includes: If there is any live broadcast topic whose topic heat representation coefficient is greater than the topic heat representation coefficient of the current live broadcast topic, the live broadcast content will be adjusted.

[0016] Compared with the prior art, the present invention extracts keywords of each live broadcast topic by calling several live broadcast topics stored in a database and the live broadcast content corresponding to each topic; extracts the hot trend amplitude of the keywords corresponding to each live broadcast topic collected by the data monitoring unit and the search volume for the keywords in the platform to calculate the topic heat characterization coefficient, so as to set corresponding heat labels for each live broadcast topic and classify them into levels; obtains the corresponding live broadcast topic in the order of heat labels and broadcasts it through a digital human, and adjusts the live broadcast content in real time. The present invention broadcasts live through a novel digital human image to stimulate the audience's viewing interest, broadcasts topics based on topic heat, and on the basis of ensuring the heat of the live broadcast room, increases the live broadcast time of the current hot content and enhances the interaction with the audience, thereby improving the heat and exposure of the live broadcast room.

[0017] In particular, the present invention calculates the topic heat characterization coefficient through the hot trend amplitude of the keyword corresponding to the live broadcast topic and the search volume for the keyword in the platform. In actual situations, hot topics are updated in real time. For example, the attention of a large number of followers in different time periods and the sensitivity of the topic will cause any topic to become a new hot topic. The current hot topics within a predetermined time range are summarized and the corresponding keywords are extracted. The hot trend amplitude of the above keywords is monitored. For example, the new or hotter hot topic is determined by the magnitude of the rising trend amplitude of any keyword hot spot. At the same time, the search volume for the above keywords in the platform is obtained. The search volume of the keyword can also show the attention of the followers to the topic from another perspective, and further reflect the heat of the topic. The live broadcast of the content with high topic heat by the digital human can not only rely on the new digital human live broadcast method to stimulate the viewing interest of the followers, but also obtain more traffic for the live broadcast room through the live broadcast heat of the high topic heat. Therefore, the present application calculates the topic heat characterization coefficient to characterize the heat and attention corresponding to the current hot topic within a certain time range, and provides data support for the subsequent setting of corresponding heat tags for the live broadcast topic, and divides the level of heat tags so that the corresponding live broadcast can be carried out by the digital human.

[0018] In particular, the present invention analyzes the live broadcast exposure characterization coefficient through the audience retention characteristics and the number of bullet comments during the live broadcast of the digital human. Under normal circumstances, in addition to the layout of the live broadcast room scene, more importantly, the live broadcast method and live broadcast content allow the audience to enter the live broadcast room and stay to watch, and the length of time they stay to watch can bring greater traffic and higher popularity to the live broadcast room. At the same time, the audience in the live broadcast room raises questions related to the live broadcast content in the form of bullet comments. The digital human can extract keywords from the bullet comment content and answer questions through all the preset information about the live broadcast content. It can also input questions into the voice model and output the voice model. The text is converted into voice and output by digital humans for answers. Then, after reaching a certain level, the platform will also give the live broadcast room more exposure. Therefore, this application analyzes the live broadcast exposure characterization coefficient through the audience retention within the predetermined time and the average retention viewing time combined with the number of barrages to further characterize the popularity and exposure of the live broadcast. When the live broadcast room obtains a certain exposure and enters the exposure stage, the live broadcast time of the current live content is appropriately increased and the enthusiasm of the barrages received in reply is increased. On the basis of ensuring the popularity of the live broadcast room, the live broadcast time of the current hot content is increased and the interaction with the audience is enhanced to increase the popularity and exposure of the live broadcast room.

[0019] In particular, the present invention uses novel digital human images for live broadcasting to stimulate audience interest in the live broadcast room and increase the possibility of audience entering the live broadcast room to watch. Furthermore, the present application adjusts the live broadcast content according to the trend of real-time topics to increase the popularity and exposure of the live broadcast room while ensuring the existing popularity of the live broadcast room. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the steps of the self-learning digital human live broadcast method according to an embodiment of the invention; Figure 2 A logical determination diagram of whether the live broadcast of an embodiment of the invention enters the exposure stage; Figure 3 A logical decision diagram for whether to adjust the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet screen keywords in the embodiment of the invention; Figure 4 A logical decision diagram for whether to adjust live broadcast content according to an embodiment of the invention. DETAILED DESCRIPTION

[0021] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0023] See also Figures 1 to 4 As shown, Figure 1 This is a schematic diagram of the steps of the self-learning digital human live broadcast method according to an embodiment of the present invention. Figure 2 This is a logic determination diagram of whether the live broadcast of an embodiment of the present invention enters the exposure stage. Figure 3 The logical decision diagram of whether to adjust the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet screen keywords in the embodiment of the invention, Figure 4 The self-learning digital human live broadcast method of the embodiment of the present invention includes: S1, calling a number of live broadcast topics and live broadcast contents corresponding to the live broadcast topics stored in a database, extracting keywords of the live broadcast topics, wherein the live broadcast contents include controlling the digital human to make predetermined actions and emit predetermined voices; S2, extracting the hot trend amplitude of the keyword corresponding to each of the live broadcast topics collected by the data monitoring unit and the search volume for the keyword in the platform to calculate the topic heat characterization coefficient, and setting a corresponding heat label for each of the live broadcast topics according to the topic heat characterization coefficient; S3, obtaining the corresponding live broadcast topic in the order of the hot tags and broadcasting it through the digital human, and adjusting the live broadcast content in real time, including: Obtain the audience retention characteristics and the number of bullet comments during the live broadcast of the digital human to analyze the live broadcast exposure representation coefficient, so as to determine whether the live broadcast has entered the exposure stage, and adjust the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet comment keywords according to the topic heat representation coefficient; Or, randomly selecting each of the live broadcast topics stored in the database, calling the live broadcast content corresponding to the randomly selected live broadcast topic for live broadcast, and adjusting the live broadcast content according to the topic heat representation coefficient; The audience retention features include the audience retention amount within a predetermined time and the average retention viewing time.

[0024] It can be understood that the live broadcast content is the narration of the topic itself according to the real-time hot spots, the extended content of the topic, and the setting of corresponding body movements for the digital human according to the various contents, and the live broadcast of the digital human's movements and voices is generated through the big model drive. Among them, the type of large model drive is not limited. The LLM large voice model can be used to drive and control the digital human to perform predetermined actions and make predetermined voices. At the same time, the reply method can be to extract the barrage questions and input them into the LLM large voice model, and convert the text output by the LLM large voice model into the voice output of the digital human to reply to the content of the barrage. Of course, the keywords in the barrage content can also be extracted to answer them according to all the information about the pre-set topics. I will not go into details here.

[0025] Specifically, there is no limitation on the specific structure of the data monitoring unit. It only needs to be able to monitor the hot trend amplitude of the keywords corresponding to the live broadcast topic and the keyword search volume in the platform. It can be monitored through data mining tools. For example, HotDetect is used to monitor hot topics in real time and perform trend analysis through data mining and machine learning algorithms. Of course, other methods can also be used, which will not be repeated here.

[0026] The present invention uses novel digital human images for live broadcasting to stimulate audience interest in the live broadcast room and increase the possibility of audience entering the live broadcast room to watch. Furthermore, the present application adjusts the live broadcast content according to the trend of real-time topics, thereby increasing the popularity and exposure of the live broadcast room while ensuring the existing popularity of the live broadcast room.

[0027] Specifically, the process of calculating the topic heat representation coefficient includes: Obtain the hot trend amplitude of the keywords of each live broadcast topic; Calculating the ratio of the hotspot trend amplitude to a preset hotspot trend threshold and assigning a corresponding weight coefficient as a first heat feature; Calculating the ratio of the search volume for the keyword in the platform to the search volume threshold and assigning a corresponding weight coefficient as the second heat feature; The sum of the first heat feature and the second heat feature is determined as the topic heat characterization coefficient.

[0028] The hot trend amplitude is the amplitude of the increase in keyword search volume within the benchmark time, and the benchmark time can be set to 1h~5h.

[0029] In this embodiment, when performing weighted summation, the weight of the first heat feature is set to 0.55, and the weight of the second heat feature is set to 0.45; The keyword hot trend amplitude variance threshold K0 and the keyword search volume threshold S0 are pre-determined. The hot trend amplitude of each keyword within a predetermined time range and the relevant data on the search volume of each keyword in the platform are obtained, and the hot trend amplitude variance mean of each keyword and the mean search volume of each keyword are solved. The keyword hot trend amplitude variance threshold is set to between 1.02 and 1.11 times the hot trend amplitude variance mean of the keyword, and the keyword search volume threshold is set to between 1.12 and 1.21 times the keyword search volume mean.

[0030] It can be understood that the predetermined time range is set between the interval [12h, 36h].

[0031] The present invention calculates the topic heat characterization coefficient through the hot trend amplitude of the keyword corresponding to the live broadcast topic and the search volume for the keyword in the platform. In actual situations, hot topics are updated in real time. For example, the attention of a large number of followers in different time periods and the sensitivity of the topic will cause any topic to become a new hot topic. The current hot topics within a predetermined time range are summarized and the corresponding keywords are extracted. The hot trend amplitude of the above keywords is monitored. For example, the new or hotter hot topic is determined by the magnitude of the rising trend amplitude of any keyword hot spot. At the same time, the search volume for the above keywords in the platform is obtained. The search volume of the keyword can also show the attention of the followers to the topic from another perspective, and further reflect the heat of the topic. The live broadcast of the content with high topic heat by the digital human can not only rely on the live broadcast of the emerging digital human to stimulate the viewing interest of the followers, but also obtain more traffic for the live broadcast room through the live broadcast heat of the high topic heat. Therefore, the present application calculates the topic heat characterization coefficient to characterize the heat and attention corresponding to the current hot topic within a certain time range, and provides data support for the subsequent setting of corresponding heat tags for the live broadcast topic, and divides the level of heat tags so that the digital human can perform the corresponding live broadcast.

[0032] Specifically, the process of setting a corresponding heat label for each live broadcast topic according to the topic heat representation coefficient includes: Presetting the correspondence between the heat label and the predetermined topic heat representation coefficient interval; Determine the topic heat representation coefficient interval to which the topic heat representation coefficient corresponding to the keyword belongs; Setting the heat label corresponding to the topic heat characterization coefficient interval as the heat label of the keyword; Among them, the heat labels correspond one to one to the topic heat representation coefficient intervals.

[0033] Each of the heat tags includes a heat tag serial number, and the heat tag serial number is positively correlated with the upper limit of the topic heat representation coefficient interval.

[0034] In this embodiment, optionally, the hot tag corresponding to the live broadcast topic is determined in the following manner: The topic heat representation coefficient is divided into three preset intervals, and three levels of heat labels are set at the same time; If the topic heat representation coefficient corresponding to the live broadcast topic is within the first preset interval [1.18, 1.27), a first level heat label 1 is set; If the topic heat representation coefficient corresponding to the live broadcast topic is within the second preset interval [1.27, 1.39), a second level heat label 2 is set; If the topic heat characterization coefficient corresponding to the live broadcast topic is within the third preset interval [1.39, 1.59], a third level heat label 3 is set.

[0035] Specifically, obtaining the corresponding live broadcast topic in the order of the hot tags and broadcasting it through the digital human includes: Live broadcast topics corresponding to higher popularity tag numbers will be broadcasted first through digital humans.

[0036] In this embodiment, a larger serial number of a heat label corresponding to a live broadcast topic represents a higher hotspot.

[0037] Specifically, the process of obtaining the audience retention characteristics and the number of comments during the live broadcast of the digital human and analyzing the live broadcast exposure representation coefficient includes: Determine the sum of the ratio of the audience retention amount to the retention amount threshold and the ratio of the average retention viewing time to the retention viewing time threshold as the first exposure feature; Determining the ratio of the number of bullet comments to the number of bullet comments as a second exposure feature; The sum of the first exposure feature and the second exposure feature is determined as a live broadcast exposure characterization coefficient.

[0038] It can be understood that the audience retention is the number of viewers in the live broadcast room, the retained viewing time is the average retention time of the audience in the live broadcast room, the number of barrages is the number of barrages sent by the audience within the benchmark comparison time, and the benchmark comparison time is 5 minutes.

[0039] In this embodiment, the audience retention threshold L0, the retention viewing time threshold G0 and the barrage number threshold D0 are obtained in advance, and the relevant data of the digital human completing several live broadcasts are obtained, and the audience retention, retention viewing time and barrage number data are called to solve the audience retention mean, retention viewing time mean and barrage number mean. The audience retention threshold is set to be between 1.01 and 1.12 times the audience retention threshold, the retention viewing time threshold is set to be between 1.13 and 1.24 times the retention viewing time threshold, and the barrage number threshold is set to be between 1.14 and 1.22 times the barrage number mean.

[0040] The present invention analyzes the live broadcast exposure characterization coefficient through the audience retention characteristics and the number of bullet comments during the live broadcast of the digital human. Generally speaking, in addition to the layout of the live broadcast room scene, more importantly, the live broadcast method and live broadcast content allow the audience to enter the live broadcast room and stay to watch, and the length of time they stay to watch can bring greater traffic and higher popularity to the live broadcast room. At the same time, the audience in the live broadcast room raises questions related to the live broadcast content in the form of bullet comments. The digital human can extract keywords from the bullet comment content and answer questions based on all the preset information about the live broadcast content. It can also input questions into the voice model and output the voice model. The text is converted into voice and the answers are output by digital humans. Then, after reaching a certain level, the platform will also give the live broadcast room more exposure. Therefore, this application analyzes the live broadcast exposure characterization coefficient through the audience retention within the predetermined time and the average retention viewing time combined with the number of barrages to further characterize the popularity and exposure of the live broadcast. When the live broadcast room obtains a certain exposure and enters the exposure stage, the live broadcast time of the current live content is appropriately increased and the enthusiasm of the barrages received in reply is increased. On the basis of ensuring the popularity of the live broadcast room, the live broadcast time of the current hot content is increased and the interaction with the audience is enhanced to increase the popularity and exposure of the live broadcast room.

[0041] Specifically, determining whether a live broadcast has entered the exposure stage includes: If the live broadcast exposure characterization coefficient is greater than or equal to the live broadcast exposure characterization coefficient threshold, it is determined that the live broadcast enters the exposure stage; If the live broadcast exposure representation coefficient is less than the live broadcast exposure representation coefficient threshold, it is determined that the live broadcast has not entered the exposure stage; The live broadcast exposure characterization coefficient threshold is selected in the interval [3.29,3.71].

[0042] Specifically, the process of adjusting the live broadcast time of the live broadcast content corresponding to the current live broadcast and the determination process of the response frequency of the digital human to the bullet screen keywords according to the topic heat representation coefficient includes: If the live broadcast enters the exposure stage, the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet screen keywords are adjusted according to the topic heat representation coefficient; If the live broadcast has not entered the exposure stage, there is no need to adjust the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the barrage keywords.

[0043] Specifically, adjusting the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet screen keywords according to the topic heat representation coefficient includes: Extending the live broadcast time of the live broadcast content corresponding to the current live broadcast, where the extension amount of the live broadcast time is positively correlated with the topic heat representation coefficient; The response frequency of the digital human to the barrage keywords is increased, and the increase in the response frequency is positively correlated with the topic heat representation coefficient.

[0044] In this embodiment, optionally, The topic heat characterization coefficient H is compared with the first topic heat characterization coefficient comparison threshold H1 and the second topic heat characterization coefficient comparison threshold H2. If H>H2, the extension amount of the live broadcast time is determined to be the first extension amount of the live broadcast time a1, and a1=0.43a0 is set; If H1≤H≤H2, the live broadcast time extension amount is determined to be the second live broadcast time extension amount a2, and a2=0.36a0 is set; If H<H2, the live broadcast time extension amount is determined to be the third live broadcast time extension amount a3, and a3=0.28a0 is set; Among them, a0 represents the initial live broadcast time, H1=1.2H0, H2=1.4H0.

[0045] It is understandable that the initial live broadcast time can be determined based on the live broadcast content corresponding to the current live broadcast topic. The live broadcast content includes the digital human making predetermined actions and emitting predetermined voices. Extending the live broadcast time can be achieved by repeating the predetermined actions and predetermined voices, which will not be elaborated here.

[0046] The topic heat characterization coefficient H is compared with the first topic heat characterization coefficient comparison threshold H1 and the second topic heat characterization coefficient comparison threshold H2. If H>H2, the response frequency increase amount is determined to be the first response frequency increase amount b1, and b1=0.55b0 is set; If H1≤H≤H2, the response frequency increase amount is determined to be the second response frequency increase amount b2, and b2=0.42b0 is set; If H<H2, the response frequency increase amount is determined to be the third response frequency increase amount b3, and b3=0.37b0 is set; Among them, b0 represents the initial response frequency, H1=1.2H0, H2=1.4H0.

[0047] The initial response frequency can be selected within the range [5 times / 10min, 20 times / 10min].

[0048] It is understandable that the digital human's response to the barrage keywords may be to extract the barrage keywords for voice reply or to perform predetermined actions based on the keywords. Those skilled in the art can pre-set this and will not be elaborated here.

[0049] Specifically, adjusting the live broadcast content according to the topic heat representation coefficient includes: If there is any live broadcast topic whose topic heat representation coefficient is greater than the topic heat representation coefficient of the current live broadcast topic, the live broadcast content is adjusted; If there is no live broadcast topic whose topic heat representation coefficient is greater than the topic heat representation coefficient of the current live broadcast topic, there is no need to adjust the live broadcast content.

[0050] Specifically, the adjustment of the live broadcast content includes changing to a more popular live broadcast topic for live broadcast. It can be understood that there is no need to adjust the live broadcast content to keep the current live broadcast content unchanged until the current live broadcast content is completed, obtain the topic heat representation coefficients corresponding to the remaining live broadcast topics, extract the live broadcast topic corresponding to the highest value of the topic heat representation coefficient, and obtain the corresponding live broadcast content for live broadcast.

[0051] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A self-learning digital human live broadcast method, characterized in that: include: Calling a number of live broadcast topics and live broadcast contents corresponding to the live broadcast topics stored in the database, extracting keywords of the live broadcast topics, the live broadcast contents including controlling the digital human to make predetermined actions and emit predetermined voices; Extracting the hot trend amplitude of the keyword corresponding to each of the live broadcast topics collected by the data monitoring unit and the search volume for the keyword in the platform to calculate the topic heat characterization coefficient, and setting a corresponding heat label for each of the live broadcast topics according to the topic heat characterization coefficient; Obtain the corresponding live broadcast topic in the order of the hot tags and broadcast it live through the digital human, and adjust the live broadcast content in real time, including: Obtain the audience retention characteristics and the number of bullet comments during the live broadcast of the digital human to analyze the live broadcast exposure representation coefficient, so as to determine whether the live broadcast has entered the exposure stage, and adjust the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet comment keywords according to the topic heat representation coefficient; Or, randomly selecting each of the live broadcast topics stored in the database, calling the live broadcast content corresponding to the randomly selected live broadcast topic for live broadcast, and adjusting the live broadcast content according to the topic heat representation coefficient; The audience retention features include the audience retention amount within a predetermined time and the average retention viewing time.

2. The self-learning digital human live broadcast method according to claim 1, characterized in that: The process of calculating the topic heat representation coefficient includes: Obtain the hot trend amplitude of the keywords of each live broadcast topic; Calculating the ratio of the hotspot trend amplitude to a preset hotspot trend threshold and assigning a corresponding weight coefficient as a first heat feature; Calculating the ratio of the search volume for the keyword in the platform to the search volume threshold and assigning a corresponding weight coefficient as the second heat feature; The sum of the first heat feature and the second heat feature is determined as the topic heat characterization coefficient.

3. The self-learning digital human live broadcast method according to claim 1, characterized in that: The process of setting a corresponding heat label for each live broadcast topic according to the topic heat representation coefficient includes: Presetting the correspondence between the heat label and the predetermined topic heat representation coefficient interval; Determine the topic heat representation coefficient interval to which the topic heat representation coefficient corresponding to the keyword belongs; Setting the heat label corresponding to the topic heat characterization coefficient interval as the heat label of the keyword; Among them, the heat labels correspond one to one to the topic heat representation coefficient intervals.

4. The self-learning digital human live broadcast method according to claim 3 is characterized in that: Each of the heat tags includes a heat tag serial number, and the heat tag serial number is positively correlated with the upper limit of the topic heat representation coefficient interval.

5. The self-learning digital human live broadcast method according to claim 1, characterized in that: Obtaining the corresponding live broadcast topic in the order of the hot tags and broadcasting it live through the digital human, including: Live broadcast topics corresponding to higher popularity tag numbers will be broadcasted first through digital humans.

6. The self-learning digital human live broadcast method according to claim 1, characterized in that: The process of obtaining the audience retention characteristics and the number of comments during the live broadcast of the digital human and analyzing the live broadcast exposure representation coefficient includes: Determine the sum of the ratio of the audience retention amount to the retention amount threshold and the ratio of the average retention viewing time to the retention viewing time threshold as the first exposure feature; Determining the ratio of the number of bullet comments to the number of bullet comments as a second exposure feature; The sum of the first exposure feature and the second exposure feature is determined as a live broadcast exposure characterization coefficient.

7. The self-learning digital human live broadcast method according to claim 1, characterized in that: Determine whether the live broadcast has entered the exposure stage, including: If the live broadcast exposure characterization coefficient is greater than or equal to the live broadcast exposure characterization coefficient threshold, it is determined that the live broadcast enters the exposure stage.

8. The self-learning digital human live broadcast method according to claim 1, characterized in that: The process of adjusting the live broadcast time of the live broadcast content corresponding to the current live broadcast and the determination process of the response frequency of the digital human to the bullet screen keywords according to the topic heat representation coefficient includes: If the live broadcast enters the exposure stage, the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the barrage keywords are adjusted according to the topic heat representation coefficient.

9. The self-learning digital human live broadcast method according to claim 1, characterized in that: Adjusting the live broadcast time of the live broadcast content corresponding to the current live broadcast and the response frequency of the digital human to the bullet screen keywords according to the topic heat representation coefficient, including: Extending the live broadcast time of the live broadcast content corresponding to the current live broadcast, where the extension amount of the live broadcast time is positively correlated with the topic heat representation coefficient; The response frequency of the digital human to the barrage keywords is increased, and the increase in the response frequency is positively correlated with the topic heat representation coefficient.

10. The self-learning digital human live broadcast method according to claim 1, characterized in that: Adjusting the live broadcast content according to the topic heat representation coefficient includes: If there is any live broadcast topic whose topic heat representation coefficient is greater than the topic heat representation coefficient of the current live broadcast topic, the live broadcast content will be adjusted.

Images