Live broadcast interaction method and device, electronic equipment and computer readable storage medium

Abstract

The embodiment of the invention provides a live broadcast interaction method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of computers. Obtaining a first interaction parameter corresponding to each watching user in the current live broadcast; inputting the first interaction parameter corresponding to each watching user into a pre-trained user value model for processing to obtain a conversion value corresponding to each watching user; the conversion value represents the promotion influence of the interaction behavior between the anchor and the watching user on the conversion of the watching user; determining a first target user whose conversion value reaches a preset value from the plurality of watching users according to the conversion value corresponding to each watching user; and for each first target user, determining an interaction strategy corresponding to the first target user through a pre-trained reply model, and sending the interaction strategy to the anchor, so that the anchor interacts with the first target user according to the interaction strategy. Therefore, user conversion can be promoted, and the live broadcast effect is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more specifically, to a live interactive method, device, electronic device, and computer-readable storage medium. Background Art

[0002] With the development of the live broadcast industry, functional live broadcast rooms and live broadcast sales have become mature applications in all walks of life. This type of live broadcast can often enable the host to interact with the users watching the live broadcast, so that some users can achieve certain conversions, such as following the host, purchasing goods, etc.

[0003] In the prior art, the conversion intention of the user itself can be analyzed and the users can be ranked according to the conversion intention. However, this method can only realize the classification of users and cannot guide the live broadcast interaction process to promote user conversion. Therefore, there is a problem of poor live broadcast effect. Summary of the invention

[0004] In view of this, the purpose of this application is to provide a live broadcast interactive method, device, electronic device and computer-readable storage medium to promote user conversion and improve the live broadcast effect.

[0005] In order to achieve the above purpose, the technical solution adopted in the embodiment of the present application is as follows:

[0006] In a first aspect, the present application provides a live interactive method, the method comprising:

[0007] Obtain the first interaction parameter corresponding to each viewing user in the current live broadcast;

[0008] Inputting the first interaction parameter corresponding to each viewing user into a pre-trained user value model for processing, and obtaining a conversion value corresponding to each viewing user; the conversion value represents the promotion effect of the interaction behavior between the anchor and the viewing user on the conversion of the viewing user;

[0009] Determine, from among the plurality of viewing users, a first target user whose conversion value reaches a preset value according to the conversion value corresponding to each viewing user;

[0010] For each of the first target users, an interaction strategy corresponding to the first target user is determined through a pre-trained response model, and the interaction strategy is sent to the anchor, so that the anchor interacts with the first target user according to the interaction strategy.

[0011] In an optional implementation manner, the obtaining of the first interaction parameter corresponding to each viewing user in the current live broadcast includes:

[0012] Acquire live broadcast parameters of the current live broadcast; the live broadcast parameters include user messages of multiple viewing users and multiple segments of live broadcast content of the anchor;

[0013] The first interaction parameter corresponding to each of the viewing users is determined according to each of the user messages and the timestamps corresponding to the user messages, each of the host live broadcast content and the timestamps corresponding to the host live broadcast content.

[0014] In an optional implementation manner, inputting the first interaction parameter corresponding to each viewing user into a pre-trained user value model for processing to obtain a conversion value corresponding to each viewing user includes:

[0015] For each of the viewing users, inputting the first interaction parameter corresponding to the viewing user into the user value model for processing to obtain a first conversion probability and a second conversion probability corresponding to the viewing user;

[0016] The first conversion probability represents the conversion probability of the viewing user when there is an interactive behavior between the anchor and the viewing user, and the second conversion probability represents the conversion probability of the viewing user when there is no interactive behavior between the user and the viewing user;

[0017] A difference between the first conversion probability and the second conversion probability is determined as a conversion value of the viewing user.

[0018] In an optional implementation manner, determining the interaction strategy corresponding to each first target user through a pre-trained response model for each first target user respectively includes:

[0019] Determining a reply order corresponding to each of the first target users according to the conversion value corresponding to each of the first target users;

[0020] According to the reply order corresponding to each of the first target users, the users to be replied are determined in turn, the first interaction parameters corresponding to the users to be replied are input into the reply model for processing, and the interaction strategies corresponding to the users to be replied are obtained.

[0021] In an optional embodiment, the method further comprises:

[0022] If the first target user does not exist, at least one second target user is determined from the multiple viewing users according to the conversion value of each viewing user, and the first interaction parameter corresponding to the second target user is input into the response model for processing to obtain the interaction strategy corresponding to the second target user.

[0023] In an optional embodiment, the response model is trained by the following steps:

[0024] Obtaining second interaction parameters and industry knowledge data between multiple converted users and the anchor;

[0025] Determine the interaction parameter to be processed randomly from the plurality of the second interaction parameters according to a preset ratio, truncate the interaction parameter to be processed, and obtain the truncate interaction parameter to be processed;

[0026] The second interaction parameter, the truncated interaction parameter to be processed and the industry knowledge data are input into a preset initial response model for training to obtain a trained response model.

[0027] In an optional implementation manner, the user value model is trained by the following steps:

[0028] Obtaining third interaction parameters corresponding to multiple viewing users;

[0029] According to the conversion status and interaction status of each viewing user, adding a training label to the third interaction parameter corresponding to each viewing user;

[0030] The third interaction parameter with the training label added thereto is input into a preset initial user value model for training to obtain a trained user value model.

[0031] In a second aspect, the present application provides a live interactive device, the device comprising:

[0032] An acquisition module, used to acquire the first interaction parameter corresponding to each viewing user in the current live broadcast;

[0033] A processing module, used to input the first interaction parameter corresponding to each viewing user into a pre-trained user value model for processing, and obtain a conversion value corresponding to each viewing user; the conversion value represents the promotion effect of the interaction behavior between the anchor and the viewing user on the conversion of the viewing user;

[0034] A determination module, configured to determine, from among the plurality of viewing users, a first target user whose conversion value reaches a preset value according to the conversion value corresponding to each viewing user;

[0035] The determination module is further used to determine the interaction strategy corresponding to each of the first target users through a pre-trained response model, and send the interaction strategy to the anchor so that the anchor interacts with the first target user according to the interaction strategy.

[0036] In a third aspect, the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program that can be executed by the processor, and the processor can execute the computer program to implement the method described in any one of the aforementioned embodiments.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in any one of the aforementioned implementations.

[0038] The live broadcast interaction method, device, electronic device and computer-readable storage medium provided in the embodiments of the present application can determine the conversion value corresponding to each viewing user through the first interaction parameters corresponding to each viewing user and the pre-trained user value model. Since the conversion value can characterize the promotion effect of the interactive behavior between the host and the viewing user on the conversion of the viewing user, the first target user who should be interacted with first can be determined according to the conversion value, and the interaction strategy corresponding to each first target user can be determined through the pre-trained reply model and sent to the host. In this way, the host can interact with each first target user first according to the interaction strategy corresponding to each first target user, thereby promoting user conversion and improving the live broadcast effect.

[0039] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A block diagram of an electronic device provided by an embodiment of the present application is shown;

[0042] Figure 2 A schematic diagram of a flow chart of a live interactive method provided in an embodiment of the present application is shown;

[0043] Figure 3 A schematic diagram of a user value model is shown;

[0044] Figure 4 A schematic diagram of conversion value calculation is shown;

[0045] Figure 5 A functional module diagram of a live interactive device provided in an embodiment of the present application is shown.

[0046] Icon: 100 - memory; 110 - processor; 120 - communication module; 200 - acquisition module; 210 - processing module; 220 - determination module. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0048] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0049] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0050] At present, with the development of the live broadcast industry, functional live broadcast rooms and live broadcast sales have become mature applications in all walks of life. This type of live broadcast can often enable the host to interact with the users watching the live broadcast, so that some users can achieve certain conversions, such as following the host, purchasing goods, etc.

[0051] Current live broadcast assistants and other software are often only used to optimize the live broadcast process to improve the quality of live broadcast content. However, when faced with a large number of users, the anchor or the live broadcast team still needs to select some users to reply.

[0052] In the prior art, although some algorithms can analyze the user's own conversion intention and sort users according to the conversion intention, this method can only realize the classification of users and cannot determine the causal impact of the host's interactive behavior on user conversion. Therefore, it is often impossible to guide the live broadcast interaction process, and therefore cannot promote user conversion, resulting in poor live broadcast effects.

[0053] Based on this, the embodiments of the present application provide a live interactive method, device, electronic device and computer-readable storage medium to solve the above problems. Next, the live interactive method, device, electronic device and computer-readable storage medium provided by the embodiments of the present application are introduced in combination with the diagram.

[0054] Figure 1 A block diagram of an electronic device provided in an embodiment of the present application is shown in FIG. Figure 1 , the electronic device includes a memory 100, a processor 110 and a communication module 120. The memory 100, the processor 110 and the communication module 120 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0055] The memory 100 is used to store computer programs or data that can be executed by the processor. The memory 100 can be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc.

[0056] The processor 110 is used to read / write data or computer programs stored in the memory and execute corresponding functions.

[0057] The communication module 120 is used to establish a communication connection between the electronic device and other communication terminals through a network, and to send and receive data through the network.

[0058] It should be understood that Figure 1 The structure shown is only a schematic diagram of the structure of the electronic device. The electronic device may also include Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown. Figure 1 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0059] Next, Figure 1 The electronic device in the embodiment is the execution subject, and the live interactive method provided by the embodiment of the present application is introduced exemplarily in combination with the flowchart. Specifically, Figure 2 A flowchart of a live interactive method provided in an embodiment of the present application is shown in FIG. Figure 2 , the method comprising:

[0060] Step S20, obtaining the first interaction parameter corresponding to each viewing user in the current live broadcast.

[0061] Optionally, the first interactive parameter may be a conversation combination corresponding to the viewing user, including the question content asked by the viewing user since entering the live broadcast room and the corresponding host's reply content.

[0062] Optionally, the electronic device may start to obtain in real time the first interaction parameters corresponding to each viewing user in the current live broadcast for processing after a preset duration of time from the start of the live broadcast.

[0063] Step S21: input the first interaction parameter corresponding to each viewing user into a pre-trained user value model for processing to obtain a conversion value corresponding to each viewing user.

[0064] Among them, the conversion value represents the impact of the interactive behavior between the anchor and the viewing users on the conversion of the viewing users.

[0065] In this embodiment, the conversion value can reflect the causal impact of the host's interactive behavior on user conversion. That is, the conversion value can be used to determine whether the host's interactive measures for the viewing user can promote the user to convert, such as purchasing goods, following the host, etc.

[0066] Optionally, different live broadcast rooms may have different conversion goals. For example, some live broadcast rooms may hope to achieve user attention, some live broadcast rooms may hope to achieve user purchases, some live broadcast rooms may hope that users leave some sales leads, etc. Therefore, the user value model can be set according to the specific conversion goals of the live broadcast room in order to obtain the conversion value corresponding to the relevant conversion goal.

[0067] Step S22: determining a first target user whose conversion value reaches a preset value from among a plurality of viewing users according to the conversion values ​​corresponding to the respective viewing users.

[0068] Optionally, viewing users can generally be divided into four categories, including a user group that is disgusted by the language, a non-target user group, a high-value user group, and an interactive conversion group.

[0069] Among them, the user group that is averse to the host's active interactive guidance behavior refers to those who resist the host's active interactive guidance behavior, which will lead to a decrease in the professionalism and trust in the live broadcast, while users who can complete natural conversion if the host does not actively interact with the audience; the non-target user group refers to users who only watch the live broadcast room and cannot complete the conversion regardless of whether the live broadcast interacts with them; the high-value user group refers to users who will complete the conversion regardless of whether the host gives priority to replying to their questions or actively initiating interaction; the interactive conversion group refers to users who need to give priority to replying or interacting to complete the conversion, otherwise they will jump out of the live broadcast room.

[0070] Optionally, the first target user refers to the viewing user whose conversion is greatly promoted by the host's interactive behavior, that is, the above-mentioned interactive conversion group.

[0071] It is understandable that for this type of users, if the host does not give priority to replying to their questions or actively guide them, the conversion probability of these users is low. On the contrary, if the host gives priority to replying to their questions or actively guides them, these users are likely to complete the conversion. Therefore, it is necessary to identify these first target users from multiple viewing users for priority interaction.

[0072] Optionally, the preset value can be set according to the actual application situation, and this application does not make too many restrictions on this. For example, if the number of people in the live broadcast room is small, only 500 people, a smaller preset value can be set to ensure that 50% of the viewing users are the first target users; if the number of people in the live broadcast room is large, such as 100,000 people, a higher preset value can be set to ensure that 0.5% of the viewing users are the first target users.

[0073] Step S23, for each first target user, determine the interaction strategy corresponding to the first target user through the pre-trained response model, and send the interaction strategy to the anchor, so that the anchor interacts with the first target user according to the interaction strategy.

[0074] Optionally, the response model may be a large language model.

[0075] Optionally, the interactive strategy may be responses to unanswered questions raised by the first target user, or may be guidance on topics of interest to the first target user, such as raising guiding questions of interest to the first target user, introducing content of interest to the first target user, etc.

[0076] For example, if a first target user asks an active question, the electronic device can generate at least one reply corresponding to the question through the reply model and send it to the anchor, who then selects one reply to interact with the first target user in the live broadcast room; if a first target user has not asked an active question, guiding questions about the content that the first target user is interested in can be generated.

[0077] The live broadcast interaction method provided in the embodiment of the present application can determine the conversion value corresponding to each viewing user through the first interaction parameter corresponding to each viewing user and the pre-trained user value model. Since the conversion value can characterize the promotion effect of the interaction behavior between the host and the viewing user on the conversion of the viewing user, the first target user who should be interacted with first can be determined according to the conversion value, and the interaction strategy corresponding to each first target user can be determined through the pre-trained reply model and sent to the host. In this way, the host can interact with each first target user first according to the interaction strategy corresponding to each first target user, thereby promoting user conversion and improving the live broadcast effect.

[0078] Optionally, for ease of application, the reply model and the user value model may be trained first.

[0079] In this embodiment, a corresponding reply model and user value model can be trained for each live broadcast room respectively, or a common reply model and user value model can be trained for multiple live broadcast rooms. The specific settings can be made according to the actual application situation, and this application does not impose too many restrictions on this.

[0080] Next, a possible implementation method is provided for how to train the response model.

[0081] Specifically, the electronic device can obtain the second interaction parameters and industry knowledge data between multiple converted users and the anchor, randomly determine the interaction parameters to be processed from the multiple second interaction parameters according to a preset ratio, truncate the interaction parameters to be processed, and obtain the truncated interaction parameters to be processed, and then input the second interaction parameters, the truncated interaction parameters to be processed and the industry knowledge data into a preset initial response model for training to obtain a trained response model.

[0082] Optionally, converted users refer to users who have achieved relevant conversion goals, such as having purchased relevant products, following a host, etc.

[0083] Optionally, the second interaction parameter may include all interaction parameters from the time the converted user enters the live broadcast room to the time of conversion, i.e., the user comments or questions of the converted user and the host's replies. The industry knowledge data may include conversation records between industry experts and users. In addition, it may also include industry-related documents, etc.

[0084] For example, for a car live broadcast room, the industry data can be the conversation between car industry experts and users. This part of the data can improve the performance of the response model in industry logical reasoning ability. If the user asks "I am in the Northeast, recommend a car", the reply can be "The winter in the Northeast is relatively cold. If it is only used in the city and there are charging stations, consider an electric car. Otherwise, it is recommended to buy a large-displacement gasoline car."

[0085] Optionally, in order to ensure that the reply model can predict questions that the user may be interested in so as to actively ask questions to the corresponding user, such as predicting the user's next question and predicting the host's effective reply content, the electronic device can also perform a truncated process on the second interaction parameter. In this embodiment, the electronic device can randomly select some of the multiple second interaction parameters as interaction parameters to be processed, and divide the interaction parameters to be processed into first interaction parameters to be processed and second interaction parameters to be processed, wherein the first interaction parameters to be processed are truncated according to the host's reply, and the second interaction parameters to be processed are randomly truncated, so as to obtain the truncated interaction parameters to be processed.

[0086] In one example, if the host's reply is characterized as A and the user's question or comment is characterized as Q, the second interaction parameter can be characterized as Q->A->Q->…->Q->(A) (that is, the user completes the conversion after the host replies (A)), the first interaction parameter to be processed can be characterized as Q->A->Q->…->A->(Q) (used to predict the user's next question (Q)), and the second interaction parameter to be processed can be characterized as Q->A->Q->…->Q->(A) (the second interaction parameter is randomly truncated and used to predict the host's reply content (A)).

[0087] Optionally, the preset ratio can be set according to actual application conditions. In one possible implementation, the second interaction parameter, the first interaction parameter to be processed, the second interaction parameter to be processed, and the industry knowledge data can be mixed in a ratio of 2:1:2:1.

[0088] In one possible implementation, the initial response model may be 10000 questions 2.514B. In addition, the electronic device may also perform Lora fine-tuning on the response model by means of supervised fine-tuning. The fine-tuning hardware may use Nvidia A800*12, 500GB RAM training cluster, and the training rounds may be a preset round, such as 3 rounds.

[0089] Optionally, after the response model is trained, it can be deployed according to actual application requirements.

[0090] It should be noted that the reply information generated by the large model in the existing technology is often not natural enough, and it is often unable to combine the user's interaction context and live broadcast context information to generate personalized replies. Therefore, it is impossible to solve user problems while engaging in targeted interactions with users and guide users to complete conversions.

[0091] The live interactive method provided in the embodiment of the present application can input the second interactive parameters and industry knowledge data of the converted users into the initial reply model for training. Therefore, the trained reply model can combine product expertise and conversation history in the corresponding style and tone to generate a suitable interactive strategy. At the same time, by intercepting part of the second interactive parameters and inputting them into the initial reply model for training, the trained reply model can also have the ability to actively ask questions, guide, and provide high-quality replies. In this way, personalized replies can be generated in the current live broadcast, and user problems can be solved while interacting with users in a targeted manner, guiding users to complete the conversion.

[0092] Next, a possible implementation method is provided for how to train the user value model.

[0093] Specifically, the electronic device can obtain the third interaction parameters corresponding to multiple viewing users, add training labels to the third interaction parameters corresponding to each viewing user according to the conversion and interaction status of each viewing user, and then input the third interaction parameters with the added training labels into a pre-set initial user value model for training to obtain a trained user value model.

[0094] Optionally, due to the characteristics of live broadcasting, the electronic device cannot collect information about whether the same viewing user has completed conversion when interacting or not interacting at the same time. Therefore, the interaction parameters between the anchor and the user during the live broadcasting process can be collected. For converted users, all interaction parameters from the time they enter the live broadcasting room to the time of conversion can be retained, and for non-converted users, all interaction parameters during the entire live broadcasting can be retained.

[0095] It can be understood that the third interaction parameter may include all interaction parameters of the converted user from the time of entering the live broadcast room to the conversion time point, as well as all interaction parameters of the unconverted user during the entire live broadcast.

[0096] Optionally, the conversion situation refers to whether the viewing user has a conversion, and the interaction situation refers to whether there is an interaction between the viewing user and the anchor.

[0097] In one possible implementation, the interaction situation can be characterized as T and the conversion situation can be characterized as Y. Then, according to the conversion situation and interaction situation of each viewing user, a training label can be added to the third interaction parameter corresponding to each viewing user, including (T=1, Y=1), (T=1, Y=0), (T=0, Y=1), and (T=0, Y=0).

[0098] Among them, (T=1, Y=1) indicates that there is interaction between the viewing user and the anchor and the user is converted, (T=1, Y=0) indicates that there is interaction between the viewing user and the anchor and the user is not converted, (T=0, Y=1) indicates that there is no interaction between the viewing user and the anchor and the user is converted, and (T=0, Y=0) indicates that there is no interaction between the viewing user and the anchor and the user is not converted.

[0099] It can be understood that the situation where there is no interactive behavior refers to the situation where the viewing user actively asks questions or comments, leaving interactive parameters, but the anchor does not respond to them.

[0100] In one possible implementation, considering that in an actual live broadcast scenario, most of the viewing users are non-interactive and non-converted users, the data volume of the third interaction parameter of (T=0, Y=0) may be large. In order to ensure the model training effect, the electronic device can randomly sample the third interaction parameter of (T=0, Y=0), and select part of the third interaction parameters of (T=0, Y=0) to input into the initial user value model for training.

[0101] Optionally, the initial user value model may be an Uplift+Transformer structure.

[0102] Optionally, considering that there may be multiple conversion goals, such as attention, purchase, transaction leads, etc., multi-task training can be adopted, that is, multiple training labels are set for the third interaction parameter, such as Y1 represents purchase, Y2 represents attention, etc. On this basis, the user value model can output values ​​corresponding to multiple conversion goals.

[0103] Optionally, after the user value model is trained, it needs to be deployed. At this time, one or more different conversion goals can be selected for different live broadcast rooms.

[0104] In one example, Figure 3 For a user value model diagram, see Figure 3 The electronic device can input multiple third interaction parameters with added training labels into the embedding layer to generate a sentence embedding vector, input the sentence embedding vector into the Transformer layer for processing, perform context encoding on the sentence embedding vector, output a word vector containing context information, and then splice vectors from different sources through the splicing layer to output a high-dimensional vector of comprehensive information, and finally perform nonlinear transformation on the spliced ​​vector through the fully connected layer to output the final prediction result.

[0105] Optionally, considering that there are many live broadcast parameters in the actual live broadcast process, the electronic device first obtains the live broadcast parameters in the current live broadcast, and obtains the first interactive parameters after performing data cleaning on them.

[0106] In this embodiment, the live broadcast parameters may include user messages from multiple viewing users and multiple segments of the host's live broadcast content.

[0107] Optionally, the live broadcast content of the host may include the host's oral broadcast content and the host's comment information on the messages left by viewing users.

[0108] Optionally, the electronic device may determine the first interaction parameter corresponding to each viewing user based on each user's message and the timestamp corresponding to the user's message, each host's live broadcast content and the timestamp corresponding to the host's live broadcast content.

[0109] In this embodiment, the electronic device can clean and integrate the live broadcast parameters to obtain a conversation combination corresponding to each viewing user, that is, a first interaction parameter.

[0110] Next, a possible implementation method is provided for how to input the first interaction parameter corresponding to each viewing user into a pre-trained user value model for processing to obtain the conversion value corresponding to each viewing user.

[0111] Specifically, the electronic device can input the first interaction parameter corresponding to each viewing user into the user value model for processing, obtain the first conversion probability and the second conversion probability corresponding to the viewing user, and determine the difference between the first conversion probability and the second conversion probability as the conversion value of the viewing user.

[0112] Among them, the first conversion probability represents the conversion probability of the viewing user when there is an interactive behavior between the anchor and the viewing user, and the second conversion probability represents the conversion probability of the viewing user when there is no interactive behavior between the anchor and the viewing user.

[0113] Optionally, the user value model may process the first interaction parameter to output a conversion probability corresponding to the viewing user.

[0114] In this embodiment, in order to determine the promoting effect of the host's interactive behavior on the conversion of viewing users, two prediction processes can be performed through the user value model, and different interaction variables are set for the two prediction processes, thereby outputting the first conversion probability and the second conversion probability respectively.

[0115] It can be understood that the interaction variable is a binary variable, which refers to whether the anchor interacts with the viewing users.

[0116] In an example, see Figure 4, which is a schematic diagram of conversion value calculation, the electronic device can input the first interaction parameter of the viewing user into the user value model, and set the interaction variable T to 1, so as to obtain the first conversion probability of the conversion variable Y=1 under the condition of T=1 for the viewing user. In addition, the electronic device also needs to set the interaction variable T to 0, so as to obtain the second conversion probability of the conversion variable Y=1 under the condition of T=0 for the viewing user, and then determine the difference between the two as the conversion value corresponding to the viewing user.

[0117] Optionally, in order to further improve the conversion rate of viewing users, the anchor needs to first interact with the first target user with a high conversion value. Therefore, when the electronic device determines the interaction strategy corresponding to the first target user, it can first sort the first target user.

[0118] Next, a possible implementation method is provided for how to determine the interaction strategy corresponding to the first target user through a pre-trained response model for each first target user.

[0119] Specifically, the electronic device can determine the reply order corresponding to each first target user according to the conversion value corresponding to each first target user, determine the users to be replied to in turn according to the reply order corresponding to each first target user, input the first interaction parameters corresponding to the users to be replied to the reply model for processing, and obtain the interaction strategies corresponding to the users to be replied.

[0120] Optionally, the electronic device may sort the first target users according to the conversion value from high to low, so as to determine the reply order corresponding to the first target users.

[0121] In this embodiment, if the user value model is set with only one conversion target, it can be directly sorted according to the conversion value corresponding to the conversion target. However, if the user value model is set with multiple conversion targets, for example, both purchase and attention, the electronic device can perform a mixed sorting of the first target user according to the preset recommendation algorithm and the conversion value corresponding to each conversion target.

[0122] In a possible implementation, the preset recommendation algorithm may be a click-through rate (CTR) algorithm.

[0123] Optionally, in order to ensure the generation effect of the interactive strategy, the electronic device can also input the output result of the user value model, the interactive goal, the length of time from the last record of the first interactive parameter of the first target user to the current moment, the length of time the first target user has entered the live broadcast room, etc. into the reply model.

[0124] The output result of the user value model refers to the conversion probability of the first target user, and the interaction goal may be to guide the user to pay attention, guide the user to purchase, guide the user to leave sales leads, etc.

[0125] It is understandable that the interaction strategy needs to be related to the interaction goal, for example, prompting users with shopping information, guiding users' attention, analyzing the products that users are interested in, etc.

[0126] Optionally, in order to ensure the processing accuracy of the response model, the response model needs to be iteratively trained according to the interaction parameters generated during the actual application process.

[0127] In this embodiment, in order to improve the richness and diversity of interactive strategies, the electronic device can not only generate interactive strategies through a reply model and send them to the host, but also encourage the host to reply to the user on his own to explore more and better reply methods.

[0128] In one possible implementation, the model interaction ratio and the host interaction ratio can be set according to actual application conditions. In one example, the model interaction ratio can be 80%, and the host interaction ratio can be 20%, that is, the electronic device can input 80% of the first interaction parameters of the first target user into the reply model for processing, and send 20% of the first interaction parameters of the first target user to the host, who can reply to the questions of the first target user or initiate topic guidance for the first target user.

[0129] Optionally, considering the possibility that there may be no first target user, the electronic device can determine at least one second target user from multiple viewing users based on the conversion value of each viewing user, input the first interaction parameter corresponding to the second target user into the reply model for processing, and obtain the interaction strategy corresponding to the second target user.

[0130] In this embodiment, if there is no conversion value greater than the preset value, the electronic device can find at least one second target user with a larger conversion value among multiple viewing users, and generate an interaction strategy corresponding to the second target user.

[0131] In a possible implementation, the electronic device may lower a preset value, and screen the second target user from a plurality of viewing users according to the lowered preset value.

[0132] In order to execute the corresponding steps in the above embodiments and various possible methods, a method for implementing a live interactive device is provided below. Figure 5 , Figure 5A functional module diagram of a live interactive device provided in an embodiment of the present application. It should be noted that the basic principle and technical effects of the live interactive device provided in this embodiment are the same as those of the above embodiments. For the sake of brief description, for parts not mentioned in this embodiment, reference can be made to the corresponding contents in the above embodiments. The live interactive device includes: an acquisition module 200, a processing module 210, and a determination module 220.

[0133] The acquisition module 200 is used to acquire the first interaction parameter corresponding to each viewing user in the current live broadcast.

[0134] It can be understood that the acquisition module can be used to execute the above step S20.

[0135] The processing module 210 is used to input the first interaction parameter corresponding to each viewing user into a pre-trained user value model for processing to obtain a conversion value corresponding to each viewing user; the conversion value represents the promotion effect of the interaction behavior between the anchor and the viewing user on the conversion of the viewing user.

[0136] It can be understood that the processing module 210 can be used to execute the above step S21.

[0137] The determination module 220 is used to determine a first target user whose conversion value reaches a preset value from a plurality of viewing users according to the conversion values ​​corresponding to the respective viewing users.

[0138] It can be understood that the determination module 220 can be used to execute the above step S22.

[0139] The determination module 220 is also used to determine the interaction strategy corresponding to each first target user through a pre-trained response model, and send the interaction strategy to the anchor so that the anchor can interact with the first target user according to the interaction strategy.

[0140] It can be understood that the determination module 220 can be used to execute the above step S23.

[0141] Optionally, the acquisition module 200 is also used to obtain live broadcast parameters in the current live broadcast; the live broadcast parameters include user messages from multiple viewing users and multiple segments of live broadcast content by the host; based on each user message and the timestamp corresponding to the user message, each live broadcast content of the host and the timestamp corresponding to the live broadcast content of the host, the first interaction parameter corresponding to each viewing user is determined.

[0142] Optionally, the processing module 210 is further used to input the first interaction parameter corresponding to each viewing user into the user value model for processing, so as to obtain the first conversion probability and the second conversion probability corresponding to the viewing user; wherein the first conversion probability represents the conversion probability of the viewing user when there is an interactive behavior between the anchor and the viewing user, and the second conversion probability represents the conversion probability of the viewing user when there is no interactive behavior between the anchor and the viewing user; and the difference between the first conversion probability and the second conversion probability is determined as the conversion value of the viewing user.

[0143] Optionally, the determination module 220 is also used to determine the reply order corresponding to each first target user according to the conversion value corresponding to each first target user; according to the reply order corresponding to each first target user, the users to be replied to are determined in turn, and the first interaction parameters corresponding to the users to be replied to are input into the reply model for processing to obtain the interaction strategies corresponding to the users to be replied.

[0144] Optionally, the determination module 220 is also used to determine at least one second target user from multiple viewing users based on the conversion value of each viewing user if the first target user does not exist, input the first interaction parameter corresponding to the second target user into the response model for processing, and obtain the interaction strategy corresponding to the second target user.

[0145] Optionally, the live broadcast interaction model also includes a model training module, which is used to obtain second interaction parameters and industry knowledge data between multiple converted users and the anchor; randomly determine the interaction parameters to be processed from the multiple second interaction parameters according to a preset ratio, truncate the interaction parameters to be processed, and obtain truncated interaction parameters to be processed; input the second interaction parameters, the truncated interaction parameters to be processed and the industry knowledge data into a preset initial response model for training to obtain a trained response model.

[0146] Optionally, the model training module is also used to obtain third interaction parameters corresponding to multiple viewing users; add training labels to the third interaction parameters corresponding to each viewing user according to the conversion and interaction status of each viewing user; input the third interaction parameters with the added training labels into a pre-set initial user value model for training, so as to obtain a trained user value model.

[0147] Optionally, the above modules can be stored in the form of software or firmware. Figure 1 The memory shown in the figure or solidified in the operating system (OS) of the electronic device, and can be Figure 1 Meanwhile, the data and program codes required for executing the above modules can be stored in the memory.

[0148] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the live broadcast interactive method provided in the embodiment of the present application can be implemented.

[0149] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0150] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0151] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as 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.

[0152] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A live interactive method, characterized in that: The method comprises: Obtain the first interaction parameter corresponding to each viewing user in the current live broadcast; Inputting the first interaction parameter corresponding to each viewing user into a pre-trained user value model for processing, and obtaining a conversion value corresponding to each viewing user; the conversion value represents the promotion effect of the interaction behavior between the anchor and the viewing user on the conversion of the viewing user; Determine, from among the plurality of viewing users, a first target user whose conversion value reaches a preset value according to the conversion value corresponding to each viewing user; For each of the first target users, an interaction strategy corresponding to the first target user is determined through a pre-trained response model, and the interaction strategy is sent to the anchor, so that the anchor interacts with the first target user according to the interaction strategy.

2. The method according to claim 1, characterized in that The obtaining of the first interaction parameter corresponding to each viewing user in the current live broadcast includes: Acquire live broadcast parameters of the current live broadcast; the live broadcast parameters include user messages of multiple viewing users and multiple segments of live broadcast content of the anchor; The first interaction parameter corresponding to each of the viewing users is determined according to each of the user messages and the timestamps corresponding to the user messages, each of the host live broadcast content and the timestamps corresponding to the host live broadcast content.

3. The method according to claim 1, characterized in that The step of inputting the first interaction parameter corresponding to each of the viewing users into a pre-trained user value model for processing to obtain the conversion value corresponding to each of the viewing users includes: For each of the viewing users, inputting the first interaction parameter corresponding to the viewing user into the user value model for processing to obtain a first conversion probability and a second conversion probability corresponding to the viewing user; The first conversion probability represents the conversion probability of the viewing user when there is an interactive behavior between the anchor and the viewing user, and the second conversion probability represents the conversion probability of the viewing user when there is no interactive behavior between the user and the viewing user; A difference between the first conversion probability and the second conversion probability is determined as a conversion value of the viewing user.

4. The method according to claim 1, characterized in that: The step of determining, for each of the first target users, a corresponding interaction strategy for each of the first target users through a pre-trained response model includes: Determining a reply order corresponding to each of the first target users according to the conversion value corresponding to each of the first target users; According to the reply order corresponding to each of the first target users, the users to be replied are determined in turn, the first interaction parameters corresponding to the users to be replied are input into the reply model for processing, and the interaction strategies corresponding to the users to be replied are obtained.

5. The method according to claim 1, characterized in that: The method further comprises: If the first target user does not exist, at least one second target user is determined from the multiple viewing users according to the conversion value of each viewing user, and the first interaction parameter corresponding to the second target user is input into the response model for processing to obtain the interaction strategy corresponding to the second target user.

6. The method according to claim 1, characterized in that The response model is trained by the following steps: Obtaining second interaction parameters and industry knowledge data between multiple converted users and the anchor; Determine the interaction parameter to be processed randomly from the plurality of the second interaction parameters according to a preset ratio, truncate the interaction parameter to be processed, and obtain the truncate interaction parameter to be processed; The second interaction parameter, the truncated interaction parameter to be processed and the industry knowledge data are input into a preset initial response model for training to obtain a trained response model.

7. The method according to claim 1, characterized in that The user value model is trained through the following steps: Obtaining third interaction parameters corresponding to multiple viewing users; According to the conversion status and interaction status of each viewing user, adding a training label to the third interaction parameter corresponding to each viewing user; The third interaction parameter with the training label added thereto is input into a preset initial user value model for training to obtain a trained user value model.

8. A live interactive device, characterized in that: The device comprises: An acquisition module, used to acquire the first interaction parameter corresponding to each viewing user in the current live broadcast; A processing module, used to input the first interaction parameter corresponding to each viewing user into a pre-trained user value model for processing, and obtain a conversion value corresponding to each viewing user; the conversion value represents the promotion effect of the interaction behavior between the anchor and the viewing user on the conversion of the viewing user; A determination module, configured to determine, from among the plurality of viewing users, a first target user whose conversion value reaches a preset value according to the conversion value corresponding to each viewing user; The determination module is further used to determine the interaction strategy corresponding to each of the first target users through a pre-trained response model, and send the interaction strategy to the anchor so that the anchor interacts with the first target user according to the interaction strategy.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program executable by the processor, and the processor can execute the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images