JP7355309B1 - Systems, methods, and computer-readable media for contribution prediction - Google Patents

Abstract

The present invention provides a contribution prediction system, a viewer contribution prediction method, and a non-transitory computer-readable medium that analyzes viewers and identifies contributor candidates for a distributor.
A system for predicting contribution degree receives a contribution prediction request, determines a viewer/distributor pair, and determines whether the contribution amount of the viewer has already reached a threshold value. Then, if it is determined that the threshold has been reached, the user is displayed as a top spender, and if it is determined that the threshold has not been reached, the user's attributes/contribution data/parameters are obtained and the machine learning model is displayed as a top spender. Determine the prediction (spender candidate or non-spender candidate) and display the determined contribution prediction.
[Selection diagram] Figure 10

Description

The present invention relates to viewer contribution prediction in the field of live streaming.

Real-time interactions on the Internet are permeating daily life, as exemplified by live streaming services. There are many different platforms and providers offering live streaming services, and competition is fierce. It is important for content providers to identify users who are likely to achieve higher contribution levels, such as revenue contribution rates. Thereby, content providers can focus more on high-potential users and increase revenue in a more efficient manner.

China Patent Application Publication No. 114268841 discloses a method for selecting users for live interaction.

A method according to an embodiment of the present invention is a viewer contribution prediction method executed by one or more computers, wherein the amount of contribution of a viewer to a first distributor is determined by the contribution of the first distributor. a step of determining that the value is less than a threshold; a step of obtaining the value of the interaction parameter of the viewer with respect to the first distributor; and a step of determining the value of the interaction parameter of the viewer with respect to the first distributor using a machine learning model. and determining a predicted degree of contribution of the viewer to the first distributor based on the method.

A system according to one embodiment of the invention is a system for audience contribution prediction that includes one or more processors, the one or more computer processors executing machine-readable instructions to a step of determining that the viewer's contribution amount to is less than a contribution threshold of the first distributor; a step of obtaining the value of an interaction parameter of the viewer with respect to the first distributor; determining a prediction of the degree of contribution of the viewer to the first distributor based on the interaction parameter value of the viewer with respect to the first distributor using the model.

A computer-readable medium according to an embodiment of the present invention includes a program for predicting viewer contribution level, and the program is configured to cause one or more computers to predict the amount of viewer contribution to a first distributor. a step of determining that the contribution level of the broadcaster is less than a threshold; a step of obtaining the value of an interaction parameter of the viewer for the first broadcaster; and determining a prediction of the degree of contribution of the viewer to the first distributor based on the value of the interaction parameter.

1 is a schematic diagram showing the configuration of a live streaming system 1 based on some embodiments of the present invention. 2 is a block diagram illustrating the functionality and configuration of user terminal 30 of FIG. 1, according to some embodiments of the present invention. FIG. FIG. 2 is a block diagram illustrating the functionality and configuration of the server of FIG. 1 in accordance with some embodiments of the present invention. 4 is a table showing an exemplary data structure of stream DB 310 of FIG. 3. FIG. 4 is a table showing an exemplary data structure of user DB 312 of FIG. 3. FIG. 4 is a table illustrating an exemplary data structure of gift DB 314 of FIG. 3. FIG. 4 is a table showing an exemplary data structure of the contribution degree DB 350 of FIG. 3. FIG. 3 is a graph illustrating exemplary daily contribution data between viewers and broadcasters. 4 is a table showing an exemplary data structure of the interaction parameter DB 352 of FIG. 3. FIG. 1 is an exemplary flowchart illustrating a method according to some embodiments of the invention. 3 is an example of how a machine learning model 370 determines or calculates contribution predictions in accordance with some embodiments of the present invention. 3 is a schematic diagram illustrating exemplary training and inference stages of a machine learning model 370 according to some embodiments of the present invention. FIG. FIG. 2 is a schematic diagram illustrating an example of a live streaming screen on a broadcaster's user terminal according to some embodiments of the present invention. FIG. 2 is a schematic diagram illustrating an example of a live streaming screen on a broadcaster's user terminal according to some embodiments of the present invention. 15(a), (b), and (c) are graphs illustrating examples of contribution prediction based on some embodiments of the present invention.

Hereinafter, the same or similar components, members, procedures, or signals shown in each drawing will be denoted by the same reference numerals in all the drawings, and therefore, redundant explanation will be omitted as appropriate. In addition, some members that are not important in the explanation of each drawing are omitted.

Traditionally, content providers (distributors, broadcasters, streamers, livers, anchors, etc.) on real-time interaction platforms (such as live streaming platforms) have the ability to make high contributions (such as revenue contributions) to the platform (or content provider). They don't have good tools to identify highly gendered audiences. The distributor cannot know who the potential contributors are and therefore cannot act accordingly.

The present invention provides a system or method for analyzing viewers and identifying potential contributors for a broadcaster. Therefore, streamers can direct their limited resources (such as time, awareness, or events) to higher potential viewers, increasing streamer rewards in a more accurate and more efficient manner. be able to.

FIG. 1 is a schematic diagram showing the configuration of a live streaming system 1 based on some embodiments of the present invention. The live streaming system 1 provides live streaming for streaming streamers (also called livers, anchors, distributors) LV and viewers (also called audiences) AUs (AU1, AU2...) to interact or communicate in real time. provide services; As shown in FIG. 1, the live streaming system 1 includes a server 10, a user terminal 20, and user terminals 30 (30a, 30b...). In some implementations, such streamers and viewers may be collectively referred to as users. The server 10 can include one or more information processing devices connected to the network NW. The user terminals 20 and 30 may be, for example, mobile terminal devices such as a smartphone, tablet, notebook PC, recorder, portable game machine, or wearable terminal, or may be a stationary device such as a desktop PC. The server 10, the user terminal 20, and the user terminal 30 are connected to be able to communicate with each other via various wired or wireless networks NW.

A distributor LV, a viewer AU, and an administrator (or an application provider, not shown) who manages the server 10 participate in the live streaming system 1. The distributor LV records the content on its own user terminal 20 and directly or indirectly uploads the content to the server 10, thereby distributing the content in real time. Examples of the content may be the distributor's own song, talk, performance, game play, and any other content. The administrator provides a platform for live streaming the content on the server 10, and mediates or manages real-time interactions between the broadcaster LV and the viewer AU. The viewer AU accesses the platform using his or her user terminal 30, selects and views desired content. The viewer AU performs operations such as commenting, cheering, sending gifts, etc. via the user terminal 30 during live streaming of the selected content. The distributor LV distributing the content may respond to those comments, cheers, or gifts. The response is sent in video and/or audio to the viewer AU and two-way communication is established.

The term "live streaming" refers to a data transmission mode that allows content recorded on the user terminal 20 of the broadcaster LV to be played back and viewed in substantially real time on the user terminal 30 of the viewer AU. However, it may also refer to a live broadcast realized by such a transmission mode. The live streaming uses HTTP live streaming, CMAF (Common Media Application Format), WebRTC (Web Real-Time Communications), and RTMP (Real-Time Messaging Protocol). , is realized using existing live streaming technologies such as MPEG DASH. It's okay. Live streaming includes a transmission mode in which the viewer AU can view the content with a predetermined delay at the same time as the content is recorded by the distributor LV. The length of the delay may be such that an interaction between the distributor LV and the viewer AU can be established. Note that the live streaming is distinguished from so-called on-demand distribution in which all recorded data of the content is once stored on the server and then provided from the server to the user in response to the user's request.

"Video data" here refers to image data (also called video data) generated using the imaging function of the user terminal 20 or 30 and data generated using the audio input function of the user terminal 20 or 30. Refers to data that includes voice data. The video data is played back on the user terminal 20, 30 so that the user can view the content. In some embodiments, the video data may be compressed, expanded, encoded, or decoded between the generation of the video data on the user terminal of the distributor and the playback of the video data on the user terminal of the viewer. , it is assumed that processing that changes its format, size, or data specifications, such as transcoding, will be performed. However, since the content (e.g., video and audio) represented by the video data does not substantially change before and after such processing, in this specification, the video data after such processing is referred to as its It is expressed as the same as the video data before such processing. In other words, when video data is generated on the user terminal of the distributor and then played back on the user terminal of the viewer via the server 10, the video data generated on the user terminal of the distributor, the The video data passing through the server 10 and the video data received and played back by the viewer's user terminal are the same video data.

In the example shown in FIG. 1, the distributor LV provides live streaming data. The user terminal 20 of the distributor LV records video and audio of the distributor LV to generate the streaming data, and the generated data is transmitted to the server 10 via the network NW. At the same time, the user terminal 20 displays the recorded video VD of the distributor LV on the display of the user terminal 20, so that the live streaming content currently being performed by the distributor LV can be confirmed.

The respective user terminals 30a and 30b of the viewers AU1 and AU2, who have requested the platform to view the live streaming of the distributor LV, transmit video data (hereinafter referred to as (which may also be referred to as "live streaming video data"), reproduces the received video data, displays the videos VD1 and VD2 on the display, and outputs audio from the speaker. The videos VD1 and VD2 displayed on the user terminals 30a and 30b, respectively, are substantially the same as the video VD captured by the user terminal 20 of the distributor LV, and are output on the user terminals 30a and 30b. The audio is substantially the same as the audio recorded at the user terminal 20 of the distributor LV.

Recording of video and audio on the user terminal 20 of the distributor LV and reproduction of video data on the user terminals 30a and 30b of the viewers AU1 and AU2 are performed substantially simultaneously. When the viewer AU1 inputs a comment regarding the content provided by the distributor LV into the user terminal 30a, the server 10 displays the comment in real time on the user terminal 30a of the distributor LV, and The comments are also displayed on the user terminals 30a and 30b of viewers AU1 and AU2, respectively. When the distributor LV reads the comment and develops a talk corresponding to the comment, the video and audio of the talk are displayed on the user terminals 30a and 30b of the viewers AU1 and AU2, respectively. This interactive operation is recognized as a conversation being established between the broadcaster LV and the viewer AU1. Thereby, the live streaming system 1 realizes live streaming that enables two-way communication instead of one-sided communication.

FIG. 2 is a block diagram illustrating the functionality and configuration of user terminal 30 of FIG. 1, according to some embodiments of the invention. The user terminal 20 has the same or similar functions and configuration as the user terminal 30. In each block in FIG. 2 and the subsequent block diagrams, the hardware may be realized by elements such as a CPU of a computer or a mechanical device, and the software may be realized by a computer program. Functional blocks may be realized by cooperative operations between these elements. Therefore, those skilled in the art will understand that these functional blocks can be implemented in various forms by combining hardware and software.

The distributor LV and the viewer AU may download a live streaming application program (hereinafter referred to as a live streaming application) from a download site via the network NW and install it on the user terminals 20 and 30. Alternatively, the live streaming application may be preinstalled on the user terminals 20 and 30. When the live streaming application is executed on the user terminals 20, 30, the user terminals 20, 30 communicate with the server 10 via the network NW and implement or execute various functions. Hereinafter, functions implemented by the user terminals 20 and 30 (processors such as CPUs) on which the live streaming application is executed will be described as functions of the user terminals 20 and 30. These functions are actually realized by the live streaming application on the user terminals 20, 30. In some embodiments, these functions are written in a programming language such as HTML (HyperText Markup Language), and are sent from the server 10 to the web browser of the user terminal 20, 30 via the network NW, and It may also be implemented by a computer program executed by a web browser.

The user terminal 30 includes a distribution unit 100 and a viewing unit 200. The distribution unit 100 generates video data in which video and audio of the user (or the user's side) are recorded, and provides the video data to the server 10. The viewing unit 200 receives video data from the server 10 and plays the video data. The user activates the distribution unit 100 when performing live streaming, and activates the viewing unit 200 when viewing video. The user terminal on which the distribution unit 100 is activated is the distributor's terminal, that is, the user terminal that generates the video data. The user terminal on which the viewing unit 200 is activated is the viewer's terminal, that is, the user terminal on which the video data is reproduced and played.

The distribution unit 100 includes an imaging control unit 102, an audio control unit 104, a video transmission unit 106, and a distributor side UI control unit 108. The imaging control unit 102 is connected to a camera (not shown in FIG. 2) and controls imaging performed by the camera. The imaging control unit 102 acquires image data from the camera. The audio control unit 104 is connected to a microphone (not shown in FIG. 2) and controls audio input from the microphone. The audio control unit 104 acquires the audio data from the microphone. The video transmission unit 106 transmits video data including the image data acquired by the imaging control unit 102 and the audio data acquired by the audio control unit 104 to the server 10 via the network NW. do. The video data is transmitted in real time by the video transmission unit 106. That is, the generation of the video data by the imaging control unit 102 and the audio control unit 104 and the transmission of the generated video data by the video transmission unit 106 are executed substantially simultaneously. The distributor side UI control unit 108 controls the UI (user interface) of the distributor. The distributor side UI control unit 108 may be connected to a display (not shown in FIG. 2), and displays the video on the display by playing the video data transmitted by the video transmission unit 106. do. The distributor side UI control unit 108 may display an operation object or an instruction permission object on the display, and may receive an input from the distributor who taps the object.

The viewing unit 200 includes a viewer-side UI control unit 202, a superposition information generation unit 204, and an input information transmission unit 206. The viewing unit 200 receives video data related to live streaming in which the distributor, the viewer who is the user of the user terminal 30, and other viewers participate from the server 10 via the network NW. . The viewer-side UI control unit 202 controls the UI of the viewer. The viewer-side UI control unit 202 is connected to a display and a speaker (not shown in FIG. 2), plays the received video data, displays the video on the display, and outputs audio from the speaker. A state in which the video is output to the display and the audio is output from the speaker can be referred to as a state in which "video data is being played back." The viewer-side UI control unit 202 is also connected to input means (not shown in FIG. 2) such as a touch panel, keyboard, and display, and obtains user input via the input means. The superposition information generation unit 204 superimposes a predetermined frame image on the image generated from the video data from the server 10. The frame image includes various user interface objects (hereinafter simply referred to as "objects") for accepting input from the user, comments input by the viewer, and/or information obtained from the server 10. etc. are included. The input information transmission unit 206 transmits the user input acquired by the viewer-side UI control unit 202 to the server 10 via the network NW.

FIG. 3 shows a block diagram illustrating the functionality and configuration of the server of FIG. 1, according to some embodiments of the invention. The server 10 includes a distribution information providing unit 302, a relay unit 304, a gift processing unit 306, a payment processing unit 308, a stream DB 310, a user DB 312, a gift DB 314, a user identification unit 330, and a distribution time proposal unit. It includes a unit 332, a contribution DB 350, and an interaction parameter DB 352. The server 10 communicates with a machine learning (ML) model 370.

Upon receiving a live streaming start notification or request from the user terminal 20 of the distributor via the network NW, the distribution information providing unit 302 sends the stream ID for identifying this live streaming and the live streaming. The distributor ID of the distributor who performs the broadcast is registered in the stream DB 310.

When the distribution information providing unit 302 receives a request for providing information regarding a live stream from the viewing unit 200 of the user terminal 30 on the viewer side via the network NW, the distribution information providing unit 302 Obtain or check the currently available live streams from the DB 310 and create a list of available live streams. The distribution information providing unit 302 transmits the created list to the requesting user terminal 30 via the network NW. The viewer-side UI control unit 202 of the requesting user terminal 30 generates a live stream selection screen based on the received list, and displays it on the display of the user terminal 30 .

When the input information transmission unit 206 of the user terminal 30 receives the selection result of the viewer on the live stream selection screen, the input information transmission unit 206 sends a distribution request including the stream ID of the selected live stream. and transmits the request to the server 10 via the network NW. The distribution information providing unit 302 starts providing the live stream specified by the stream ID included in the received distribution request to the user terminal 30 that is the request source. The distribution information providing unit 302 updates the stream DB 310 to include the user ID of the viewer of the requesting user terminal 30 in the (or corresponding) viewer ID of the stream ID.

The relay unit 304 relays the video data from the user terminal 20 on the distributor side to the user terminal 30 on the viewer side in the live streaming started by the distribution information providing unit 302. The relay unit 304 receives a signal representing a user input by a viewer from the input information transmitting unit 206 during the live streaming or playing the video data. The signal representing the user input may be an object designation signal that designates an object displayed on the display of the user terminal 30. The object designation signal may include the viewer ID of the viewer, the distributor ID of the distributor of the live stream that the viewer is viewing, and an object ID that specifies the object. When the object is a gift, the object ID is a gift ID. Similarly, the relay unit 304 receives from the distribution unit 100 of the user terminal 20 a signal representing a user input performed by the distributor during playback (or live streaming) of the video data. The signal may be an object designation signal.

Further, the signal representing the user input may be a comment input signal including a comment input by the viewer into the user terminal 30 and the viewer ID of the viewer. Upon receiving the comment input signal, the relay unit 304 transmits the comment and the viewer ID included in the signal to the user terminal 20 of the distributor and the user terminal 30 of other viewers. In these user terminals 20 and 30, the viewer-side UI control unit 202 and the superimposition information generation unit 204 display the received comment on the display associated with the viewer ID that was also received.

The gift processing unit 306 increases the points of the distributor and updates the user DB 312 based on the points of the gift specified by the gift ID included in the object designation signal. Specifically, the gift processing unit 306 refers to the gift DB 314 and specifies points to be given to the gift ID included in the received object designation signal. Thereafter, the gift processing unit 306 updates the user DB 312 and adds the specified points to the points of the distributor ID (or corresponding to the distributor ID) included in the object designation signal.

The payment processing unit 308 processes payment for the gift from the viewer in response to receiving the object designation signal. Specifically, the payment processing unit 308 refers to the gift DB 314 to identify the price point of the gift specified by the gift ID included in the object designation signal. Thereafter, the payment processing unit 308 updates the user DB 312 and subtracts the identified price points from the points of the viewer identified by the viewer ID included in the object designation signal.

FIG. 4 is a table showing an exemplary data structure of the stream DB 310 of FIG. The stream DB 310 holds information regarding the live stream currently being performed. The stream DB 310 stores stream IDs, distributor IDs, and viewer IDs in association with each other. The stream ID is an ID for identifying a live stream on the live streaming platform provided by the live streaming system 1. The distributor ID is a user ID for identifying the distributor who provides the live stream. The viewer ID is a user ID for identifying the viewer of the live stream. In the live streaming platform provided by the live streaming system 1 according to some embodiments, when a user starts a live stream, the user becomes a broadcaster, and when the same user watches a live stream broadcast by another user, the user becomes a broadcaster. , the user also becomes a viewer. Therefore, the distinction between a distributor and a viewer is not fixed, and a user ID that is registered as a distributor ID at one time may be registered as a viewer ID at another time.

FIG. 5 is a table illustrating an exemplary data structure of user DB 312 of FIG. The user DB 312 holds information regarding users. The user DB 312 stores a user ID, points, and one or more attributes (attributes 1, 2, . . . ) in association with each other. The user ID identifies the user. The points correspond to points held by the corresponding user. The points are electronic value that circulates within the live streaming platform. In some implementations, when a broadcaster receives a gift from a viewer during a live stream, the broadcaster's points increase by the value corresponding to the gift. The points are used, for example, to determine the amount of compensation (such as money) that the broadcaster receives from the administrator of the live streaming platform. In some embodiments, when the broadcaster receives a gift from a viewer, an amount corresponding to the gift may be given instead of the points.

Examples of attributes may include gender, country, number of followers, number of followers, billing amount, and/or viewer level. The number of followers is the number of distributors that the corresponding user follows. The number of followers is the number of viewers who follow the corresponding user. The charged amount is the amount charged to the platform by the corresponding user. The viewer level may increase with the corresponding user's billing amount. Some attributes, such as gender, may be input by the user when registering on the platform. Some attributes, such as country, may be detected within the server 10 or outside the server 10 by a detection unit (not shown in FIG. 3).

FIG. 6 is a table illustrating an exemplary data structure of gift DB 314 of FIG. The gift DB 314 holds information about gifts available to the viewer during the live streaming. The gift is electronic data. Gifts may be purchased with points or money, or may be provided free of charge. Gifts can be given by viewers to broadcasters. Giving a gift to a streamer is also called using a gift, sending a gift, or throwing a gift. Some gifts can be used at the same time as the purchase, while others can be used at any time by the viewer after purchase. When a viewer gives a gift to a broadcaster, points corresponding to the gift are given to the broadcaster. When a gift is used, effects associated with the gift may occur due to its use. For example, effects (visual effects, auditory effects, etc.) corresponding to the gift are displayed on the live streaming screen.

The gift DB 314 stores gift IDs, awarded points, and price points in association with each other. The gift ID is for identifying the gift. The given points are the amount of points given to the distributor when a gift is given to the distributor. The price point is the amount of points paid for the use (purchase) of the gift. While viewing the live stream, the viewer can give the desired gift to the broadcaster by paying the desired gift's price point. Payment of such price points may be made by any suitable electronic payment means. For example, payment may be made by the viewer paying the administrator the price points. Alternatively, payment by bank transfer or credit card may be used. The administrator can arbitrarily set the relationship between the granted points and the price points. For example, it may be set such that awarded points=price points. Alternatively, points obtained by multiplying the given points by a predetermined coefficient such as 1.2 may be set as the price points, or points obtained by adding predetermined fee points to the given points may be set as the price points. .

The user identification unit 330 is configured to identify (or determine or predict) whether a particular viewer is a top contributor candidate (or high contributor candidate) for a particular broadcaster. The user identification unit 330 may refer to the stream DB 310, the user DB 312, the gift DB 314, the contribution DB 350, and the interaction parameter DB 352, and use the machine learning model 370 for determination or prediction. The result of the determination or prediction may be stored or registered in the contribution degree DB 350 as a top contributor tag and/or a top contributor candidate tag. Details will be described later.

FIG. 7 is a table showing an exemplary data structure of the contribution degree DB 350 of FIG. The contribution DB 350 holds the broadcaster ID, the viewer ID, the past contribution, the threshold (or the contribution threshold), the current contribution, the top contributor tag, and the top contributor candidate tag in association with each other. do.

The past contribution level is the amount of contribution from the viewer to the distributor in a predetermined period (for example, the past month).

The threshold value or contribution level threshold value is used to determine whether the viewer is a top contributor (or high contributor) for the distributor. The threshold is the top T% (or first T% quantile) of the total contribution to the broadcaster from all viewers of the broadcaster. In some implementations, the threshold may be the first Y quantile of the total contribution to the broadcaster from the broadcaster's total audience. Therefore, the threshold value is different for each distributor. Further, the threshold value is updated every predetermined time.

The current contribution degree is updated in real time based on the contribution status from the viewer to the distributor in the current calculation period.

The top contributor tag is a tag that indicates whether or not the viewer has already been registered as a top contributor (or high contributor, or top spender) with the distributor. When the current contribution exceeds the contribution threshold, the corresponding viewer is registered as a top contributor of the corresponding broadcaster. For example, viewer V2's current degree of contribution [2500 points] has already exceeded the threshold value [2400 points] of distributor D1, so viewer V2 is registered as a top contributor of distributor D1. However, viewer V2 may not be a top contributor for other broadcasters. The registration of the top contributor tag may be performed by the user identification unit 330.

The top contributor candidate tag indicates that a viewer who has not yet become a top contributor of a certain broadcaster may become a top contributor of that broadcaster at some point in the future (for example, half a month or one month later). This is a tag that indicates whether there is a gender or not. In some implementations, the top contributor candidate tag indicates whether contributions from viewers will exceed a threshold corresponding to the broadcaster at some point in the future. The tag determination (or prediction) may be performed by the user identification unit 330 using the machine learning model 370. At the time of determining the top contributor candidate tag, if the viewer is already a top contributor of the broadcaster (eg, viewer V2 for broadcaster D1), the tag may be marked NA (not applicable). In this embodiment, at the time of determining the top contributor candidate tag, viewer V3 has not yet been registered as a top contributor to distributor D1, but has been tagged as a top contributor candidate. In other words, based on the prediction result by the machine learning model 370, there is a high possibility that the contribution of viewer V3 to distributor D1 will exceed the contribution threshold corresponding to distributor D1 at some point in the future (or the probability is (is above the threshold value). In some implementations, the unit of contribution may be points. In some implementations, the unit of contribution may be any monetary unit or currency.

FIG. 8 is a graph illustrating exemplary daily contribution data between viewers and broadcasters. The daily contribution data indicates the contribution of each day, such as points given by viewers to the distributor. Similar contribution data is stored for each viewer-contributor pair. The contribution data may be stored as part of the contribution DB 350.

FIG. 9 is a table showing an exemplary data structure of the interaction parameter DB 352 of FIG. The interaction parameter DB 352 holds a distributor ID, a viewer ID, and one or more interaction parameters (interaction parameters 1, 2, . . . ).

The interaction parameters shown in FIG. 9 are associated with each broadcaster/viewer pair. Examples of the interaction parameters may include a contribution rate, a total contribution rate, an average number of comments, an audience rating, a contribution correlation coefficient, a comment correlation coefficient, and an audience rating correlation coefficient.

The contribution rate may be defined as [the contribution from the viewer to the distributor within a certain period] divided by the [contribution level threshold of the distributor]. The total contribution rate may be defined as [total contributions from viewers within a certain period of time (including contributions to other broadcasters)] divided by [contribution level threshold of the broadcaster]. The average number of comments may be defined as the number of comments from viewers to the broadcaster (or on the broadcaster's live stream) divided by the number of live streams started by the broadcaster. The audience rating may be defined as the amount of time viewers watched a broadcaster's live stream divided by the total length of the broadcaster's live stream.

The contribution correlation coefficient indicates the tendency of contribution (from the viewer to the distributor) over time. The value of the contribution correlation coefficient can be set in the range of -1 to 1. A positive value means that the contribution increases over time, a negative value means that the contribution decreases over time.

The comment correlation coefficient indicates the tendency of the amount of comments (from viewers to distributors) to fluctuate over time. The value of the comment correlation coefficient can be set in the range of -1 to 1. A positive value means that comments increase over time, and a negative value means that comments decrease over time.

The audience rating correlation coefficient indicates the tendency of the audience rating to fluctuate over time. The value of the audience rating correlation coefficient can be set in the range of -1 to 1. A positive value means viewers are watching the broadcaster's live stream less (or at a higher percentage), and a negative value means viewers are watching the broadcaster's live stream less (or at a higher rate). or to a lesser extent).

The machine learning model 370 may be a machine learning DB containing one or more machine learning models. In some implementations, the machine learning model 370 may be implemented within the server 10. The machine learning model 370 may include or utilize machine learning algorithms such as supervised learning algorithms or gradient boosting algorithms. A Light Gradient Boosting Machine (LightGBM) may be implemented. A decision tree algorithm may be implemented.

FIG. 10 shows an exemplary flowchart illustrating a method according to some embodiments of the invention.

In step S1000, the server 10 (or the distribution information providing unit 302, etc.) receives a contribution prediction request. The request is associated with a viewer and a distributor, and includes a viewer ID and a distributor ID. The request may be from the broadcaster (or the broadcaster's terminal) or the operator of the streaming platform.

In step S1002, the server 10 determines/identifies the viewer and the distributor associated with the contribution prediction request based on the viewer ID and the distributor ID included in the contribution prediction request. do. This process may be performed by the user identification unit 330, for example.

In step S1004, the server 10 determines whether the current degree of contribution of the viewer (in the current calculation session) to the distributor has already reached the threshold of the distributor. If “yes”, the flow advances to step S1006. If "no", the flow advances to step S1008. The server 10 makes the determination by referring to the contribution degree DB 350. The determination process may be executed by the user identification unit 330, for example.

In step S1006, the viewer is displayed as a top contributor on the requesting side. For example, the viewer is displayed as a top contributor on the screen of the user terminal of the distributor who initiated the contribution prediction request. Specifically, the server 10 can transmit the result of step S1004 indicating that the viewer is already a top contributor of the broadcaster to the requesting side, and display the viewer as a top contributor on the requesting side. Do it like this.

In step S1008, the user identification unit 330 refers to the user DB 312 for user attributes (attributes of the viewer and the distributor, attributes of other viewers and other distributors). The user identification unit 330 refers to the contribution DB 350 for contribution data between viewers and distributors. The user identification unit 330 refers to the interaction parameter DB 352 for interaction parameters (between the viewer and the distributor, and/or between the viewer and other distributors, etc.).

In step S1010, the user identification unit 330 inputs the data or information received in step S1008 to the machine learning model 370. The machine learning model 370 determines or calculates a prediction of the viewer's contribution to the broadcaster based on the above-mentioned input, and sends the contribution prediction result to the user identification unit 330. In some implementations, the contribution prediction results may be calculated based on the viewer's contribution (to the broadcaster) (or the current It may also be a probability (or probability value) of the possibility that the contribution (degree of contribution) exceeds the contribution threshold of the distributor. Thereafter, the user identification unit 330 determines the value (or type/class) of the top contributor candidate tag (in the contribution DB 350) of the viewer for the broadcaster by comparing the probability with the probability threshold. may be determined. The probability threshold may be determined by the streaming platform operator. The probability threshold may be, for example, 50%, 75%, or 90%.

In step S1012, the contribution prediction result is displayed on the request side. For example, on the screen of the user terminal of the broadcaster who initiated the contribution prediction request, the viewer ID of the viewer and the predicted probability value (or the top contribution indicated by "yes" or "no") are displayed. person candidate tag) may be displayed. Specifically, the server 10 transmits the prediction result of step S1010 to the requesting side, so that the prediction result can be displayed on the requesting side.

In some implementations, the contribution prediction process described above is performed in real time. That is, when the server 10 receives a contribution prediction request regarding a viewer/broadcaster pair (for example, the viewer participates in the chat room of the broadcaster, the broadcaster requests a contribution prediction request regarding the viewer/broadcaster), (initiating a request), the user identification unit 330 accesses the user DB 312, the contribution DB 350, and the interaction parameter DB 352 in real time for attribute data, contribution data, and interaction parameter data. Thereafter, the above-described feature amount data is input to the machine learning model 370, and a prediction result is generated in real time. The stream DB 310, the user DB 312, the gift DB 314, the contribution DB 350, and/or the interaction parameter DB 352 are also updated in real time. The update of the database may be realized by communication with a service or database external to the server 10, such as a cloud service or a BigQuery service.

In some implementations, batch processing may be incorporated or utilized in the contribution prediction process. For example, the stream DB 310, the user DB 312, the gift DB 314, the contribution DB 350, and/or the interaction parameter DB 352 are updated once a day at a constant timing. The updated data (attribute data, contribution data, interaction parameter data, etc.) is used by the machine learning model 370 as a feature amount to generate a contribution prediction result for each viewer/distributor pair. Thereafter, the contribution prediction result is stored in the contribution DB 350, for example. After that, when the viewer participates in the broadcaster's chat room and the broadcaster starts a contribution prediction request regarding the viewer, the user identification unit 330 identifies the prediction result in the contribution degree DB 350 (prepared on the previous day or (which may have been calculated) and respond to the distributor. In some implementations, a batch-based prediction process may lack up-to-date information in prediction calculations, but may save database update time and/or machine learning model inference time. In some implementations, the batch-based prediction process provides prediction results more quickly and may lead to a better user experience.

FIG. 11 is an example of how machine learning model 370 determines or calculates contribution predictions in accordance with some implementations of the invention.

As shown in FIG. 11, the feature amounts input to the machine learning model 370 include interaction parameters between the viewer and the distributor, attributes of the distributor, and attributes of the viewer. The feature amount is acquired from various databases such as the stream DB 310, the user DB 312, the contribution DB 350, and/or the interaction parameter DB 352. Feature quantities (or feature quantity values) for the past X days are input to the machine learning model 370. The purpose is to determine whether the viewer's contribution to the distributor exceeds the distributor's contribution threshold at a future timing (for example, after Y days). In some implementations, the objective is to calculate a probability that indicates how likely the viewer's contribution to the broadcaster will be over the broadcaster's contribution threshold at a future time (Y days later). It consists in generating.

FIG. 12 is a schematic diagram illustrating exemplary training and inference stages of a machine learning model 370 according to some implementations of the invention. In this embodiment, the machine learning model 370 is retrained periodically (eg, weekly, monthly, yearly, etc.) using the latest available data. For example, the machine learning model 370 is retrained on the first of each month. In the (re)training phase, data obtained in the last month is used for training and validation, and data obtained two months ago is used to calculate the threshold. (Re)learning may be performed manually or automatically. In the example shown in FIG. 12, the training phase includes a period in June and a period in July, which are in the past. The inference phase includes a period of August, which is in the future.

For the June period (or at the end of the June period), the top 10% (or first 10% quintile, or top 10 contributions) of each broadcaster's total received contributions for the June period. The contribution threshold of each distributor is calculated using .

During the July period, the contribution data between each viewer/broadcaster pair, the interaction parameters between each viewer/broadcaster pair, the attributes of each viewer, and the attributes of each broadcaster were analyzed by the machine learning model 370. Obtained as training data. The machine learning model 370 may use training data as a training set and a validation set, perform contribution prediction, and perform self-learning based on ground truth contribution data. The contribution prediction is based on the probability that the viewer's contribution to the broadcaster will exceed the broadcaster's contribution threshold (calculated from data for the June period) at a timing within the July period. It may also be a probability indicating that.

For the August period, which is the inference stage, we use the top 10% (or first 10% quantile, or top 10 contributions) of each broadcaster's total received contribution for the July period to A new contribution threshold for the distributor is calculated. Contribution data between each viewer/broadcaster pair, interaction parameters between each viewer/broadcaster pair, attributes of each viewer, and attributes of each broadcaster are new features that are input into the machine learning model 370. Obtained as a quantity. The machine learning model 370 then provides an inference result that includes the probability that the viewer's contribution to the broadcaster will exceed the broadcaster's new contribution threshold at a future time. .

In some implementations, feature data from the past 10 days (X=10) is used by the machine learning model 370 to generate contribution predictions for the future 14 days (or 14 days from now, Y=14). used for. For example, when calculating contribution prediction on August 14th, the feature amount data input to the machine learning model 370 is calculated from August 5th to August 14th (or from August 4th to August 13th). ), and the output of the machine learning model 370 includes a contribution prediction for August 27th (or August 28th).

FIG. 13 shows a schematic diagram illustrating an example of a live streaming screen on a broadcaster's user terminal according to some embodiments of the present invention.

The distributor LV can request the contribution prediction by tapping the viewer ID (or viewer name) V14. After that, the contribution prediction result calculated by the machine learning model 370 is displayed. In this embodiment, the probability that viewer V14 will become a top contributor of broadcaster LV at a future timing (that is, the contribution from viewer V14 will exceed the contribution threshold of broadcaster LV) is (0.7) =70%). In some embodiments, a "top contributor candidate tag" may be displayed that indicates whether viewer V14 has a possibility of becoming a top contributor of broadcaster LV.

FIG. 14 shows a schematic diagram illustrating an example of a live streaming screen on a broadcaster's user terminal according to some embodiments of the present invention.

During the live stream, the distributor's device transmits a current viewer list acquisition request to the server 10 in response to an instruction from the distributor to display the current viewer list. The request includes the stream ID of the live stream. The server 10 specifies a list of viewer IDs corresponding to the stream ID included in the request by referring to the stream DB 310. Thereafter, the server 10 refers to the contribution degree DB 350, and for each viewer ID in the specified list, the server 10 identifies the viewer (of each viewer ID) as a top contributor or a top contributor candidate. It is determined whether it is attached or not. The server 10 adds the determination result to each of the viewer IDs in the specified list. The server 10 transmits the specified list to the device of the requesting distributor. The device generates a current viewer list area R1 and displays it within the live stream room screen. The current viewer list is displayed in this area. In this area, viewers tagged as top contributors are displayed at the top, and viewers tagged as top contributor candidates are displayed below. The top contributors and/or top contributor candidates are displayed more prominently than others. This allows the distributor to easily recognize who to pay attention to and with whom to communicate more.

In some embodiments, the batch-type contribution prediction process described above may be used to determine the top contributor tags and/or top contributor candidate tags of the contribution DB 350 in the embodiment of FIG. . This can save time compared to determining or calculating those tags in real time for all current viewers.

As shown in FIGS. 7, 8, and 9, the feature amounts input to the machine learning model 370 include contribution data from the contribution DB 350 and interaction parameters from the interaction parameter DB 352. The contribution data and the interaction parameters are unique or associated for each viewer/broadcaster pair. Therefore, the prediction of the degree of contribution of one viewer to one broadcaster provided by the machine learning model 370 is not only correlated to the interaction between the viewer and the broadcaster, but also correlated with the interaction between the viewer and other broadcasters. It may also be correlated to interactions between distributors. That is, the predicted contribution of one viewer to one distributor may be determined by the machine learning model 370 based on interaction parameters and/or contribution data between the viewer and other distributors.

For example, in some embodiments, the machine learning model 370 is based on the contribution data in the contribution DB 350 (or based on the contribution correlation between the viewer V1 and the broadcaster D2 in the interaction parameter DB 352). (based on the value of the relationship number), an increase in the degree of contribution (or an increasing tendency in the degree of contribution) of the viewer V1 to the distributor D2 is detected or determined. The machine learning model 370 determines that the predicted contribution of the viewer V1 to the distributor D1 is large even if the increasing tendency of the viewer V1's contribution to the distributor D1 is not clear at the time of prediction. It's okay. In some implementations, an increase in contribution from the viewer V1 to the broadcaster D2 may indicate an increase in the viewer V1's ability or desire to contribute. Therefore, by presenting to the broadcaster D1 that the predicted contribution value of the viewer V1 is increasing, the broadcaster D1 can pay more attention to the viewer V1 and increase interaction with the viewer V1. By doing so, it is possible to increase the possibility of receiving a contribution from the viewer V1.

In some implementations, the machine learning model 370 detects or determines that the attributes of the distributor D1 are different or opposite to the attributes of the distributor D2. In that case, the predicted degree of contribution of the viewer to the distributor D1 determined by the machine learning model 370 and the interaction parameter value of the viewer to the distributor D2 may be inversely correlated. For example, if the attributes of the streamer D1 and the streamer D2 indicate that they usually stream political content and have diametrically opposed political positions, the viewer V1 may express interest in the streamer D2. Increasing may indicate that interest in the distributor D1 is decreasing. Therefore, the machine learning model 370 assumes that interaction parameters (contribution correlation coefficient, comment correlation coefficient, audience rating correlation coefficient, etc.) between viewer V1 and broadcaster D2 indicate an increase in interaction or an increase in contribution. In this case, the predicted contribution level of the viewer V1 to the distributor D1 may be determined to be low. After that, the distributor D1 may pay more attention to other viewers who have a higher degree of contribution or increase interaction with them based on the prediction result. In some implementations, the user's attributes may include the user's political stance, which is determined by a machine learning model based on the user's statements and comments detected by the detection unit. Good too.

15(a), 15(b), and 15(c) are graphs illustrating examples of contribution predictions based on some embodiments of the present invention. The X-axis is the daily contribution. The Y axis is the timeline. Contribution prediction indicates the probability that a viewer will become a top contributor to a broadcaster at some point in the future.

In Example 1 shown in FIG. 15(a), the contribution prediction increased from 15.9% calculated at 8/28 to 67.1% calculated at 9/2. The viewer may be tagged as a top contributor candidate as of 9/2. As shown in the figure, the actual contribution level increased after that, and the viewer became a top contributor as of September 10th.

In Example 2 shown in FIG. 15(b), the contribution prediction decreased from 36.9% calculated at 9/1 to 1.2% calculated at 9/19. The viewer will not be tagged as a potential top contributor. As shown in the figure, the actual degree of contribution continued to be low, and the viewer did not become a top contributor.

In Example 3 shown in FIG. 15(c), the contribution prediction increased from 25.3% calculated at 9/5 to 92.5% calculated at 9/20. The viewer may be tagged as a top contributor candidate as of 9/20. As shown in the figure, the actual contribution has increased slightly since then. In reality, the broadcaster will refer to the prediction results for 9/20 to pay more attention to (or deepen interaction with) that viewer, maximize the possibility of contribution from that viewer, and/or ) You can actually make that viewer a top contributor.

When a broadcaster starts a live stream, live stream notifications (or push notifications) are sent to viewers who follow the broadcaster. In some implementations, when a broadcaster begins a live stream, the streaming system identifies the broadcaster's top potential contributors (with reference to contribution DB 350) and selects only those top contributors; or Send notifications with higher priority. Thereby, the effect of notification can be maximized in an environment where resources related to notification processing (bandwidth, CPU usage rate, etc. of the notification system that communicates with the server 10) are limited. In addition, the batch-type contribution prediction process can be used to pre-update top contributor candidate tags in the contribution DB 350, thereby speeding up the process of identifying top contributor candidates and the notification process. can.

By identifying a broadcaster's top contributor candidates, the present invention can enable the broadcaster to interact more with those people or allow the broadcaster to focus more on those people. The present invention can identify viewers who are already top or loyal contributors to a broadcaster and allow the broadcaster to maintain current interactions with those people. The present invention can identify viewers who are less likely to become top contributors for a broadcaster, allowing the broadcaster to interact more with other viewers who have higher potential to contribute.

The processes and procedures described in this invention, in addition to those explicitly described, may be implemented by software, hardware, or any combination thereof. For example, the processes and procedures described herein may be implemented in a medium such as an integrated circuit, volatile memory, non-volatile memory, non-transitory computer-readable medium, magnetic disk, or the like. This can be realized by Further, the processes and procedures described in this specification can be realized as a computer program corresponding to the processes and procedures, and can be executed by various computers.

Additionally, the systems or methods described in the above embodiments may be integrated into programs stored on non-transitory computer-readable media, such as solid state storage, optical disk storage, magnetic disk storage, and the like. Alternatively, the program may be downloaded from a server via the Internet and executed by the processor.

Although the technical contents and features of the present invention have been described above, those with ordinary knowledge in the technical field to which the present invention pertains will appreciate that many variations and modifications can be made without departing from the teachings and disclosure of the present invention. It can be performed. Therefore, the scope of the invention is not limited to the embodiments already disclosed, but includes other variations and modifications that do not depart from the invention, as falling within the scope of the claims.

1 Communication system 10 Server 20 User terminals 30, 30a, 30b User terminal LV Distributor AU1, AU2 Viewer VD, VD1, VD2 Video NW Network 30 User terminal 100 Distribution unit 102 Imaging control unit 104 Audio control unit 106 Video transmission unit 108 Distributor side UI control unit 200 Viewing unit 202 Viewer side UI control unit 204 Overlay information generation unit 206 Input information transmission unit 302 Distribution information provision unit 304 Relay unit 306 Gift processing unit 308 Payment processing unit 310 Stream DB
312 User DB
314 Gift DB
330 User identification unit 332 Delivery time suggestion unit 350 Contribution DB
352 Interaction parameter DB
370 Machine learning models S1000, S1002, S1004, S1006, S1008, S1010, S1012 Process R1 Current viewer list

Claims (10)

A viewer contribution prediction method performed by a server, the method comprising:
determining that the viewer's contribution to the first distributor is less than the first distributor's contribution threshold;
In response to determining that the viewer's contribution to the first distributor is less than the first distributor's contribution threshold, the value of the interaction parameter of the viewer with respect to the first distributor is determined. The process of acquiring
using a machine learning model to determine a predicted degree of contribution of the viewer to the first distributor based on the value of the interaction parameter of the viewer to the first distributor;
A viewer contribution prediction method characterized by comprising: moreover,
comprising the step of acquiring attributes of the viewer,
The step of determining a prediction of the viewer's contribution to the first distributor is based on the attribute of the viewer and the value of the interaction parameter of the viewer with respect to the first distributor, using the machine learning model. The viewer contribution prediction method according to claim 1, further comprising determining the contribution prediction. moreover,
obtaining a value of an interaction parameter of the viewer with a second broadcaster;
The step of determining a prediction of the viewer's contribution to the first broadcaster includes determining, by the machine learning model, the value of the interaction parameter of the viewer for the second broadcaster and the prediction of the viewer's contribution level for the first broadcaster. The viewer contribution prediction method according to claim 1, further comprising determining the contribution prediction based on the value of the interaction parameter of the viewer . The step of determining a contribution prediction of the viewer to the first broadcaster further comprises:
determining an increase in the viewer's contribution to the second distributor based on the value of the interaction parameter of the viewer with the second distributor;
4. The method according to claim 3, further comprising the step of determining a degree of contribution prediction in which the degree of contribution of the viewer to the first distributor increases in response to the determination of the increase in degree of contribution. A method for predicting viewer contribution. moreover,
acquiring attributes of the first distributor and attributes of the second distributor from a user database in the server;
determining that the attribute of the first distributor is different from the attribute of the second distributor;
including;
The step of determining the prediction of the degree of contribution of the viewer with respect to the first broadcaster includes: determining a contribution prediction that is inversely correlated with a value of the interaction parameter of the viewer for the second broadcaster;
The viewer contribution prediction method according to claim 3, characterized in that: The viewer contribution prediction method according to claim 1, wherein the machine learning model is trained with interaction parameters of the viewer with respect to the distributor and values of the viewer's contribution with respect to the distributor. . The viewer contribution prediction according to claim 1, wherein the contribution prediction is a probability that the viewer will reach the contribution threshold of the first distributor at some point in the future. Method. moreover,
Displaying the determined contribution prediction of the viewer to the first distributor on a display of the first distributor's terminal during live streaming of the first distributor. The viewer contribution prediction method according to claim 1, characterized in that: A system for viewer contribution prediction, comprising one or more processors, the one or more processors executing machine-readable instructions;
determining that the viewer's contribution to the first distributor is less than the first distributor's contribution threshold;
In response to determining that the viewer's contribution to the first distributor is less than the first distributor's contribution threshold, the value of the interaction parameter of the viewer with respect to the first distributor is determined. The process of acquiring
using a machine learning model to determine a predicted degree of contribution of the viewer to the first distributor based on the value of the interaction parameter of the viewer to the first distributor;
A system for predicting viewer contribution, characterized by performing the following. A non-transitory computer-readable medium comprising a program for predicting viewer contribution, wherein the program is stored on one or more computers;
determining that the viewer's contribution to the first distributor is less than the first distributor's contribution threshold;
In response to determining that the viewer's contribution to the first distributor is less than the first distributor's contribution threshold, the value of the interaction parameter of the viewer with respect to the first distributor is determined. The process of acquiring
using a machine learning model to determine a predicted degree of contribution of the viewer to the first distributor based on the value of the interaction parameter of the viewer to the first distributor;
1. A non-transitory computer-readable medium containing a program for predicting viewer contribution, characterized in that the program is executed.

Images