User account matching method and apparatus, and model training method and apparatus

Abstract

This disclosure provides a user account matching method and apparatus, as well as a model training method and apparatus. The user account matching method may include: acquiring multiple user account sets for a target user account; obtaining multiple user account matching models corresponding to multiple live streaming data scenarios based on the multiple user account sets; performing live streaming matching on candidate user accounts using the multiple user account matching models to obtain the live streaming matching probability of the candidate user account in each of the multiple live streaming data scenarios, wherein the live streaming matching probability represents the probability of watching the target user account's live stream through the candidate user account; and determining the user account that matches the target user account's live stream based on the live streaming matching probability of the candidate user account in each live streaming data scenario.

Description

Technical Field

[0001] This disclosure relates to the field of data classification and processing technology, and in particular to a user account matching method and apparatus, as well as a training method and apparatus for a user account matching model. Background Technology

[0002] Currently, before or during a live stream, user account matching is typically performed to identify the streamer's potential audience (those willing to watch the live stream). The stream is then promoted within this identified potential audience, ensuring these users can access the streamer's information promptly and accurately, thereby increasing viewership. Therefore, the accuracy of user account matching is directly related to the streamer's performance. Consequently, improving user account matching to enhance the accuracy of potential audience targeting has become a hot research topic. Summary of the Invention

[0003] This disclosure provides a user account matching method, a user account matching apparatus, and a training method and apparatus for a user account matching model, to at least solve the aforementioned problems.

[0004] According to a first aspect of this disclosure, a user account matching method is provided, which may include: obtaining multiple user account sets for a target user account, wherein the multiple user account sets correspond to multiple live streaming data scenarios of the target user account, and each user account set in the multiple user account sets includes features corresponding to the multiple user accounts; obtaining multiple user account matching models corresponding to the multiple live streaming data scenarios based on the multiple user account sets; performing live streaming matching on candidate user accounts using the multiple user account matching models to obtain the live streaming matching probability of the candidate user account in each of the multiple live streaming data scenarios, wherein the live streaming matching probability represents the probability of watching the live stream of the target user account through the candidate user account; and determining the user account that matches the live stream of the target user account based on the live streaming matching probability of the candidate user account in each live streaming data scenario.

[0005] Optionally, obtaining multiple user account matching models corresponding to the multiple live streaming data scenarios based on the multiple user account sets may include: performing the following data processing on each user account set: obtaining multiple sets of negative sample data by sampling negative sample data in the user account set; combining positive sample data in the user account set with the multiple sets of negative sample data to obtain multiple sets of training samples; and training a user account matching model for the user account set based on the multiple sets of training samples, wherein the positive sample data represents data in which the user account in the user account set watches the live stream of the target user account or performs an operation while watching the live stream in the corresponding live streaming data scenario, and the negative sample data represents data in which the user account does not watch the live stream of the target user account or does not perform an operation while watching the live stream in the corresponding live streaming data scenario.

[0006] Optionally, training a user account matching model for the user account set based on the multiple sets of training samples may include: setting sample weights for each set of training samples in the multiple sets of training samples; and using the multiple sets of training samples with sample weights to train a user account matching model for the user account set.

[0007] Optionally, for each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

[0008] Optionally, determining the user account that matches the live stream of the target user account based on the live stream matching probability of the candidate user account in each live stream data scenario may include: determining a target live stream data scenario from the plurality of live stream data scenarios; applying a first weight to the live stream matching probability output by the user account matching model corresponding to the target live stream data scenario and applying a second weight to the live stream matching probabilities output by other user account matching models, wherein the first weight is greater than the second weight; determining the integrated live stream matching probability of the candidate user account in the target live stream data scenario based on the weighted live stream matching probability; and determining the user account that matches the live stream of the target user account in the target live stream data scenario based on the integrated live stream matching probability.

[0009] Optionally, determining the user account that matches the live stream of the target user account in the target live stream data scenario based on the integrated live stream matching probability may include: sorting the integrated live stream matching probability of each candidate user account in the candidate user accounts from largest to smallest; selecting the top preset percentage of candidate user accounts as the user accounts that match the live stream of the target user account.

[0010] Optionally, the multiple live streaming data scenarios may include at least one of the following: sending live streaming notifications to users of the target user account; allocating traffic to the target user account for live streaming promotion; daily live streaming data of the target user account; and daily video consumption data of the target user account.

[0011] According to a second aspect of this disclosure, a method for training a user account matching model is provided, which may include: obtaining multiple user account sets for a target user account, wherein the multiple user account sets correspond to multiple live streaming data scenarios of the target user account, and each user account set in the multiple user account sets includes features corresponding to the multiple user accounts; training multiple sub-user account matching models corresponding to the multiple live streaming data scenarios based on the multiple user account sets; and obtaining the user account matching model by assigning weights to the multiple sub-user account matching models, wherein the user account matching model is used to determine user accounts that match the live streams of the target user account.

[0012] Optionally, training multiple sub-user account matching models corresponding to the multiple live streaming data scenarios based on the multiple user account sets may include: performing the following data processing on each user account set in the multiple user account sets: obtaining multiple sets of negative sample data by sampling negative sample data in the user account sets; combining positive sample data in the user account sets with the multiple sets of negative sample data to obtain multiple sets of training samples; and training a sub-user account matching model for the user account sets based on the multiple sets of training samples, wherein the positive sample data represents data in which user accounts in the user account sets watch the live stream of the target user account in the corresponding live streaming data scenario or perform operational actions while watching the live stream, and the negative sample data represents data in which user accounts do not watch the live stream of the target user account in the corresponding live streaming data scenario or do not perform operational actions while watching the live stream.

[0013] Optionally, training a sub-user account matching model for the user account set based on the multiple sets of training samples may include: setting sample weights for each set of training samples in the multiple sets of training samples; and using the multiple sets of training samples with sample weights to train a gradient boosting model for the user account set.

[0014] Optionally, for each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

[0015] Optionally, obtaining the user account model by assigning weights to the plurality of sub-user account matching models may include: determining a target live streaming data scenario from the plurality of live streaming data scenarios; assigning a first weight to the sub-user account matching model corresponding to the target live streaming data scenario, and assigning a second weight to the other sub-user account matching models among the plurality of sub-user account matching models, wherein the first weight is greater than the second weight; and obtaining the user account matching model by combining the plurality of sub-user account matching models with assigned weights.

[0016] According to a third aspect of this disclosure, a user account matching apparatus is provided, which may include: an acquisition module configured to acquire multiple sets of user accounts for a target user account, wherein the multiple sets of user accounts correspond to multiple live streaming data scenarios of the target user account, and each set of user accounts includes features corresponding to the multiple user accounts; a training module configured to obtain multiple user account matching models corresponding to the multiple live streaming data scenarios based on the multiple sets of user accounts; a prediction module configured to perform live streaming matching on candidate user accounts using the multiple user account matching models to obtain the live streaming matching probability of the candidate user account in each of the multiple live streaming data scenarios, wherein the live streaming matching probability represents the probability of watching the live stream of the target user account through the candidate user account; and a determination module configured to determine the user account that matches the live stream of the target user account based on the live streaming matching probability of the candidate user account in each live streaming data scenario.

[0017] Optionally, the training module can be configured to perform the following data processing for each of the plurality of user account sets: obtain multiple sets of negative sample data by sampling negative sample data in the user account set; combine positive sample data in the user account set with the multiple sets of negative sample data to obtain multiple sets of training samples; and train a user account matching model for the user account set based on the multiple sets of training samples, wherein the positive sample data represents data in which the user account in the user account set watches the live stream of the target user account or performs an operation while watching the live stream in the corresponding live stream data scenario, and the negative sample data represents data in which the user account does not watch the live stream of the target user account or does not perform an operation while watching the live stream in the corresponding live stream data scenario.

[0018] Optionally, the training module can be configured to: set sample weights for each of the plurality of training samples; and use the plurality of training samples with sample weights to train a user account matching model for the user account set.

[0019] Optionally, for each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

[0020] Optionally, the determining module may be configured to: determine a target live streaming data scenario from the plurality of live streaming data scenarios; apply a first weight to the live streaming matching probability output by the user account matching model corresponding to the target live streaming data scenario and apply a second weight to the live streaming matching probability output by other user account matching models, wherein the first weight is greater than the second weight; determine the integrated live streaming matching probability of the candidate user account in the target live streaming data scenario based on the weighted live streaming matching probability; and determine the user account that matches the live stream of the target user account in the target live streaming data scenario based on the integrated live streaming matching probability.

[0021] Optionally, the determining module can be configured to: sort the integrated live streaming matching probability of each candidate user account in the candidate user accounts from largest to smallest; and select the top preset percentage of candidate user accounts in the candidate user accounts as user accounts that match the live stream of the target user account.

[0022] Optionally, the multiple live streaming data scenarios may include at least one of the following: sending live streaming notifications to users of the target user account; allocating traffic to the target user account for live streaming promotion; daily live streaming data of the target user account; and daily video consumption data of the target user account.

[0023] According to a fourth aspect of this disclosure, a training apparatus for a user account matching model is provided, which may include: an acquisition module configured to acquire multiple user account sets for a target user account, wherein the multiple user account sets correspond to multiple live streaming data scenarios of the target user account, and each user account set in the multiple user account sets includes features corresponding to the multiple user accounts; and a training module configured to: train multiple sub-user account matching models corresponding to the multiple live streaming data scenarios based on the multiple user account sets; and obtain the user account matching model by assigning weights to the multiple sub-user account matching models, wherein the user account matching model is used to determine user accounts that match the live streams of the target user account.

[0024] Optionally, the training module can be configured to perform the following data processing for each of the plurality of user account sets: obtain multiple sets of negative sample data by sampling the negative sample data in the user account set; combine the positive sample data in the user account set with the multiple sets of negative sample data to obtain multiple sets of training samples; and train a sub-user account matching model for the user account set based on the multiple sets of training samples, wherein the positive sample data represents data in which the user account in the user account set watches the live stream of the target user account or performs an operation while watching the live stream in the corresponding live stream data scenario, and the negative sample data represents data in which the user account does not watch the live stream of the target user account or does not perform an operation while watching the live stream in the corresponding live stream data scenario.

[0025] Optionally, the training module can be configured to: set sample weights for each of the plurality of training samples; and use the plurality of training samples with sample weights to train a sub-user account matching model for the user account set using a gradient boosting model.

[0026] Optionally, for each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

[0027] Optionally, the training module can be configured to: determine a target live streaming data scenario from the plurality of live streaming data scenarios; assign a first weight to the sub-user account matching model corresponding to the target live streaming data scenario, and assign a second weight to other sub-user account matching models among the plurality of sub-user account matching models, wherein the first weight is greater than the second weight; and obtain the user account matching model by combining the plurality of sub-user account matching models with assigned weights.

[0028] According to a fifth aspect of the present disclosure, an electronic device is provided, the electronic device may include: at least one processor; at least one memory storing computer-executable instructions, wherein, when executed by the at least one processor, the computer-executable instructions cause the at least one processor to perform the user account matching method or training method as described above.

[0029] According to a sixth aspect of the present disclosure, a computer-readable storage medium is provided storing instructions that, when executed by at least one processor, cause the at least one processor to perform the user account matching method or training method as described above.

[0030] According to a seventh aspect of the present disclosure, a computer program product is provided, wherein instructions in the computer program product are executed by at least one processor in an electronic device to perform the user account matching method or training method as described above.

[0031] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0032] By comprehensively considering multiple possible factors influencing the live streaming of target user accounts, modeling techniques are used to match user accounts willing to watch the streamer's live stream in different scenarios. This not only improves the accuracy of user account matching but also significantly saves manpower and time costs. Furthermore, by sampling minority class samples (such as negative samples in this invention) and setting sample weights, the trained user account matching model can avoid excessively biasing the majority class samples, thereby overcoming the problem of inaccurate model predictions caused by imbalanced data.

[0033] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0034] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0035] Figure 1This is a flowchart of a user account matching method according to an embodiment of the present disclosure;

[0036] Figure 2 This is a flowchart illustrating a user account matching method according to another embodiment of the present disclosure;

[0037] Figure 3 This is a block diagram of a user account matching device according to an embodiment of the present disclosure;

[0038] Figure 4 This is a schematic diagram of the structure of a user account matching device according to an embodiment of the present disclosure;

[0039] Figure 5 This is a block diagram of an electronic device according to embodiments of the present disclosure;

[0040] Figure 6 This is a flowchart of a training method for a user account matching model according to an embodiment of the present disclosure;

[0041] Figure 7 This is a block diagram of a training apparatus for a user account matching model according to an embodiment of the present disclosure.

[0042] Throughout the accompanying drawings, it should be noted that the same reference numerals are used to denote the same or similar elements, features, and structures. Detailed Implementation

[0043] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0044] The following description, provided with reference to the accompanying drawings, is intended to aid in a full understanding of embodiments of the present disclosure as defined by the claims and their equivalents. Various specific details are included to aid understanding, but these details are to be considered exemplary only. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Furthermore, for clarity and brevity, descriptions of well-known functions and structures are omitted.

[0045] The terms and words used in the following description and claims are not limited to their literal meaning, but are intended solely by the inventors to achieve a clear and consistent understanding of this disclosure. Therefore, it will be apparent to those skilled in the art that the following description of various embodiments of this disclosure is provided for illustrative purposes only and is not intended to limit the purpose of this disclosure as defined by the claims and their equivalents.

[0046] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0047] In related technologies, current user account matching is based on two aspects: firstly, the accounts a user follows; and secondly, several hypothetical factors. After verifying these hypothetical factors, matching user accounts are selected based on the effective factors found. For example, in targeting a streamer's audience, if it is assumed that age, region, and gender affect the streamer's livestream performance, then data analysis is conducted using these three factors to find key effective factors. For instance, it is found that men aged 20 to 30 are more likely to watch, and these factors are used to select the streamer's target audience.

[0048] However, both the focus on accounts and the conjecture of hypothetical factors have two obvious drawbacks. First, conjecture of hypothetical factors are found by humans based on experience, but some hypothetical factors are beyond human detection, and sometimes the hypothetical factors found are invalid. Second, this method has high human and time costs, and the number of hypothetical factors that the human brain can handle is limited, and it is impossible to combine multiple hypothetical factors.

[0049] Based on this, this disclosure employs machine learning modeling to more accurately identify user accounts that match the target user account (such as the streamer's account) in the live stream (i.e., user accounts willing to watch the streamer's live stream). Furthermore, in the actual modeling process, due to the extreme imbalance between the number of user accounts that watched and did not watch the streamer's live stream history, with watching user accounts representing a very small proportion compared to non-watching user accounts, this disclosure utilizes ensemble learning and cost sensitivity to overcome the impact of imbalanced data on user account matching.

[0050] In the following, the methods, apparatus and devices of this disclosure will be described in detail with reference to the accompanying drawings, according to various embodiments of this disclosure.

[0051] Figure 1This is a flowchart of a user account matching method according to an embodiment of the present disclosure. The user account matching method of the present disclosure can be applied to audience targeting for live streamers, audience targeting for advertising, etc. For example, two hours before a streamer goes live, the user account matching method of the present disclosure can be used to predict the streamer's target audience, and then a notification that the streamer will be going live can be sent to the target audience, or a portion of organic traffic can be allocated specifically for promoting the streamer's live stream within the target audience.

[0052] The user account matching method disclosed herein can be executed by any electronic device with data processing capabilities. The electronic device can be at least one of a smartphone, tablet, laptop, or desktop computer. The electronic device can also be a server, etc.

[0053] Reference Figure 1 In step S101, multiple user account sets for the target user account are obtained. The multiple user account sets may correspond to multiple live streaming data scenarios of the target user account, and each user account set may include features corresponding to the multiple user accounts.

[0054] As an example, the target user account could be the account of the streamer who will be conducting the live broadcast. Multiple sets of user accounts related to the streamer's live broadcast can be segmented according to different live broadcast data scenarios. Live broadcast data scenarios could include scenarios such as sending live broadcast notifications to users about the target user account, allocating traffic to the target user account for live broadcast promotion, the target user account's daily live broadcast data, and the target user account's daily video consumption data.

[0055] For example, user account sets can be categorized from historical user accounts related to a streamer's live streams based on scenarios such as sending live stream notifications to user accounts (referred to as push scenarios), allocating traffic to streamer accounts for live stream promotion (referred to as live stream follower scenarios), daily live stream data of streamer accounts, and daily video consumption data of streamer accounts. However, the above examples are merely illustrative, and multiple user account sets can be obtained based on the channels through which streamer live stream information is delivered or other live stream data. User account sets associated with the desired channel or scenario can be obtained based on the channel or scenario for which live stream information is to be delivered.

[0056] The "push" scenario can be understood as sending notifications to user accounts about a streamer's live stream, allowing users to access and watch the streamer's live broadcast. The "live streamer fan" scenario can be understood as allocating a portion of the streamer's organic traffic to promote their live stream, allowing users to access and watch it. Live stream fans can be tools used to promote a streamer's live stream, enabling users to watch it. The streamer's daily live stream data scenario can be understood as the effective playback of user accounts during the longest-running live streams within a specific timeframe. Daily live stream data can also be understood as user actions such as liking or following the streamer during the live stream.

[0057] For example, the user account set in the push scenario (also known as push data) can be the user accounts that pushed the streamer's account in the past seven days; the user account set in the live stream fan data scenario (also known as live stream fan data) can be the user accounts that used live stream fans to promote the streamer's live stream in the past two months; the user account set in the daily live stream data scenario (also known as daily live stream data) can be the user accounts that played the live stream in the two live streams with the longest push duration in the past two months, as well as the user accounts that did not play the live stream; the user account set in the daily video consumption data scenario (also known as daily video consumption data) can include the user accounts that consumed the short videos under the streamer's account in the past three days, as well as the user accounts that did not play them.

[0058] Positive and negative sample data (also known as positive samples) can be determined according to different rules for different live streaming data scenarios. Data representing user accounts in a user account set watching the target user account's live stream or performing actions while watching the live stream in the corresponding live streaming data scenario can be determined as positive samples, while data representing user accounts not watching the target user account's live stream or not performing actions while watching the live stream in the corresponding live streaming data scenario can be determined as negative samples.

[0059] As examples, in push data, the account data of users who clicked the push notification (i.e., users who clicked the notification to enter the streamer's live stream room and watch the streamer) can be used as positive samples, and the account data of users who did not click the push notification can be used as negative samples. In live stream follower data, the account data of users who clicked the live stream follower to enter the live stream room (i.e., users who clicked the promotional tool to enter the streamer's live stream room) can be used as positive samples, and the account data of users who did not click the follower can be used as negative samples. In daily live stream data, the account data of users with more than 0 effective views in the two live streams with the longest push duration in the past two months can be used as positive samples, and the account data of users who did not play can be used as negative samples. In daily video consumption data, the account data of users who generated behaviors such as liking and following can be used as positive samples, and the account data of users who did not generate behaviors such as liking and following can be used as negative samples, or the account data of users who were exposed but did not play can be used as negative samples. However, the above examples are merely illustrative, and this disclosure is not limited thereto.

[0060] After segmenting user account sets under different scenarios, features can be constructed for each user account within each user account set, so that each user account set can include features corresponding to multiple user accounts. Here, the features corresponding to a user account can refer to the static and dynamic attributes of the user corresponding to that user account. The user's static attributes may include the user's age, gender, region, etc. The user's dynamic attributes may include user interest tags, video consumption data, etc. The above example is merely exemplary, and this disclosure is not limited thereto.

[0061] In this disclosure, user account sets across different scenarios can share a set of features. For example, various factors, attributes, and characteristics related to user accounts can be used to construct multi-dimensional features corresponding to each user account. In other words, each user account corresponds to a single data entry, which includes various static and dynamic attributes of the user.

[0062] In step S102, multiple user account matching models corresponding to multiple live streaming data scenarios are obtained based on the acquired user account sets. A user account matching model can be trained using a single user account set. For example, for a user account set corresponding to a push scenario, this user account set can be used to obtain a user account matching model for predicting the people who click on the streamer's push notifications.

[0063] Due to the highly imbalanced nature of live streaming data, for example, in push data, positive samples account for only about 0.3% of the total sample on average, and even as low as 0.08%. This means that out of 1,240,000 data points, only 1,000 are positive samples. To address this extreme data imbalance, the following data processing can be performed on each of the multiple user account sets: Multiple sets of negative sample data are obtained by sampling the negative sample data from that user account set; the positive sample data from that user account set is combined with the multiple sets of negative sample data to obtain multiple sets of training samples; and a user account matching model for that user account set is trained based on these multiple sets of training samples.

[0064] Taking push data as an example, negative samples can be sampled from the push data, and then the positive samples from the push data can be combined with the sampled negative samples to generate multiple sets of training samples. Each user account set of data can be sampled for negative samples in this way to reduce the imbalance between negative and positive samples.

[0065] For a set of user account data, after generating multiple sets of training samples, considering the different cost sensitivities of the samples during model training, sample weights can be set for each training sample in each set of training samples. Then, multiple sets of training samples with sample weights are used to train a user account matching model for that set of user accounts.

[0066] As an example, for a set of training samples, the sample weight of a positive sample in the set is the ratio of the number of positive samples to the number of negative samples, and the sample weight of a negative sample in the set is the ratio of the number of negative samples to the number of positive samples. However, the above example is merely illustrative, and the sample weights can be set differently depending on design requirements.

[0067] After obtaining training samples, a gradient boosting model can be used to train a user account matching model for each set of user accounts. For example, processed training samples can be used to train an XGBoost model to obtain a user account matching model. By setting sample weights, during training, the positive samples can have a greater impact on the model parameters when optimizing the loss function.

[0068] By performing negative sample sampling and setting sample weights, the trained user account matching model can avoid over-biasing the majority class samples, thereby overcoming the problem of inaccurate model predictions caused by imbalanced data.

[0069] In step S103, multiple user account matching models are used to perform live streaming matching on candidate user accounts, obtaining the live streaming matching probability of each candidate user account in each of the multiple live streaming data scenarios. Here, the live streaming matching probability represents the probability of watching the target user account's live stream through the candidate user account. Candidate user accounts can be user accounts that have been active on the live streaming platform in the past month or user accounts from the overall user base, etc.

[0070] As an example, after obtaining the user account matching models for the four live streaming data scenarios mentioned above, each user account matching model can be used to predict the probability of a candidate user account watching a live stream in the corresponding live streaming data scenario. For example, the user account matching model corresponding to the push scenario can be used to predict the probability of a user watching a live stream by clicking on a live stream notification sent to them. The user account matching model corresponding to the live stream fan bar scenario can be used to predict the probability of a user watching a live stream by clicking on a live stream fan bar. The user account matching model corresponding to the daily video consumption data scenario can be used to predict the probability of a user performing an action on a live stream.

[0071] In step S104, the user account that matches the live stream of the target user account is determined based on the live stream matching probability of the candidate user account in each live stream data scenario.

[0072] As an example, a target live streaming data scenario can be determined from multiple live streaming data scenarios. A first weight is applied to the live streaming matching probability output by the user account matching model corresponding to the target live streaming data scenario, and a second weight is applied to the live streaming matching probability output by other user account matching models. Here, the first weight is greater than the second weight, and the second weight may include multiple values. The integrated live streaming matching probability of the candidate user account in the target live streaming data scenario is determined based on the weighted live streaming matching probability. The integrated live streaming matching probability of each candidate user account can be sorted from large to small, and the candidate user accounts in the top preset percentage are selected as the user accounts that match the live stream of the target user account.

[0073] Each user account matching model can have its own weights, which can be assigned differently based on the channel used to deliver live stream messages or the live stream data scenario. The desired live stream data scenario can be predetermined, such as the channel from which the live stream information will be delivered. Then, higher weights can be assigned to the user account matching models corresponding to the determined live stream data scenario. For example, if you want to utilize a push notification channel, you need to assign higher weights to the user account matching models corresponding to the push notification scenario. The weights of each user account matching model can be pre-set, or user accounts can customize the weights for each user account matching model.

[0074] After predicting the live streaming matching probability of each candidate user account in each live streaming data scenario using the matching model for each user account, corresponding model weights can be assigned to the live streaming matching probabilities. Then, the live streaming matching probabilities are weighted and summed to obtain the integrated live streaming matching probability.

[0075] Candidate user accounts that meet preset conditions can be selected from the candidate user accounts based on the integrated live streaming matching probability of each candidate user account. For example, candidate user accounts with an integrated live streaming matching probability reaching a preset threshold can be formed into targeted user accounts that match the target user account. Alternatively, the integrated live streaming matching probabilities of each candidate user account can be sorted from largest to smallest, and then the top 30% of candidate user accounts can be selected as targeted user accounts that match the target user account.

[0076] According to embodiments of this disclosure, by segmenting live streaming data scenarios across multiple different distribution channels, the generation of matching accounts for different scenarios can be more accurately adapted. Furthermore, by considering not only the desired distribution channels or scenarios but also the matching probabilities of user accounts from other distribution channels or scenarios, matching user accounts can be located more efficiently, bringing more viewers to the live streaming room under the broadcaster's account.

[0077] Figure 2 Figure 2 is a flowchart illustrating a user account matching method according to another embodiment of this disclosure. The user account matching method shown in Figure 2 predicts which user accounts to send live broadcast notifications to in a scenario (i.e., a push scenario) where live broadcast notifications for target user accounts (such as broadcaster accounts) are sent to user accounts (such as viewer accounts).

[0078] Reference Figure 2 Based on the historical user accounts associated with the streamer's account, four types of data were divided according to the live streaming data scenario: data for scenarios where live streaming notifications were sent to users (which can be called push data), data for scenarios where traffic was allocated to the streamer's account for live streaming promotion and publicity (which can be called live streaming fan data), daily live streaming data under the streamer's account, and daily video consumption data under the streamer's account.

[0079] For example, push data could be the account data of users who pushed content to the streamer over the past 7 days; live stream fan data could be the order data of live stream fans over the past 2 months; daily live stream data could come from the two live streams with the longest push duration over the past two months; and daily video consumption data could include short video consumption data under the streamer's account over the past 3 days. In push data, the account data of users who clicked on the push notification (i.e., users who entered the streamer's live stream through the sent notification) can be used as positive samples, and the account data of users who did not click on the push notification can be used as negative samples. In live stream fan data, the account data of users who entered the live stream by clicking on the fan link (i.e., users who entered the streamer's live stream through promotional tools) can be used as positive samples, and the account data of users who did not click on the fan link can be used as negative samples. In daily live stream data, the account data of users with more than 0 valid plays can be used as positive samples, and the account data of users who did not play can be used as negative samples. In daily video consumption data, account data of users who generate behaviors such as liking and following can be used as positive samples, while account data of users who do not generate behaviors such as liking and following can be used as negative samples, or account data of users whose videos are exposed but not played can be used as negative samples.

[0080] Features can be constructed for the four types of training data separately. All four types of training data can share the same set of features. For example, this set of features may include static and dynamic attribute data of the user corresponding to the user account. Static attribute data may include the user's age, gender, and region, while dynamic attribute data may include video consumption data and user interest tags. In other words, the training data corresponding to each user account may include multi-dimensional features.

[0081] After feature construction, the training data enters the model training process. Figure 2 The process of negative sample sampling and generating multiple training samples (each training sample includes multiple positive samples and multiple negative samples) is described for the training of a model.

[0082] For example, when training a model, the bagging method is used to sample negative samples to generate multiple training data sets. Then, boosting combined with cost-sensitive methods is used to set sample weights for each training sample, so that the trained model does not overly bias the majority class samples.

[0083] Since there are four different sets of training data, four different user account matching models can be trained. The difference lies in the different positive and negative labels in the original training data, which means that the probability predicted by each user account matching model after training is different.

[0084] Each user account matching model disclosed herein employs a bagging + boosting + cost-sensitive approach to overcome the impact of imbalanced data on model training.

[0085] After training different user account matching models using training samples with sample weights, the weighted ensemble process can be initiated. Figure 2 The weighted ensemble module in the code defines the modules that follow it. In other words, after training each user account matching model, different weights are assigned to each model, and then the assigned model weights are applied to the directional probabilities output by each model.

[0086] Each model has its own weights, and different weights can be assigned to each user account based on the different distribution channels. For example... Figure 2 As shown, when utilizing push channels, the push matching model needs to be given higher weights. For example, a first weight, such as 0.7, can be applied to the push matching model, and a second weight, such as 0.1, can be applied to the live stream fan matching model, the daily live stream matching model, and the daily video consumption matching model. The above weights are merely illustrative, and this disclosure is not limited thereto. The final push probability is obtained by weighted summation of the predicted probabilities of the four user account matching models.

[0087] Candidate user accounts with a push probability reaching a preset threshold can be designated as user accounts targeted by the streamer's account. Alternatively, the push probabilities of each candidate user account can be sorted from highest to lowest, and the top 30% of candidate user accounts can be selected as targeted user accounts, i.e., user accounts matched with the streamer's live broadcast. Push notifications are then used to reach these targeted user accounts, allowing each of these user accounts to enter the streamer's live broadcast room via push notification, thereby attracting more viewers to the live broadcast room.

[0088] This disclosure addresses the issue of extremely imbalanced data by integrating multiple possible factors and using modeling to explore a set of effective feature combinations to find user accounts that match a streamer's live stream. On one hand, it utilizes ensemble learning and cost-sensitive techniques to overcome the impact of imbalanced data on the classification model; on the other hand, it uses the model to predict the matching probability of a streamer's live stream, which is both effective and saves significant manpower and time costs.

[0089] Figure 3 This is a block diagram of a user account matching apparatus according to an embodiment of the present disclosure.

[0090] Reference Figure 3The user account matching device 300 may include an acquisition module 301, a training module 302, a prediction module 303, and a determination module 304. Each module in the user account matching device 300 may be implemented by one or more modules, and the name of the corresponding module may vary depending on the type of module. In various embodiments, some modules in the user account matching device 300 may be omitted, or additional modules may be included. Furthermore, modules / elements according to various embodiments of this disclosure may be combined to form a single entity, and thus perform the functions of the respective modules / elements before combination.

[0091] The acquisition module 301 can acquire multiple user account sets for a target user account. Each user account set corresponds to multiple live streaming data scenarios for the target user account, and each user account set includes features corresponding to the multiple user accounts. The multiple live streaming data scenarios may include at least one of the following: sending live streaming notifications to users for the target user account; allocating traffic to the target user account for live streaming promotion; daily live streaming data scenarios for the target user account; and daily video consumption data scenarios for the target user account.

[0092] Training module 302 can obtain multiple user account matching models corresponding to multiple live streaming data scenarios based on multiple user account sets.

[0093] The prediction module 303 can use multiple user account matching models to perform live streaming matching on candidate user accounts respectively, and obtain the live streaming matching probability of candidate user accounts in each live streaming data scenario in multiple live streaming data scenarios. The live streaming matching probability represents the probability of watching the live stream of the target user account through the candidate user account.

[0094] The determination module 304 can determine the user account that matches the live stream of the target user account based on the live stream matching probability of the candidate user account in each live stream data scenario.

[0095] According to the publicly available examples, the training module 302 can perform the following data processing for each user account set in multiple user account sets: obtain multiple sets of negative sample data by sampling negative sample data in the user account set; combine positive sample data in the user account set with the multiple sets of negative sample data to obtain multiple sets of training samples; and train a user account matching model for the user account set based on the multiple sets of training samples.

[0096] Here, positive sample data represents data where user accounts in the user account set watch the target user account's live stream or perform actions while watching the live stream in the corresponding live stream data scenario, while negative sample data represents data where user accounts do not watch the target user account's live stream or perform no actions while watching the live stream in the corresponding live stream data scenario.

[0097] According to the publicly available example, training module 302 can set sample weights for each of the multiple sets of training samples; and use the multiple sets of training samples with sample weights to train a user account matching model for a set of user accounts.

[0098] For each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

[0099] According to the publicly available examples, the determination module 304 can determine the target live streaming data scenario from multiple live streaming data scenarios; apply a first weight to the live streaming matching probability output by the user account matching model corresponding to the target live streaming data scenario and apply a second weight to the live streaming matching probability output by other user account matching models, wherein the first weight is greater than the second weight; determine the integrated live streaming matching probability of the candidate user account in the target live streaming data scenario based on the weighted live streaming matching probability; and determine the user account that matches the live stream of the target user account in the target live streaming data scenario based on the integrated live streaming matching probability.

[0100] Based on the publicly available example, the determining module 304 can sort the integrated live streaming matching probability of each candidate user account in the candidate user accounts from largest to smallest; and select the top preset percentage of candidate user accounts in the candidate user accounts as the user accounts that match the live stream of the target user account.

[0101] According to another aspect of this disclosure, the user account matching device 300 may include an acquisition module and a determination module. The acquisition module may acquire each user account matching model as described above from an external device, and the determination module may use each acquired user account matching model to predict the live streaming matching probability under different live streaming data scenarios, and then determine the live streaming matching user account of the target user account based on the live streaming matching probability.

[0102] The above has been based on Figure 1 and Figure 2 The user account matching process has been described in detail and will not be repeated here.

[0103] Figure 4 This is a schematic diagram of the structure of a user account matching device in the hardware operating environment of an embodiment of this disclosure.

[0104] like Figure 4As shown, the user account matching device 400 may include: a processing component 401, a communication bus 402, a network interface 403, an input / output interface 404, a memory 405, and a power supply component 406. The communication bus 402 is used to enable communication between these components. The input / output interface 404 may include a video display (such as a liquid crystal display), a microphone and speaker, and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). Optionally, the input / output interface 404 may also include a standard wired interface or a wireless interface. The network interface 403 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 405 may be a high-speed random access memory or a stable non-volatile memory. The memory 405 may also optionally be a storage device independent of the aforementioned processing component 401.

[0105] Those skilled in the art will understand that Figure 4 The structure shown does not constitute a limitation on the user account matching device 400, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0106] like Figure 4 As shown, the memory 405, which serves as a storage medium, may include an operating system (such as a MAC operating system), a data storage module, a network communication module, a user interface module, a program corresponding to the user account matching method of this disclosure, a model training method, and a database.

[0107] exist Figure 4 In the user account matching device 400 shown, the network interface 403 is mainly used for data communication with external electronic devices / terminals; the input / output interface 404 is mainly used for data interaction with user accounts; the processing component 401 and the memory 405 in the user account matching device 400 can be set in the user account matching device 400. The user account matching device 400 calls the crowd targeting program, model training method and various APIs provided by the operating system stored in the memory 405 through the processing component 401 to execute the user account matching method and model training method provided in the embodiments of this disclosure.

[0108] Processing component 401 may include at least one processor, and memory 405 stores a set of computer-executable instructions. When the set of computer-executable instructions is executed by at least one processor, the user account matching method and model training method according to embodiments of the present disclosure are executed. However, the above examples are merely exemplary, and the present disclosure is not limited thereto.

[0109] For example, processing component 401 can determine viewer accounts that match the broadcaster's account based on the user account matching method of this disclosure. Input / output interface 404 can push information about the broadcaster's live stream to each of the matched viewer accounts. Furthermore, processing component 401 can train multiple user account matching models based on the model training method of this disclosure.

[0110] The processing component 401 can control the components included in the user account matching device 400 by executing a program.

[0111] As an example, the user account matching device 400 may be a PC, tablet, personal digital assistant, smartphone, or other device capable of executing the aforementioned set of instructions. Here, the user account matching device 400 is not necessarily a single electronic device, but may be any collection of devices or circuits capable of executing the aforementioned instructions (or instruction sets) individually or in combination. The user account matching device 400 may also be part of an integrated control system or system manager, or may be configured to interface with a portable electronic device locally or remotely (e.g., via wireless transmission).

[0112] In the user account matching device 400, the processing component 401 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller, or a microprocessor. By way of example and not limitation, the processing component 401 may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0113] Processing component 401 can execute instructions or code stored in memory, wherein memory 405 can also store data. Instructions and data can also be sent and received over a network via network interface 403, wherein network interface 403 can employ any known transport protocol.

[0114] The memory 405 can be integrated with the processing component 401, for example, by placing RAM or flash memory within an integrated circuit microprocessor. Alternatively, the memory 405 can include a separate device, such as an external disk drive, a storage array, or other storage device that can be used by any database system. The memory and processing component 401 can be operatively coupled, or can communicate with each other, for example, via I / O ports, network connections, etc., enabling the processing component 401 to read data stored in the memory 405.

[0115] According to embodiments of this disclosure, an electronic device may be provided. Figure 5This is a block diagram of an electronic device according to an embodiment of the present disclosure. The electronic device 500 may include at least one memory 502 and at least one processor 501. The at least one memory 502 stores a set of computer-executable instructions. When the set of computer-executable instructions is executed by the at least one processor 501, a user account matching method and a model training method according to an embodiment of the present disclosure are executed.

[0116] Processor 501 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller, or a microprocessor. By way of example and not limitation, processor 501 may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, etc.

[0117] The memory 502, which serves as a storage medium, may include an operating system (e.g., a MAC operating system), a data storage module, a network communication module, a user interface module, a user account matching program, a model training program, and a database.

[0118] The memory 502 may be integrated with the processor 501; for example, RAM or flash memory may be arranged within an integrated circuit microprocessor. Alternatively, the memory 502 may include a separate device, such as an external disk drive, a storage array, or other storage device that can be used by any database system. The memory 502 and the processor 501 may be operatively coupled, or may communicate with each other, for example, via I / O ports, network connections, etc., enabling the processor 501 to read files stored in the memory 502.

[0119] In addition, electronic device 500 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). All components of electronic device 500 can be interconnected via a bus and / or network.

[0120] As will be understood by those skilled in the art, Figure 5 The structure shown does not constitute a limitation on the structure and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0121] Figure 6 This is a flowchart of a training method for a user account matching model according to an embodiment of the present disclosure. Figure 6 The training method shown can be performed by the above-mentioned device 400, or by the training device described below.

[0122] In step S601, multiple user account sets for the target user account are obtained, wherein the multiple user account sets correspond to multiple live streaming data scenarios of the target user account, and each user account set includes features corresponding to the multiple user accounts.

[0123] In step S602, multiple sub-user account matching models corresponding to multiple live streaming data scenarios are trained based on multiple user account sets.

[0124] As an example, for each user account set in multiple user account sets, multiple sets of negative sample data are obtained by sampling the negative sample data in that user account set; the positive sample data in that user account set is combined with the multiple sets of negative sample data to obtain multiple sets of training samples; sample weights are set for each set of training samples in the multiple sets of training samples; and a gradient boosting model is used to train a sub-user account matching model for that user account set using the multiple sets of training samples with sample weights.

[0125] Here, positive sample data can represent data where a user account in the user account set watches the target user account's live stream or performs an action while watching the live stream in the corresponding live stream data scenario, while negative sample data can represent data where a user account does not watch the target user account's live stream or does not perform an action while watching the live stream in the corresponding live stream data scenario.

[0126] For each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

[0127] In step S603, a final user account matching model is obtained by assigning weights to multiple sub-user account matching models. A weighted ensemble of user account matching models used to predict the target user account for a live-streaming audience can be achieved by assigning weights to multiple sub-user account matching models.

[0128] As an example, a target live streaming data scenario can be determined from multiple live streaming data scenarios. A first weight is assigned to the sub-user account matching model corresponding to the target live streaming data scenario, and a second weight is assigned to the other sub-user account matching models among the multiple sub-user account matching models. The first weight is greater than the second weight. The user account matching model is obtained by combining the multiple sub-user account matching models with assigned weights. Different target live streaming data scenarios can be selected according to actual needs, and the weights of each sub-user account matching model can be set differently.

[0129] Figure 7This is a block diagram of a training apparatus for a user account matching model according to an embodiment of the present disclosure.

[0130] Reference Figure 7 The training device 700 may include an acquisition module 701 and a training module 702. Each module in the training device 700 may be implemented by one or more modules, and the names of the corresponding modules may vary depending on the type of module. In various embodiments, some modules in the training device 700 may be omitted, or additional modules may be included. Furthermore, modules / elements according to various embodiments of this disclosure may be combined to form a single entity, and thus perform the functions of the respective modules / elements before combination.

[0131] The acquisition module 701 can acquire multiple user account sets for the target user account, wherein the multiple user account sets correspond to multiple live streaming data scenarios of the target user account, and each user account set includes features corresponding to the multiple user accounts.

[0132] Training module 702 can train multiple sub-user account matching models corresponding to multiple live streaming data scenarios based on multiple user account sets. Training module 702 can obtain a user account matching model by assigning weights to multiple sub-user account matching models, wherein the user account matching model is used to determine the user account that matches the live stream of the target user account.

[0133] According to the examples of this disclosure, the training module 702 may perform the following data processing for each user account set among multiple user account sets: obtain multiple sets of negative sample data by sampling negative sample data from the user account sets; combine positive sample data from the user account sets with the multiple sets of negative sample data to obtain multiple sets of training samples; and train a sub-user account matching model for the user account sets based on the multiple sets of training samples.

[0134] Here, positive sample data represents data where user accounts in the user account set watch the target user account's live stream or perform actions while watching the live stream in the corresponding live stream data scenario, while negative sample data represents data where user accounts do not watch the target user account's live stream or perform no actions while watching the live stream in the corresponding live stream data scenario.

[0135] According to the examples in this disclosure, training module 702 can set sample weights for each of a plurality of training samples; and use a gradient boosting model to train a sub-user account matching model for a set of user accounts using the plurality of training samples with sample weights.

[0136] For each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

[0137] According to the examples in this disclosure, training module 702 can determine a target live streaming data scenario from multiple live streaming data scenarios.

[0138] Here, users can select the target live streaming data scenario using their account, or the target live streaming data scenario can be preset before training the model.

[0139] The training module 702 can assign a first weight to the sub-user account matching model corresponding to the target live data scenario, and assign a second weight to other sub-user account matching models among multiple sub-user account matching models, wherein the first weight is greater than the second weight; the final user account matching model is obtained by combining multiple sub-user account matching models with assigned weights.

[0140] According to embodiments of this disclosure, a computer-readable storage medium storing instructions may also be provided, wherein when the instructions are executed by at least one processor, they cause at least one processor to execute a user account matching method and model training program according to this disclosure. Examples of computer-readable storage media herein include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disc storage, hard disk drive (HDD), solid-state drive (SSD), card storage (such as multimedia cards, secure digital (SD) cards, or ultra-fast digital (XD) cards), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, and any other device configured to store a computer program and any associated data, data files, and data structures in a non-transitory manner and to provide the computer program and any associated data, data files, and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the aforementioned computer-readable storage medium can run in an environment deployed in computer devices such as clients, hosts, agent devices, servers, etc. Furthermore, in one example, the computer program and any associated data, data files, and data structures are distributed across a networked computer system, such that the computer program and any associated data, data files, and data structures are stored, accessed, and executed in a distributed manner through one or more processors or computers.

[0141] According to embodiments of this disclosure, a computer program product may also be provided, wherein the instructions in the computer program product can be executed by the processor of a computer device to complete the above-described user account matching method and model training method.

[0142] It should be noted that the user account information (including but not limited to user device information, user personal information, etc.) involved in this disclosure are all information authorized by the user or fully authorized by all parties.

[0143] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0144] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A user account matching method, characterized by, The method comprises: obtaining a plurality of user account sets for a target user account, wherein the plurality of user account sets respectively correspond to a plurality of live data scenarios of the target user account, and each user account set in the plurality of user account sets comprises features corresponding to a plurality of user accounts, wherein the plurality of live data scenarios comprise at least one of a scenario of sending live notification of the target user account to a user, a scenario of dividing traffic for live promotion of the target user account, a daily live data scenario of the target user account, and a daily video consumption data scenario of the target user account; obtaining a plurality of user account matching models corresponding to the plurality of live data scenarios based on the plurality of user account sets respectively; performing live matching on a candidate user account by using the plurality of user account matching models respectively to obtain a live matching probability of the candidate user account in each live data scenario of the plurality of live data scenarios, wherein the live matching probability represents a probability of watching a live broadcast of the target user account by the candidate user account; determining a user account matched with a live broadcast of the target user account based on the live matching probability of the candidate user account in each live data scenario, wherein determining a user account matched with a live broadcast of the target user account based on the live matching probability of the candidate user account in each live data scenario comprises: determining a target live data scenario from the plurality of live data scenarios; applying a first weight to a live matching probability output by a user account matching model corresponding to the target live data scenario and applying a second weight to a live matching probability output by other user account matching models, wherein the first weight is greater than the second weight; determining an integrated live matching probability of the candidate user account in the target live data scenario based on the weighted live matching probabilities; determining a user account matched with a live broadcast of the target user account in the target live data scenario based on the integrated live matching probability.

2. The user account matching method of claim 1, wherein, obtaining a plurality of user account matching models corresponding to the plurality of live data scenarios based on the plurality of user account sets respectively comprises: performing the following data processing for each user account set in the plurality of user account sets: obtaining a plurality of groups of negative sample data by sampling negative sample data in a user account set; combining positive sample data in the user account set with the plurality of groups of negative sample data respectively to obtain a plurality of groups of training sample data; training a user account matching model for the user account set based on the plurality of groups of training sample data, wherein the positive sample data represents data of watching a live broadcast of the target user account or having an operation behavior when watching a live broadcast by a user account in the user account set in a corresponding live data scenario, and the negative sample data represents data of not watching a live broadcast of the target user account or not having an operation behavior when watching a live broadcast by the user account in the corresponding live data scenario.

3. The user account matching method of claim 2, wherein, training a user account matching model for the user account set based on the plurality of groups of training sample data comprises: setting a sample weight for each of the plurality of sets of training sample data; training a user account matching model for the set of user accounts by using the plurality of sets of training sample data with the sample weights.

4. The user account matching method of claim 3, wherein, For each of the sets of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

5. The user account matching method of claim 1, wherein, Based on the integrated live broadcast matching probability, determining a user account that matches the target user account in the target live broadcast data scenario, including: sorting the integrated live broadcast matching probabilities of each of the candidate user accounts from large to small; selecting the top pre-set percentage of the candidate user accounts as the user accounts that match the live broadcast of the target user account. 6.A method for training a user account matching model, the method comprising: including: obtaining a plurality of sets of user accounts for a target user account, wherein the plurality of sets of user accounts correspond to a plurality of live broadcast data scenarios of the target user account, and each set of user accounts in the plurality of sets of user accounts includes features corresponding to a plurality of user accounts, wherein the plurality of live broadcast data scenarios include at least one of a scenario of sending a live broadcast notification of the target user account to a user, a scenario of dividing traffic for live broadcast promotion of the target user account, a daily live broadcast data scenario of the target user account, and a daily video consumption data scenario of the target user account; training a plurality of sub-user account matching models corresponding to the plurality of live broadcast data scenarios based on the plurality of sets of user accounts; obtaining the user account matching model by assigning weights to the plurality of sub-user account matching models, wherein the user account matching model is used to determine a user account that matches the live broadcast of the target user account, wherein obtaining the user account matching model by assigning weights to the plurality of sub-user account matching models includes: determining a target live broadcast data scenario from the plurality of live broadcast data scenarios; assigning a first weight to the sub-user account matching model corresponding to the target live broadcast data scenario, and assigning a second weight to the other sub-user account matching models in the plurality of sub-user account matching models, wherein the first weight is greater than the second weight; obtaining the user account matching model by combining the plurality of sub-user account matching models with weights.

7. The training method of claim 6, wherein, Training a plurality of sub-user account matching models corresponding to the plurality of live broadcast data scenarios based on the plurality of sets of user accounts, includes: For each set of user accounts in the plurality of sets of user accounts, the following data processing is performed: obtaining a plurality of sets of negative sample data by sampling negative sample data in the set of user accounts; combining the positive sample data in the set of user accounts with the plurality of sets of negative sample data to obtain a plurality of sets of training sample data; training a sub-user account matching model for the set of user accounts based on the plurality of sets of training sample data, The positive sample data indicates that a user account in the user account set watches a live broadcast of the target user account or has an operation behavior when watching a live broadcast in a corresponding live broadcast data scene, and the negative sample data indicates that the user account does not watch the live broadcast of the target user account or does not have an operation behavior when watching a live broadcast in the corresponding live broadcast data scene.

8. The training method of claim 7, wherein, The sub-user account matching model for the user account set is trained based on the multiple sets of training sample data, including: The sample weight of each set of training sample data is set; The sub-user account matching model for the user account set is trained using a gradient boosting model using the multiple sets of training sample data with sample weights.

9. The training method of claim 8, wherein, For each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

10. A user account matching apparatus, characterized by, It includes: An acquisition module configured to acquire a plurality of user account sets for a target user account, wherein the plurality of user account sets correspond to a plurality of live broadcast data scenes of the target user account, and each user account set in the plurality of user account sets includes features corresponding to a plurality of user accounts, wherein the plurality of live broadcast data scenes include at least one of a scene in which a live broadcast notification of the target user account is sent to a user, a scene in which traffic is divided for the target user account for live broadcast promotion, a daily live broadcast data scene of the target user account, and a daily video consumption data scene of the target user account; A training module configured to obtain a plurality of user account matching models corresponding to the plurality of live broadcast data scenes based on the plurality of user account sets, respectively; A prediction module configured to use the plurality of user account matching models to perform live broadcast matching on a candidate user account to obtain a live broadcast matching probability of the candidate user account in each live broadcast data scene of the plurality of live broadcast data scenes, wherein the live broadcast matching probability represents a probability of watching a live broadcast of the target user account by the candidate user account; A determination module configured to determine a user account matched with a live broadcast of the target user account based on the live broadcast matching probability of the candidate user account in each live broadcast data scene, The determination module is configured to: determine a target live broadcast data scene from the plurality of live broadcast data scenes; apply a first weight to the live broadcast matching probability output by the user account matching model corresponding to the target live broadcast data scene and a second weight to the live broadcast matching probability output by other user account matching models, wherein the first weight is greater than the second weight; determine an integrated live broadcast matching probability of the candidate user account in the target live broadcast data scene based on the weighted live broadcast matching probability; determine, based on the integrated live match probability, a user account that matches the live broadcast of the target user account in the target live data scenario.

11. The user account matching apparatus of claim 10, wherein, The training module is configured to: perform the following data processing for each user account set in the plurality of user account sets: obtain a plurality of sets of negative sample data by sampling negative sample data in a user account set; combine the positive sample data in the user account set with the plurality of sets of negative sample data respectively to obtain a plurality of sets of training samples; train a user account matching model for the user account set based on the plurality of sets of training samples, wherein the positive sample data represents data of watching the live broadcast of the target user account or having an operation behavior while watching the live broadcast by a user account in the user account set in a corresponding live data scenario, and the negative sample data represents data of not watching the live broadcast of the target user account or not having an operation behavior while watching the live broadcast by the user account in the corresponding live data scenario.

12. The user account matching apparatus of claim 11, wherein, The training module is configured to: set a sample weight for each set of training samples in the plurality of sets of training samples; train a user account matching model for the user account set using the plurality of sets of training samples with sample weights.

13. The user account matching apparatus of claim 12, wherein, For each set of training sample data, the sample weight of the positive sample data in the set of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the set of training sample data, and the sample weight of the negative sample data in the set of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the set of training sample data.

14. The user account matching apparatus of claim 10, wherein, The determination module is configured to: sort the integrated live match probabilities of each candidate user account in the candidate user accounts from large to small; select the candidate user accounts in the top preset percentage of the candidate user accounts as the user accounts that match the live broadcast of the target user account.

15. A training apparatus for a user account matching model, characterized in that, comprise: an acquisition module configured to acquire a plurality of user account sets for a target user account, wherein the plurality of user account sets correspond to a plurality of live data scenarios of the target user account respectively, and each user account set in the plurality of user account sets includes features corresponding to a plurality of user accounts, wherein the plurality of live data scenarios include at least one of a scenario of sending a live broadcast notification of the target user account to a user, a scenario of dividing traffic for live broadcast promotion of the target user account, a daily live data scenario of the target user account, and a daily video consumption data scenario of the target user account; a training module configured to: train a plurality of sub-user account matching models corresponding to the plurality of live data scenarios based on the plurality of user account sets respectively; obtain the user account matching model by assigning weights to the plurality of sub-user account matching models, wherein the user account matching model is used to determine a user account that matches the live broadcast of the target user account, wherein the training module is configured to: determine a target live data scenario from the plurality of live data scenarios; assign a first weight to a sub-user account matching model corresponding to the target live data scenario, and assign a second weight to other sub-user account matching models in the plurality of sub-user account matching models, wherein the first weight is greater than the second weight; obtain the user account matching model by combining the plurality of sub-user account matching models to which weights are assigned.

16. The exercise device of claim 15, wherein, The training module is configured to: perform the following data processing for each user account set in the plurality of user account sets: obtain a plurality of groups of negative sample data by sampling negative sample data in a user account set; combine the positive sample data in the user account set with the plurality of groups of negative sample data respectively to obtain a plurality of groups of training samples; train a sub-user account matching model for the user account set based on the plurality of groups of training samples, wherein the positive sample data represents data of watching a live broadcast of the target user account or having an operation behavior when watching a live broadcast by a user account in the user account set in a corresponding live data scenario, and the negative sample data represents data of not watching the live broadcast of the target user account or not having an operation behavior when watching a live broadcast by the user account in the corresponding live data scenario.

17. The exercise device of claim 16, wherein, The training module is configured to: set a sample weight for each group of training samples in the plurality of groups of training samples; train a sub-user account matching model for the user account set using a gradient boosting model using the plurality of groups of training samples with sample weights.

18. The exercise device of claim 17, wherein, For each group of training sample data, the sample weight of the positive sample data in the group of training sample data is the ratio of the number of positive sample data to the number of negative sample data in the group of training sample data, and the sample weight of the negative sample data in the group of training sample data is the ratio of the number of negative sample data to the number of positive sample data in the group of training sample data.

19. An electronic device, comprising: comprise: at least one processor; at least one memory storing computer-executable instructions, wherein the computer-executable instructions, when executed by the at least one processor, cause the at least one processor to perform the user account matching method of any one of claims 1 to 5 or the training method of any one of claims 6 to 9.

20. A computer-readable storage medium, characterized in that, instructions that, when executed by at least one processor, cause the at least one processor to perform the user account matching method of any one of claims 1 to 5 or the training method of any one of claims 6 to 9.

Images