Article inventory scheduling method and device based on private domain channel, equipment and medium

Abstract

Embodiments of the present disclosure disclose a private domain channel-based item inventory scheduling method, device, equipment and medium. A specific implementation of the method includes: obtaining a user private domain channel behavior information set; performing aggregation processing on the user private domain channel behavior information set to obtain a user behavior information set; performing abnormal processing on the user behavior information set to obtain a target user behavior information set; performing user value layering processing on the user identifier information set to obtain a user value level information set; generating a private domain channel content push information set; performing content detection processing on the private domain channel content push information set to obtain a push content detection result set; performing traffic task scheduling processing on the private domain channel content push information set to obtain private domain channel push scheduling information; sending the private domain channel content push information set to each user terminal; and dynamically adjusting the item inventory. The implementation can improve the push accuracy, thereby accurately predicting the change in the number of item inventories, reducing waste of items, and reducing the occurrence of item shortages or accumulation.

Description

Technical Field

[0001] The embodiments of this disclosure relate to the field of computer technology, and more specifically to a method, apparatus, device, and medium for scheduling item inventory based on private domain channels. Background Technology

[0002] Inventory management involves comparing predicted inventory turnover based on user interactions (e.g., purchases, returns, pre-set delayed shipments) with existing inventory. This comparison allows for inventory adjustments such as swapping, replenishment, and price reductions to mitigate shortages or stockpiles. Private domain refers to the digital assets of user (customer) data acquired and accumulated by an enterprise through public domains, other domains, or its own channels. These data can be repeatedly accessed and used for secondary inventory management. However, current private domain traffic push systems suffer from data silos, inefficient content distribution, and data security issues. For inventory management based on private domain channels, the typical approach is to acquire a set of user behavior information from a single private domain channel. Then, using an RFM (Recency, Frequency, Monetary Model) model, the user behavior information set is segmented by user value, resulting in a segmented user value information set. Finally, a large language model is used to generate a hierarchical content push information set tailored to the segmented user value information set and push it to various user terminals. Finally, the inventory warehouse is dynamically adjusted based on the data fed back from each user terminal.

[0003] However, in practice, it has been found that when adjusting the inventory of items based on private domain channels using the above methods, the following technical problems often occur: severe shortages or overstocking of items. The reasons are as follows: obtaining user behavior information from only a single private domain channel makes the data too one-dimensional and incomplete; the RFM model only relies on user consumption information when segmenting user value, without considering other data that affects user value, resulting in low accuracy of user value segmentation; at the same time, the large language model suffers from model illusion, and the content push information generated for users at the same level is too general and does not consider user personalization information, resulting in low quality of content push information and low user push conversion rate, which makes it impossible to predict the quantity of item inventory in a timely and accurate manner, resulting in shortages or overstocking of items.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the present disclosure concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a method, apparatus, device, and medium for scheduling item inventory based on private domain channels to address one or more of the technical problems mentioned in the background section above.

[0007] Firstly, some embodiments of this disclosure provide a method for scheduling item inventory based on private domain channels, including: in response to detecting a shortage or backlog of item inventory, obtaining a user private domain channel behavior information set from multiple private domain channels, wherein the user private domain channel behavior information set includes at least one of the following: data source information, user identification information, user item acquisition information, and user browsing behavior information; aggregating the user private domain channel behavior information set to obtain a user behavior information set; performing anomaly processing on the user behavior information set to obtain a target user behavior information set; performing user value stratification processing on the user identification information set based on the target user behavior information set to obtain a user value hierarchy information set; and performing user value stratification processing on the user identification information set based on private domain channel feature information and... The aforementioned user value hierarchy information set generates a private domain channel content push information set; the aforementioned private domain channel content push information set undergoes content detection processing to obtain a push content detection result set; based on the aforementioned push content detection result set and the aforementioned private domain channel feature information, the aforementioned private domain channel content push information set undergoes traffic task scheduling processing to obtain private domain channel push scheduling information; based on the aforementioned private domain channel push scheduling information, the aforementioned private domain channel content push information set is pushed and transmitted to the user terminal set corresponding to the aforementioned user identification information set; based on the obtained item flow information set corresponding to the aforementioned user terminal set, the item inventory is dynamically adjusted, wherein the aforementioned dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed item replenishment.

[0008] Secondly, some embodiments of this disclosure provide an item inventory scheduling device based on private domain channels, comprising: an acquisition unit configured to, in response to detecting a shortage or overstock of item inventory, acquire a user private domain channel behavior information set from multiple private domain channels, wherein the user private domain channel behavior information set includes at least one of the following: data source information, user identification information, user item acquisition information, and user browsing behavior information; an aggregation unit configured to perform aggregation processing on the user private domain channel behavior information set to obtain a user behavior information set; an anomaly handling unit configured to perform anomaly processing on the user behavior information set to obtain a target user behavior information set; a user value stratification unit configured to perform user value stratification processing on the user identification information set according to the target user behavior information set to obtain a user value stratification information set; and a generation unit configured to, based on private domain channel characteristics... The system generates a private domain channel content push information set based on the information and the aforementioned user value hierarchy information set. A content detection unit is configured to perform content detection processing on the aforementioned private domain channel content push information set to obtain a push content detection result set. A traffic task scheduling unit is configured to perform traffic task scheduling processing on the aforementioned private domain channel content push information set based on the aforementioned push content detection result set and the aforementioned private domain channel feature information to obtain private domain channel push scheduling information. A push transmission unit is configured to push and transmit the aforementioned private domain channel content push information set to the user terminal set corresponding to the aforementioned user identification information set based on the aforementioned private domain channel push scheduling information. An adjustment unit is configured to dynamically adjust the item inventory based on the obtained item flow information set corresponding to the aforementioned user terminal set, wherein the aforementioned dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed item replenishment.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, such that when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method as described in any implementation of the first aspect.

[0011] The above embodiments of this disclosure have the following beneficial effects: The item inventory scheduling method based on private domain channels in some embodiments of this disclosure can improve the accuracy of push notifications, thereby accurately predicting changes in item inventory quantities, reducing item waste, and reducing the occurrence of item shortages or stockpiles. Specifically, the reasons for the inability to predict item inventory quantities in a timely and accurate manner, resulting in item shortages or stockpiles, are as follows: The data is too one-dimensional and incomplete because only user behavior information from a single private domain channel is obtained; the RFM model only relies on user consumption information when performing user value segmentation, without considering other data affecting user value information, resulting in low accuracy in user value segmentation; at the same time, the large language model suffers from model illusion problems, and the content push information generated for users at the same level is too general and does not consider user personalization information, resulting in low quality of content push information and low user push conversion rates, leading to the inability to predict item inventory quantities in a timely and accurate manner, resulting in item shortages or stockpiles. Based on this, some embodiments of the item inventory scheduling method based on private domain channels disclosed herein can first, in response to detecting a shortage or backlog of item inventory, obtain a set of user private domain channel behavior information from multiple private domain channels. This set of user private domain channel behavior information includes at least one of the following: data source information, user identification information, user item acquisition information, and user browsing behavior information. Obtaining user private domain channel behavior information from multiple private domain channels improves data comprehensiveness, avoids data silos, and facilitates subsequent user value stratification processing. Secondly, the user private domain channel behavior information set is aggregated to obtain a user behavior information set. This improves the quality of user behavior information and reduces the occurrence and volume of conflicting data. Thirdly, the user behavior information set undergoes anomaly processing to obtain a target user behavior information set. Anomaly processing can automatically identify malicious behaviors such as fraudulent transactions, facilitating improved accuracy in subsequent user value stratification. Next, based on the target user behavior information set, the user identification information set is processed for user value stratification to obtain a user value hierarchy information set. Here, user value stratification effectively classifies users into different levels, facilitating the generation of personalized private channel content push notifications. Subsequently, based on private channel feature information and the aforementioned user value stratification information set, a private channel content push notification set is generated. This personalization ensures the private channel content push notification set meets user preferences and private channel needs, enabling subsequent content detection. Afterward, content detection processing is performed on the aforementioned private channel content push notification set to obtain a push content detection result set. This content detection process automatically identifies content in the private channel content push notifications that does not conform to the private channel's rules, improving the quality of the private channel content push notifications and the success rate of subsequent pushes.Then, based on the aforementioned push content detection result set and the aforementioned private domain channel feature information, traffic task scheduling processing is performed on the aforementioned private domain channel content push information set to obtain private domain channel push scheduling information. Here, traffic task scheduling processing can reasonably follow the inference timing and quantity of the private domain channel content push information set, increasing the accuracy and precision of scheduling while meeting the requirements of the private domain channel. Next, based on the aforementioned private domain channel push scheduling information, the aforementioned private domain channel content push information set is pushed and transmitted to the user terminal set corresponding to the aforementioned user identification information set. Here, it is pushed to each user terminal for subsequent observation of user feedback information and conversion rates. Finally, based on the obtained item flow information set corresponding to the aforementioned user terminal set, the item inventory is dynamically adjusted, whereby the aforementioned dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed item replenishment. Here, by analyzing the user private domain channel behavior information set after the push, the user conversion rate can be obtained, which can improve the accuracy of predicting item inventory changes, reduce stockouts or backlogs, increase inventory turnover, and reduce item waste. Therefore, this private domain channel-based item inventory scheduling method can improve push accuracy, thereby accurately predicting changes in item inventory quantity, reducing item waste, and minimizing item shortages or stockpiles. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 This is a flowchart of some embodiments of the item inventory scheduling method based on private domain channels according to this disclosure; Figure 2 This is a schematic diagram of the structure of some embodiments of the item inventory scheduling device based on private domain channels according to the present disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Figure 1 A flow 100 is shown illustrating some embodiments of a private domain channel-based item inventory scheduling method according to this disclosure. This private domain channel-based item inventory scheduling method includes the following steps: Step 101: In response to the detection of a shortage or backlog of item inventory, obtain a set of user private domain channel behavior information from multiple private domain channels.

[0021] In some embodiments, the executing entity (e.g., an electronic device) of the above-described private domain channel-based item inventory scheduling method can, in response to detecting an item inventory shortage or backlog, obtain a set of user private domain channel behavior information from multiple private domain channels via wired or wireless connection. This set of user private domain channel behavior information includes at least one of the following: data source information, user identification information, user item acquisition information, and user browsing behavior information. The detection of an item inventory shortage or backlog can be detected when the existing inventory exceeds a preset safe storage limit, indicating a backlog, or when the existing inventory is below a preset safe storage lower limit, indicating a shortage. The preset safe storage limit and preset safe storage lower limit can be upper and lower limits of the safe storage range for items, predetermined through expert experience. The private domain channels mentioned above can be private domain channels owned by the enterprise that can repeatedly reach users free of charge or at low cost. For example, the multiple private domain channels may include, but are not limited to, at least one of the following: enterprise-owned platform channels, enterprise offline touchpoint channels, and mini-program / official account channels. The aforementioned user private channel behavior information set can be user interaction information through various private channels and items. These interactions may include, but are not limited to, at least one of the following: clicking, purchasing, forwarding, browsing, and saving. The aforementioned data source information may be the channel identifier information of the private channel in the user's private channel behavior information. The aforementioned user identification information may be information that uniquely identifies the user. The aforementioned user item acquisition information may include information such as the purchase time, value, and method of purchasing the items purchased by the user.

[0022] In some optional implementations of certain embodiments, the aforementioned user private channel behavior information set may be obtained through the following steps: The first step is to perform the following transmission steps for each user's private channel behavior information in the user private channel behavior information set: Sub-step 1 involves performing local data preprocessing on the aforementioned user private channel behavior information to obtain preprocessed channel behavior information. This local data preprocessing can be performed on the edge server where the private channel resides, which can reduce the risk of data leakage.

[0023] Sub-step 2 involves determining the receiving end address information of the data receiving end corresponding to the aforementioned preprocessed channel behavior information. This receiving end address information can be the domain name or IP address of the server used to receive the user's private channel behavior information, using SSL (Secure Sockets Layer).

[0024] Sub-step 3: In response to confirming a successful communication connection with the data receiving end, an end-to-end covert handshake message is sent to the data receiving end using the receiving end's address information. This end-to-end covert handshake message includes: the size of the data transmission byte stream, a data integrity check value, covert channel version information, and covert channel type information. Specifically, the end-to-end covert handshake message can be used to inform the data receiving end of the identity information of the private domain channel end sending preprocessed channel behavior information and the covert channel information. The private domain channel end can be an edge server end used to send preprocessed channel behavior information. The size of the data transmission byte stream can be the size of the transmitted byte stream included in the handshake message composed of user private domain channel behavior information. The file integrity check value can be a value used to verify whether the preprocessed channel behavior information before and after transmission is consistent. The covert channel type information can be information about the type of covert channel constructed between the private domain channel end and the data receiving end. This covert channel type information can include: storage-type covert channel, sequence-type covert channel, time-type covert channel, and packet-length covert channel. The aforementioned sequence-based covert channel can be a covert channel that transmits ciphertext by using the sorting order of different cipher suites in the cipher suite list field of the Client Hello message used to establish the initial handshake connection with the data receiver. The aforementioned packet-length-based covert channel can be a covert channel that transmits ciphertext using the length information of the Client Hello message.

[0025] Sub-step 4: In response to receiving the response message of the end-to-end covert handshake message information sent by the data receiving end, and the covert channel type information is a stored covert channel, a first covert channel symmetric key is generated. This first covert channel symmetric key can be a symmetric key used for encrypting and decrypting file transmission byte streams. For example, the first covert channel symmetric key can be a 3DES (Triple Data Encryption Algorithm) key.

[0026] Sub-step 5: Based on the first covert channel symmetric key mentioned above, the preprocessed channel behavior information is cyclically encrypted to obtain an encrypted user behavior byte stream set.

[0027] Sub-step 6 involves embedding the aforementioned encrypted user behavior byte stream set and synchronization transmission identifier information into the target field set of the end-to-end covert handshake message information to obtain the handshake embedding protocol message. The synchronization transmission identifier information can be information used to prevent out-of-order execution of the encrypted user behavior byte stream within the encrypted user behavior byte stream set. The target field set can be the Random field and Session ID field from the Client Hello message. The synchronization transmission identifier information is embedded in the Session ID field. It should be noted that since the Random field and Session ID field in the Client Hello message originally store random fields generated by sending a set of files to be transmitted, modifying them will not affect message recognition and can, to some extent, prevent them from being identified as abnormal messages.

[0028] Sub-step 7 involves performing traffic anomaly detection on the aforementioned handshake embedded protocol message information to obtain traffic anomaly detection result information. This result information represents a comparison between the probability value of the embedded handshake protocol message being identified as an anomalous message and a preset anomaly threshold. The preset anomaly threshold can be a pre-defined minimum value for being identified as an anomalous message. The value of the preset anomaly threshold can be determined based on specific circumstances and is not limited here. In practice, the executing entity can first input the aforementioned handshake embedded protocol message into a denoising autoencoder to obtain a protocol feature vector set. Then, the protocol feature vector set is input into a message traffic anomaly detection model optimized based on the "moth to a flame" optimization algorithm to obtain a traffic anomaly value. This model can be a model that adds a gated recurrent unit network to a convolutional neural network. Finally, the traffic anomaly value is compared with the preset anomaly threshold to obtain the traffic anomaly detection result information.

[0029] Sub-step 8: In response to determining that the traffic anomaly detection result information indicates no anomaly, the number of times the handshake embedded protocol message information is sent is determined based on the encrypted user behavior byte stream set. Here, "no anomaly" can mean that the traffic anomaly detection value is less than or equal to a preset anomaly value threshold. In practice, the executing entity can determine the number of bytes corresponding to the encrypted user behavior byte stream set divided by the number of bytes transmitted in the handshake embedded protocol message as the number of data transmissions.

[0030] Sub-step 9 involves dividing the aforementioned handshake embedded protocol message information into several data transmission messages and transmitting them separately for the data receiving end to decrypt and receive, thereby obtaining the user's private channel behavior information.

[0031] In some optional implementations of certain embodiments, the above-mentioned byte stream cyclic encryption of the preprocessed channel behavior information based on the first covert channel symmetric key to obtain an encrypted user behavior byte stream set may include the following steps: The first step, based on the first covert channel symmetric key and preprocessed channel behavior information, is to perform the following encryption steps: Sub-step 1 involves encrypting the transmission byte stream at the beginning position of the preprocessed channel behavior information using the first covert channel symmetric key, to obtain a first encrypted user behavior byte stream. This first encrypted user behavior byte stream can be a byte stream obtained by encrypting the transmission byte stream at the beginning position using a symmetric encryption algorithm.

[0032] Sub-step 2 involves symmetrically encrypting the target transmission byte stream based on the first encrypted user behavior byte stream to obtain a second encrypted user behavior byte stream. The target transmission byte stream is the byte stream located after the start position in the preprocessed channel behavior information. In practice, the executing entity can use a symmetric encryption algorithm, employing the first encrypted user behavior byte stream as the symmetric encryption key, to symmetrically encrypt the target transmission byte stream and obtain the second encrypted user behavior byte stream.

[0033] Sub-step 3: In response to determining that the target transmission byte stream is a transmission byte stream located at the termination position of the preprocessed channel behavior information, an encrypted user behavior byte stream set is determined based on the first encrypted user behavior byte stream and the second encrypted user behavior byte stream. The aforementioned encrypted user behavior byte stream can be a byte stream obtained by encrypting the user private domain channel behavior information located before the termination position; that is, the encrypted user behavior byte stream during the third encryption is composed of the first encrypted user behavior byte stream and the second encrypted user behavior byte stream. The encrypted user behavior byte stream during the fourth encryption is composed of the first encrypted user behavior byte stream, the second encrypted user behavior byte stream, and the third encrypted user behavior byte stream, and so on, to obtain the encrypted user behavior byte stream set.

[0034] The second step involves determining, in response to the determination that the target transmitted byte stream is not a transmitted byte stream located at the termination position, that the accumulated second encrypted user behavior byte stream is identified as the first covert channel symmetric key, and that the transmitted byte streams corresponding to the accumulated first and second encrypted user behavior byte streams are removed, thus identifying the remaining transmitted byte stream set, to execute the aforementioned encryption steps again. The accumulated second encrypted user behavior byte stream can be any of all encrypted user behavior byte streams preceding the current target transmitted byte stream. The accumulated first and second encrypted user behavior byte streams can be the cumulative encrypted user behavior byte streams removed from the preprocessed channel behavior information.

[0035] Step 102: Aggregate the user's private channel behavior information set to obtain the user behavior information set.

[0036] In some embodiments, the aforementioned execution entity may aggregate the aforementioned user private channel behavior information set to obtain a user behavior information set. The user behavior information in the aforementioned user behavior information set may be information from multiple user private channel behavior sets that integrate the same user identifier information from different private channels, and remove conflicting data.

[0037] In some optional implementations of certain embodiments, the above-mentioned aggregation processing of the user private channel behavior information set to obtain the user behavior information set may include the following steps: The first step is to construct an undirected graph of user channel identifiers based on the aforementioned user private channel behavior information set. This undirected graph can be composed of nodes representing user identifiers (e.g., phone number, WeChat ID, device ID, etc.) included in the user private channel behavior information set, and connecting edges representing different channel identifiers for the same user. In practice, the executing entity can first identify the user channel identifier information set and the channel identifier association information set representing the same user on different private channels within the aforementioned user private channel behavior information set, using the identical identifier information (e.g., phone number, ID card number) across different private channels. Then, the user channel identifier information set and the channel identifier association information set are input into a graph database to obtain the undirected graph of user channel identifiers.

[0038] The second step involves performing connected component partitioning on the aforementioned undirected graph of user channel identifiers to obtain a user identifier connected component graph set. This set can be a subgraph composed of multiple user channel identifiers and associated information, consisting of identical identifier information across different private channels. The connected component partitioning can be performed using a breadth-first search algorithm.

[0039] The third step involves performing user identifier matching processing on the aforementioned user identifier connected component graph atlas and the aforementioned user private channel behavior information set to obtain the target user identifier matching information set. The target user identifier matching information in this set can be information that assigns unique global user identifiers to the aforementioned user identifier connected component graph, including setting each user node in the user identifier connected component graph to different private channel user identifiers under the global user identifier information. For example, the WeChat ID included in the user identifier connected component graph can be set as the global user identifier information _WeChat ID. In practice, the executing entity can first set global identifier information on the aforementioned user identifier connected component graph atlas to obtain a global user identifier information set. Then, using the aforementioned global user identifier information set as an index, user identifiers are configured on the aforementioned user private channel behavior information set to obtain the target user identifier matching information set.

[0040] The fourth step involves extracting information from the user's private channel behavior information set based on the target user identifier matching information set, resulting in a user attribute information set and a user behavior information set. The user attribute information in the user attribute information set can be basic information representing the user's identity. This user attribute information may include, but is not limited to, at least one of the following: user age, user gender, user login device information, and user channel profile information. The user behavior information in the user behavior information set can be information about the user's content interactions on different private channels. For example, this user behavior information may include, but is not limited to, at least one of the following: push content browsing time, likes, favorites, purchases, dwell time, and sharing.

[0041] The fifth step involves disambiguating the aforementioned user attribute information set through voting conflicts, resulting in a disambiguated user attribute information set. In practice, the executing entity can first perform attribute priority rule conflict disambiguation on the aforementioned user attribute information set, obtaining a priority-disambiguated user attribute information set. This attribute priority rule conflict disambiguation can include, but is not limited to, at least one of the following: priority of channel attribute source credibility rules (where attribute information from private channel real-name authentication is greater than user-initiated attribute information, which is greater than inferred attribute information); priority of source frequency retention rules (where attribute information with the same attribute across different private channels is retained); and priority of attribute rules (where new attributes override old attributes). Then, through a voting mechanism, the priority-disambiguated user attribute information set undergoes channel voting processing based on discrete value statistics to obtain the disambiguated user attribute information set.

[0042] Step 6: Perform indicator conflict disambiguation processing on the aforementioned user behavior information set to obtain a disambiguated user behavior information set. This disambiguated user behavior information set can be obtained by performing behavior merging and time series alignment disambiguation processing on the aforementioned user behavior information set. In practice, the executing entity can use a rule engine to perform indicator conflict disambiguation processing on the aforementioned user behavior information set to obtain a disambiguated user behavior information set. The rules included in the tree-based rule engine may include, but are not limited to, at least one of the following: hash deduplication rules based on user behavior identifier information (e.g., browsing, purchasing, etc.) and time windows; rules to eliminate conflicts based on the order of user behavior timestamps; and rules to disambiguate based on the trustworthiness of private channels if the times are the same.

[0043] Step 7: Based on the target user identifier matching information set mentioned above, aggregate the disambiguated user attribute information set and the disambiguated user behavior information set to obtain the user behavior information set. In practice, the executing entity can match the disambiguated user attribute information set, the disambiguated user behavior information set, and the target user identifier matching information set to obtain the user behavior information set.

[0044] Step 103: Perform anomaly processing on the user behavior information set to obtain the target user behavior information set.

[0045] In some embodiments, the executing entity can perform anomaly processing on the user behavior information set to obtain a target user behavior information set. The target user behavior information in the target user behavior information set may be user behavior information without any anomalies. In practice, the executing entity can use a Naive Bayes algorithm to perform anomaly processing on the user behavior information set to obtain a user anomaly behavior information set. This anomaly processing can be done on a user-by-user basis, identifying anomalies present in each user at different times. Then, the user behavior information set after removing the user anomaly behavior information set is determined as the target user behavior information set.

[0046] Step 104: Based on the target user behavior information set, perform user value stratification processing on the user identification information set to obtain the user value hierarchy information set.

[0047] In some embodiments, the executing entity may perform user value stratification processing on the user identification information set based on the target user behavior information set to obtain a user value hierarchy information set. The user value hierarchy information in the user value hierarchy information set may be information representing the user's consumption level. The user value hierarchy information may include: high-net-worth user hierarchy information, advanced user hierarchy information, and ordinary user hierarchy information.

[0048] In some optional implementations of certain embodiments, the process of performing user value stratification on the user identification information set based on the target user behavior information set to obtain a user value hierarchy information set may include the following steps: The first step involves extracting heterogeneous nodes and user behavior information from the aforementioned target user behavior information set, resulting in a heterogeneous node set and a user behavior information set. The heterogeneous node set includes at least one of the following: a target user node set, an item node set, a private domain content node set, and a private domain channel push node set. Heterogeneous nodes in the heterogeneous node set can represent entities and objects within the target user behavior information set. Private domain content nodes can be content sent to different private domain channels (e.g., WeChat Moments, Xiaohongshu, Douyin, etc.) in one or more formats such as images, audio / video, and text to push items and / or virtual items. Item nodes in the item node set can include real items or virtual items (e.g., promotional information, coupons, discount information). Private domain channel push nodes can be nodes composed of platform information that pushes private domain content. User behavior in the user behavior information set can be behavioral information extracted from target user behavior information, after removing abnormal information and data conflicts.

[0049] The second step involves feature extraction from the node attribute information set corresponding to the aforementioned heterogeneous node set, resulting in a node feature information set. This node feature information set can include information describing the basic information, characteristics, and status of user nodes; category information, value information, and semantic information of private domain content nodes; and traffic distribution information and content review information of private domain channel push nodes. For example, the node feature information may include, but is not limited to, at least one of the following: basic user information and user statistical feature information (e.g., total historical spending, number of active days, and highest single spending amount). In practice, the executing entity can first perform word embedding processing on the basic user information and statistical feature information of the user nodes included in the aforementioned heterogeneous node set, obtaining a basic word embedding vector set and a statistical word embedding vector set. Secondly, the basic word embedding vector set and the statistical word embedding vector set are input into a multilayer perceptron for mapping and then concatenated to obtain the user node feature information set, which serves as the node feature information set. Then, using the CLIP (Contrastive Language-Image Pre-training) model, multimodal feature extraction is performed on the item information and item push content information included in the node attribute information set corresponding to the aforementioned private channel push node set, resulting in a content node feature vector set, which serves as the node feature information set. Finally, using the BERT (Bidirectional Encoder Representations from Transformers) model, feature extraction is performed on the traffic distribution information and content review information corresponding to the aforementioned private channel push node set, resulting in a push node feature vector set, which also serves as the node feature information set.

[0050] The third step is to determine the edge weights of the aforementioned user behavior information set, which will be used as the behavior edge weight set. The behavior edge weights in this set characterize the degree of association, behavior intensity, and time decay between two connected nodes. In practice, the executing entity can perform the weight determination step for any two nodes connected in the heterogeneous node set: First, using a behavior type weight table, determine the behavior intensity information of the behavior type of the user behavior information set between any two connected nodes in the heterogeneous node set. This behavior type weight table can be a form used to record the weights assigned to the type of user behavior. For example, if the behavior type is a purchase behavior and the number of purchases is a factor, the behavior intensity can be the sum of the basic weight of the purchase behavior and ln(purchase count + 1), then input into a Sigmoid function to obtain the value; alternatively, for browsing behavior, the intensity can be a value normalized to the browsing duration. Then, using a behavior time decay function, determine the time decay of the user behavior information between any two nodes. The aforementioned behavior time decay function can be an exponential function with base e, using the negative of the product of the difference between the current timestamp and the event timestamp of the user behavior information, and the time decay factor as the exponent. The aforementioned time decay factor can be any value within the range (0, 1), used to determine the decay rate; a larger value indicates faster decay. Finally, the aforementioned time decay rate and the aforementioned behavior intensity information are weighted and summed to obtain the behavior edge weight.

[0051] The fourth step involves inputting the aforementioned heterogeneous node set, user behavior information set, node feature information set, and behavior edge weight set into a graph database to obtain a heterogeneous user behavior graph. Specifically, the heterogeneous user behavior graph can be a graph where the heterogeneous node set is the node set, the user behavior information set is the connecting edge, the node feature information set is the node attribute information of the heterogeneous node set, and the behavior edge weight set is the weight of the connecting edge.

[0052] The fifth step involves constructing meta-paths on the aforementioned heterogeneous user behavior graph to obtain a heterogeneous user behavior path graph. This heterogeneous user behavior path graph can be a graph with preset semantic meta-paths added to it. These preset semantic meta-paths can be predefined paths representing different perspectives, composed of different node types and connection edge types. These preset semantic meta-paths may include, but are not limited to, at least one of the following: meta-paths connecting user node purchase behavior to item nodes and user nodes that purchase the same item; meta-paths connecting user node browsing behavior to private domain content nodes, including meta-paths connecting user node browsing behavior to private domain content nodes and related products after browsing private domain content; and meta-paths connecting user node favorites behavior to private domain content nodes and related content pushed by the favorites platform belonging to the associated private domain channel push node.

[0053] Step 6: Extract behavioral interaction information from the aforementioned target user behavior information set to obtain a behavioral interaction time sequence information set. This behavioral interaction time sequence information is a time sequence triple consisting of user behavior, item information / private domain content, and interaction time. Specifically, the behavioral interaction time sequence information in this set can be obtained by sorting the target user behavior information of a user and the item / private domain content in chronological order according to timestamps within a preset time window. For example, the user behavior interaction time sequence information could be [(shoes pushing private domain content, browsing, interaction time 1), (shoes pushing private domain content, favorite, interaction time 2), (shoes, click, interaction time 3), (shoes, purchase, interaction time 4)].

[0054] The seventh step involves encoding the sequence features of the aforementioned behavioral interaction time-series information set to obtain a behavioral interaction time-series feature information set. The behavioral interaction time-series feature information in this set can be sequence information representing the user's dynamic behavioral preferences. In practice, the executing entity can utilize a Behavior Sequence Transformer (BST) model to encode the sequence features of the aforementioned behavioral interaction time-series information set to obtain the behavioral interaction time-series feature information set.

[0055] Step 8 involves fusing and comparing the aforementioned behavioral interaction temporal feature information set and the aforementioned heterogeneous user behavior path graph to obtain a fused interaction feature information set. The fused interaction feature information in this set can be information that integrates global collaborative information and temporal dynamic behavior information from various heterogeneous nodes across multiple types of behaviors. In practice, the executing entity can first input the aforementioned heterogeneous user behavior path graph into an MB-HGCN (Multi-Behavior Hierarchical Graph Convolutional Network) model for multi-behavior embedding extraction to obtain a multi-behavior global feature information set. Then, through comparative learning, a cross-view comparative learning process is performed on the aforementioned behavioral interaction temporal feature information set and the aforementioned multi-behavior global feature information set to obtain the fused interaction feature information set. Positive samples for the heterogeneous user behavior path graph in the comparative learning can be the behavioral interaction temporal feature information of the same user and the graph feature information obtained from the heterogeneous user behavior path graph using HGNN (Hierarchical Graph Neural Networks). In the above contrastive learning, negative samples for the heterogeneous graph of user behavior paths can be temporal feature information and graph feature information of behavioral interactions from different users. Positive samples for the temporal information set of behavioral interactions can be feature representations of the temporal information set of behavioral interactions from different behaviors of the same user. The loss function for the above contrastive learning can be the contrastive cross-entropy loss function.

[0056] Step nine involves classifying user value based on the aforementioned fused interaction feature information set, specifically the heterogeneous user behavior path graph and the time series graph corresponding to the aforementioned behavioral interaction time series information set, to obtain a user value hierarchy information set. In practice, the aforementioned executing entity can use the LTV (Customer Lifetime Value) model to classify user value based on the aforementioned fused interaction feature information set, specifically the heterogeneous user behavior path graph and the time series graph corresponding to the aforementioned behavioral interaction time series information set, to obtain a user value hierarchy information set. The aforementioned LTV model can be a CMLTV (Contrastive Multi-view Framework for Customer Lifetime Value Prediction) that predicts user lifetime value from the input fused interaction feature information set and outputs the user's value level information.

[0057] In some optional implementations of certain embodiments, the process of dividing the heterogeneous user behavior path graph into user value hierarchical information sets based on the aforementioned fused interaction feature information set may include the following steps: The first step involves inputting the aforementioned fused interaction feature information set into a user value probability prediction network to obtain a user lifetime value prediction set and a value grading probability set. The user lifetime value predictions can be the grading information predicted after user value stratification. The value grading probabilities in the value grading probability set can be the probability values ​​of a user being classified into the user lifetime value prediction values. The user value probability prediction network can be a model that predicts user value from the input fused interaction feature information set and outputs user lifetime value predictions and value grading probabilities. For example, the user value probability prediction network could be an LTV model based on ZILN (Zero-Inflated LogNormal).

[0058] The second step involves reconstructing the user graph from the aforementioned heterogeneous user behavior path graph and the time series graph corresponding to the aforementioned behavioral interaction time series information set, based on the fused interaction feature information set. This results in a user isomorphic weighted graph, where the connection edges in the user isomorphic weighted graph include at least one of the following types: friend relationship, item purchase relationship, or shared interest relationship. The user isomorphic weighted graph can be a graph containing only user nodes, where the similarity between adjacent user nodes is greater than or equal to a preset similarity threshold. This preset similarity threshold can be pre-set and determined as a critical value to be retained in the user isomorphic weighted graph. The value of the preset similarity threshold can be determined based on specific circumstances and will not be elaborated further here. In practice, the executing entity can first determine the cosine similarity of the fused interaction feature information sets of any two user nodes included in the set of connected user nodes, obtaining a cosine similarity set. Then, at least one cosine similarity greater than or equal to the preset similarity threshold is selected from the cosine similarity set. Subsequently, the connection edge set and user node set corresponding to at least one cosine similarity are extracted from the heterogeneous graph and time sequence graph of the above user behavior path to obtain the user isomorphic weighted graph.

[0059] The third step involves performing temporal community graph partitioning on the aforementioned isomorphic weighted graph to obtain user community clusters. These user community clusters can be obtained through clustering based on user behavior similarity. In practice, the executing entity can first perform graph Laplacian matrix analysis on the aforementioned isomorphic weighted graph to obtain a symmetric normalized Laplacian matrix. Secondly, it can perform eigenvalue decomposition on the symmetric normalized Laplacian matrix to obtain a set of non-zero eigenvectors. Then, it can concatenate these non-zero eigenvectors column-wise to obtain concatenated eigenvectors. Finally, it can perform clustering on these concatenated eigenvectors to obtain the user community clusters.

[0060] The fourth step involves adjusting the value classification of the aforementioned value grading probability set and the aforementioned user lifetime value prediction set based on the user community cluster set, resulting in an adjusted value level information set and a user value hierarchy information set. The adjusted value information in the adjusted value level information set can be obtained by further refining the clusters corresponding to the aforementioned user lifetime value prediction set.

[0061] As an example, the aforementioned implementing entity can first classify user value using the aforementioned value grading probability set and the aforementioned user lifetime value prediction value set, resulting in high-value user clusters, medium-value user clusters, and low-value user clusters. Then, using the user community cluster set, behavioral similarity is used to classify the high-value, medium-value, and low-value user clusters, resulting in an adjusted value level information set, and ultimately, a user value hierarchy information set. For example, the adjusted value level information could be derived from similar behaviors within the high-value user cluster, resulting in a brand-loyalty high-value user cluster and a stockpiling high-value user cluster.

[0062] In addressing the technical problems mentioned above, the application scenario—item inventory scheduling driven by different user behaviors within the same user value level information under different item states—often presents the following technical challenges: Inventory backlog or shortage can cause different item states (e.g., price reduction, price increase, pre-sale), which in turn affect the behavior of users with different behavioral preferences within the same value stratification. Existing methods utilize the Leiden community discovery algorithm to partition communities based on modularity and topology. Modularity optimization tends to merge smaller communities. However, in private domain scenarios, the data volume of high-value clusters is significantly smaller than that of medium- and low-value clusters, resulting in lower boundaries between different value levels. This reduces the accuracy and speed of user community cluster partitioning, lowers the accuracy and speed of user value stratification, reduces user push conversion rates, and makes it impossible to predict item inventory quantities accurately and timely, leading to item shortages or backlogs. Considering the following requirements for this application scenario: adaptability to clear user cluster boundaries, adaptability to long-tail circles (small user base but high user stickiness), adaptability to dynamic changes in user behavior, and adaptability to value heterogeneity, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the above-described temporal community graph partitioning of the user isomorphic weighted graph to obtain user community clusters may include the following steps: The first step involves performing time slicing on the aforementioned user isomorphic weighted graph to obtain a sequence of user association time-series slice graphs. These user association time-series slice graphs can be user association graphs obtained by slicing the user isomorphic weighted graph according to preset time slices. The preset time slices can be time slices with a 6-hour period. These time slices can capture the structural changes of the user isomorphic weighted graph over time. The aforementioned user association time-series slice graphs can be multi-relationship graphs consisting of user nodes and representing different relationships between any two user nodes. These different relationships can include, but are not limited to, at least one of the following: friend relationships, relationships with similar preferences, relationships with the same user value hierarchy information, and recommendation relationships.

[0063] The second step involves using a user association graph attention model to perform multi-dimensional attention aggregation on each user association time-series slice in the aforementioned user association time-series slice sequence, resulting in a sequence of node fusion feature vector groups. The aforementioned user association graph attention model can be a deep neural network that performs node-level and semantic-level attention aggregation on the user association time-series slices in the input sequence to output a set of node fusion feature vector groups. For example, the aforementioned user association graph attention model could be a relational graph attention network. The aforementioned node-level attention aggregation can be attention aggregation on multiple neighboring user nodes connected by edges of the same type. The aforementioned semantic-level attention aggregation can be attention aggregation on multiple neighboring user nodes connected by edges of different types. The node fusion feature vectors in the aforementioned sequence of node fusion feature vector groups can be node semantic information and node connection relationship information obtained by fusing edges representing different association relationships and multiple user nodes with neighbor relationships.

[0064] The third step involves performing time-aware embedding processing on the aforementioned node fusion feature vector group sequence to obtain a node time-aware feature vector group sequence. The node time-aware feature vectors in this sequence can be feature vectors that fuse node-level, semantic-level, and temporal feature information. In practice, the execution entity can use DySAT (Dynamic Self-Attention Network, a dynamic graph neural network model based on a self-attention mechanism) to perform time-aware embedding processing on the aforementioned node fusion feature vector group sequence to obtain the node time-aware feature vector group sequence.

[0065] The fourth step is to traverse the aforementioned isomorphic weighted graph of users according to the order of node appearance, thereby obtaining a set of user node sequences to be partitioned. This set of user node sequences can be obtained by traversing user nodes with connecting edges in the order of their appearance.

[0066] The fifth step is to construct user communities for the above-mentioned sequence of user nodes to be partitioned, thereby obtaining an initial set of user node communities. The initial user node communities in this initial set can be clusters of user nodes located at the initial positions in the sequence of user nodes to be partitioned.

[0067] Step 6: Based on the aforementioned sequence of node time-aware feature vectors, perform community partitioning on the initial user node community set to obtain a user node association partitioning community set. This user node association partitioning community set can be obtained by assigning each user node to be partitioned in the user node sequence after removing the initial user node community set to its corresponding user node community. The user node association partitioning community can be a community where the node time-aware feature vectors of the included user nodes are relatively similar and have dense connections, while the connections between different user node association partitioning communities are relatively sparse. In practice, the executing entity can sequentially determine each user node in the user node sequence after removing the initial user node as the target user node sequence set. Secondly, the mean cosine similarity of the node temporal-aware feature vector of each target user node to be partitioned and the node temporal-aware feature vectors of each user node to be partitioned included in the initial user node community set with interconnected relationships is determined. When the mean is greater than or equal to a preset embedding similarity threshold, the target user node to be partitioned is assigned to an initial user node community with interconnected relationships; otherwise, the target user node to be partitioned is determined as an initial user node community. This process continues until all user nodes to be partitioned are partitioned, resulting in a user node association partitioning community set. The preset embedding similarity threshold can be a pre-set minimum value used to determine whether nodes belong to the same initial user node community. For example, the preset embedding similarity threshold could be 0.7.

[0068] Step 7: Perform multi-resolution adaptive community partitioning on the aforementioned user node association community set to obtain a multi-resolution user node community set. The multi-resolution user node communities in this set can be communities composed of user value hierarchy information with significant differences in community data volume and high behavioral similarity. In practice, the executing entity can first perform a first multi-resolution adaptive community partitioning on the aforementioned user node association community set using the Leiden algorithm with a first resolution parameter value to obtain a first multi-resolution user node community set. The Leiden algorithm with the first resolution parameter value can be a formula that adds a resolution parameter value to the product of the degrees of two nodes divided by twice the total number of edges in the existing modularity formula. The first resolution parameter value can be a parameter value that tends to discover large structures to divide macroscopic user circles. For example, the first resolution parameter value can be 0.5. Then, using the Leiden algorithm with a second resolution parameter value, perform a second multi-resolution adaptive community partitioning on the first multi-resolution user node community set to obtain a third multi-resolution user node community set. The aforementioned second resolution parameter value can be used to force further fragmentation in order to delineate communities with small connectivity densities. This second resolution parameter value can be a value between 1.5 and 2.

[0069] Step 8: Based on the aforementioned node time-series-aware feature vector sequence, perform boundary allocation processing on the aforementioned user node multi-resolution partitioned community set to obtain a user community cluster set. In practice, the aforementioned execution entity can first determine the local clustering coefficient and degree centrality of each user node's multi-resolution partitioned community. The local clustering coefficient can be the ratio of the actual number of connecting edges between a node's neighbors to the number of possible connecting edges. Secondly, select at least one user node from the aforementioned user node multi-resolution partitioned community set whose values ​​are less than a preset local clustering coefficient threshold and a preset degree centrality threshold. The preset local clustering coefficient threshold and preset degree centrality threshold can be pre-set critical values ​​used to distinguish core nodes from peripheral nodes; their specific values ​​can be determined based on specific circumstances and will not be elaborated here. Then, using the node time-series-aware feature vector sequence, determine the cosine similarity between at least one user node and the central node of each adjacent user node's multi-resolution partitioned community, as the community cosine similarity set. Finally, if the cosine similarity of all communities is less than the preset community similarity threshold, the user node is removed from the multi-resolution user node partitioning community; otherwise, the user node is partitioned into the multi-resolution user node partitioning community where the similarity is greater than the preset community similarity threshold.

[0070] The above technical solution, combined with steps 106 to 109 and related content, serves as an inventive point of this disclosure, solving the technical problem mentioned in the background art: "leading to reduced accuracy in user community clustering, lower user push conversion rate, inability to predict item inventory quantity in a timely and accurate manner, resulting in item shortages or backlogs." The factors leading to reduced accuracy and speed in user community clustering, lower user push conversion rate, inability to predict item inventory quantity in a timely and accurate manner, and resulting in item shortages or backlogs are often as follows: Because the formation of the same value stratification can result in multiple different behaviors, the Leiden community discovery algorithm uses modularity and topology to divide communities. Modularity optimization tends to merge small communities. However, in private domain scenarios, the amount of data in high-value clusters is significantly less than the amount of data in medium- and low-value clusters, resulting in lower boundaries between different value levels. This reduces the accuracy and speed of user community clustering, lowers the accuracy and speed of user value stratification, reduces user push conversion rate, and leads to the inability to predict item inventory quantity in a timely and accurate manner, resulting in item shortages or backlogs. Solving the above factors can improve the accuracy of user community clustering, reduce low user push conversion rates, and accurately predict item inventory levels, thereby reducing item shortages or overstocking. To achieve this, this disclosure first performs time-slicing processing on the user isomorphic weighted graph and embeds the user-related time-series slice graph sequence at the node, semantic, and temporal levels. This allows for real-time capture of the structural changes of the user isomorphic weighted graph over time, precise measurement of heterogeneous edges in the user isomorphic weighted graph and assignment of different weights, improving the comprehensiveness and accuracy of node time-series-aware feature vectors, and avoiding semantic fragmentation of subgraphs in subsequent partitioning. Second, community partitioning based on the node appearance order is performed using the node time-series-aware feature vector sequence. Single-pass scanning partitioning reduces memory consumption and provides global graph structure guidance for subsequent community partitioning, avoiding semantic fragmentation caused by partitioning. Then, multi-resolution adaptive community partitioning can macroscopically identify behavioral user value circles and microscopically separate high-net-worth sub-circles, effectively avoiding the problem of low partitioning accuracy caused by large differences in community data volume. Subsequently, by analyzing the local clustering coefficient and degree centrality of nodes, the user nodes are further processed for boundary allocation, resulting in user community clusters. This improves the accuracy of segmentation and effectively refines the segmentation of different behavioral circles under a unified framework based on behavioral similarity. Finally, content generation scheduling and item inventory adjustments are performed using user community clusters. Segmenting communities based on similar user behaviors at the same user value level further refines user behavior preference information, improves user push conversion rates, and allows for timely and accurate prediction of item inventory, reducing item shortages or overstocking.

[0071] Step 105: Generate a private domain channel content push information set based on the private domain channel feature information and the user value hierarchy information set.

[0072] In some embodiments, the aforementioned executing entity may generate a private domain channel content push information set based on the private domain channel characteristic information and the aforementioned user value hierarchy information set. The aforementioned private domain channel characteristic information may characterize the traffic distribution information of the various private domain channels. The aforementioned private domain channel characteristic information may include, but is not limited to, at least one of the following: private domain channel geographic distribution information, peak traffic periods, private domain channel topic traffic information, user interaction rate, and private channel content review information. The private domain channel content push information in the aforementioned private domain channel content push information set may be item push text and images that conform to user preferences and user value hierarchy information. For example, the aforementioned private domain channel content push information may be information composed of one or more of the following forms: video, text and images, and audio.

[0073] In addressing the aforementioned technical challenges by employing technical solutions, the specific application scenario—item inventory scheduling based on content push notifications across different private channels during holidays—often presents the following technical issues: While users have more time to browse and engage with content during holidays, existing push content generation typically utilizes a large language model to generate uniform content based on user historical behavior preferences, resulting in low-quality content push notifications that are unsuitable for the varying requirements of different private channels. This leads to low push notification efficiency, reduced user conversion rates, difficulty in accurately predicting item inventory, and an increase in inventory shortages or overstocking. Considering the following requirements for this application scenario: adaptability to multiple private channels, adaptation to the platform's high-quality content review mechanism, adaptation to personalized user requirements, applicability to refined content push notifications, and adaptability to large volumes and diverse formats of push content, we have decided to adopt the following solution: In some optional implementations of certain embodiments, generating a private domain channel content push information set based on the private domain channel feature information and the user value hierarchy information set may include the following steps: The first step is to obtain the channel traffic dataset and traffic detection information set of the private domain channel through a third-party interface. The channel traffic data in the aforementioned traffic dataset can be traffic characteristic information of different private domain channels. For example, the aforementioned channel traffic data may include, but is not limited to, at least one of the following: traffic geographic distribution, traffic distribution information, traffic topics, interaction rate, likes, favorites, and read distribution information, and user / fan profiles. The traffic detection information in the aforementioned traffic detection information set can be the content review information of each private domain channel. The aforementioned traffic detection information may include, but is not limited to, at least one of the following: prohibited words, sensitive words, and content review rules information. The aforementioned third-party interface of the private domain channel can be a third-party interface of the platform connected to the private domain channel.

[0074] The second step involves feature extraction from the aforementioned channel traffic dataset to obtain a channel traffic feature information set. This set includes at least one of the following: traffic distribution over time periods, content type traffic share, user interaction rate, and traffic peak prediction window. The channel traffic feature information in this set can characterize the high-level semantic and contextual information of the aforementioned channel traffic dataset. Feature extraction can be performed using a large language model. The user interaction rate can be the percentage of user interactions such as likes, comments, and shares. The traffic peak prediction window can be the time period with the highest traffic in the private channel.

[0075] The third step involves extracting detection features from the aforementioned traffic detection information set to obtain a content detection feature information set. This content detection feature information set includes at least one of the following: detection dimension information, content structure requirement information, and historical detection feedback information. Specifically, the content detection feature information in this set can represent the semantic information, multimodal review dimension information, and generative content detection annotation information of the traffic detection information. The aforementioned detection feature extraction can be performed using the BERT model. The aforementioned content structure requirement information can be information regarding the push structure requirements of the private channel content push information to be published. For example, the aforementioned content structure requirement information may include, but is not limited to, at least one of the following: identifiers for generative content information, font requirements, content style requirements, and audio / video content direction requirements.

[0076] The fourth step is to determine the user traffic preference information set based on the aforementioned user value hierarchy information set. The associated user information in the aforementioned associated user information set can be user information that has a relationship with the user. The user traffic preference information in the aforementioned user traffic preference information set can be the user's personalized demand information. For example, the aforementioned user traffic preference information can include, but is not limited to, at least one of the following: new product preference information, item efficacy preference information, and item-related enterprise service preference information. In practice, the aforementioned executing entity can first use a collaborative filtering algorithm to extract traffic preferences from the aforementioned user value hierarchy information set to obtain the user traffic preference information set.

[0077] Fifth, based on the aforementioned user value hierarchy information set and the aforementioned user behavior path heterogeneous graph, the aforementioned user traffic preference information set is expanded to obtain a user preference expansion information set. This expanded user preference information can be derived by identifying user traffic preference information associated with other user information as the user's preference information. In practice, the executing entity can perform the following preference expansion steps for each user: First, using a graph embedding model, the static structural attributes of the user value graph corresponding to the aforementioned user value hierarchy information set and the aforementioned user behavior path heterogeneous graph are extracted to obtain a static structural feature vector. This graph embedding model can be a Deepwalk model or a Node2Vec model. Second, using a random walk algorithm, the group perception of the aforementioned user value graph and the aforementioned user behavior path heterogeneous graph is extracted to obtain a group perception feature vector. Third, the sum of the function values ​​of the indicator functions of the interaction values ​​of the target user corresponding to the aforementioned user traffic preference information and each user in the associated user information set included in the aforementioned user value graph and user behavior path heterogeneous graph, and the ratio of this sum to the number of associated user information included in the aforementioned user value graph and user behavior path heterogeneous graph, is determined as the group interaction feature vector. Then, the dynamic structural feature vectors and dynamic interaction feature vectors of newly added users whose changes occurred within a preset time period are determined. Next, the static structural feature vectors, group perception feature vectors, group interaction feature vectors, dynamic structural feature vectors, and dynamic interaction feature vectors are normalized, concatenated, and then input into a multi-head attention mechanism network and a feedforward neural network to obtain a set of social influence weight values. Subsequently, the user traffic preference information set corresponding to at least one social influence weight value greater than or equal to a preset social influence threshold is determined as the extended preference information set. Finally, the extended preference information set and the user traffic preference information set are deduplicated to obtain the user preference extended information set.

[0078] Step six: Based on the aforementioned user value hierarchy information set, generate user content push strategy information. This user content push strategy information can be information regarding the push time, frequency, and dimensions of content at different levels, determined by the hierarchy information in the user value hierarchy information set. In practice, the executing entity can utilize a rule engine to generate user content push strategy information based on the aforementioned user value hierarchy information set. For example, the rule engine could be the Drools rule engine.

[0079] Step 7: Based on the aforementioned channel traffic feature information set, content detection feature information set, user preference extension information set, and user content push strategy information, generate a channel push content prompt word set. The channel push content prompt words in this set can be engineered prompt word information from a large language model that applies multi-dimensional constraints such as the user preference extension information set, channel traffic feature information set, content detection feature information set, and user content push strategy information to subsequently generated user content push information. In practice, the executing entity can input the aforementioned channel traffic feature information set, content detection feature information set, user preference extension information set, and user content push strategy information into the large language model to obtain the channel push content prompt word set.

[0080] Step 8: Input the aforementioned set of prompt words for channel push content into the channel content generation model to obtain the initial user content push information set. The initial user content push information in this set can be content push information that meets the requirements of the aforementioned prompt words for channel push content. The aforementioned channel content generation model can be a large language model that generates content from the input set of prompt words for channel push content and outputs the initial user content push information set. For example, the aforementioned channel content generation model could be the DeepSeek model.

[0081] Step nine involves performing channel style adaptation processing on the initial user content push information set to obtain a private domain channel content push information set. The type information of this private domain channel content push information includes at least one of the following: image type information, audio / video type information, text type information, and multi-structure mixed type information. Specifically, the private domain channel content push information in this set can be image and audio / video content push information with matched content style added to the initial user content push information. The multi-structure mixed type information can be obtained by combining multiple types from the image, audio / video, and text types. In practice, the executing entity can utilize the Puff-Net style transfer algorithm to perform channel style adaptation processing on the initial user content push information set to obtain the private domain channel content push information set.

[0082] The above technical solution, combined with steps 107-109 and related content, serves as an inventive point of this disclosure, solving the technical problem of "low-quality generated content push information, low push efficiency, reduced user conversion rate, difficulty in accurately predicting item inventory, and increased shortage or backlog of item inventory." The factors contributing to low-quality generated content push information, low push efficiency, reduced user conversion rate, difficulty in accurately predicting item inventory, and increased shortage or backlog of item inventory are often as follows: During holidays, users have more time to browse and pay attention to content pushes. However, existing push content generation typically uses a large language model to generate unified content based on user historical behavior preferences and sends it to different private domain channels. This results in low-quality generated content push information, incompatibility with the different requirements of different private domain channels, low push efficiency, reduced user conversion rate, difficulty in accurately predicting item inventory, and increased shortage or backlog of item inventory. Solving these factors can improve the quality of generated content push information, enhance adaptability to the different requirements of different private domain channels, improve push efficiency, increase user conversion rate, improve the accuracy of item inventory prediction, and reduce the shortage or backlog of item inventory. To achieve this effect, this disclosure firstly obtains channel traffic datasets and traffic detection information sets through third-party interfaces of private channels and performs feature extraction. This allows for a comprehensive understanding of the traffic and content detection information of different private channels, thereby improving the success rate of subsequent content push notifications. Traffic feature extraction enables precise perception of traffic trends in private channels, improving the accuracy of subsequent push strategies. Detection information feature extraction quantifies content review, allowing for the avoidance of review points in private channels during the generation process and increasing the likelihood of approval. Secondly, determining user preference information at different value levels and associating it with user expansion can enhance the explicit and concrete representation of users' personalized needs. User preference expansion can comprehensively extract user preference information, which can, to some extent, prevent the emergence of information cocoons. Finally, a rule engine is used to determine user content push strategy information, ensuring the accuracy of user content push strategy information and avoiding user churn caused by excessive push notifications. Subsequently, generating multi-dimensional constraint prompts and generating initial user content push information sets based on the large model can achieve comprehensive boundary constraints on the generated content push information, making the generated content push information more in line with the requirements of private domain channels, improving its quality and avoiding the illusion phenomenon generated by the large model. The large model's large output in a short period of time can meet the traffic demand and generation speed of holidays, reducing the cost of generating content push information.Finally, channel style adaptation processing of the initial user content push information set can realize cross-modal content generation and channel visual adaptation, as well as dynamic adjustment of item inventory through private domain channel content push information set, further increasing the conversion rate and click-through rate of content push information, improving the accuracy of predicted item inventory, and reducing item inventory shortages or backlogs.

[0083] Step 106: Perform content detection processing on the private channel content push information set to obtain the push content detection result set.

[0084] In some embodiments, the aforementioned execution entity can perform content detection processing on the aforementioned private channel content push information set to obtain a push content detection result set. The push content detection results in the aforementioned push content detection result set can be information used to determine whether the private channel content push information meets the private channel's review requirements. In practice, the aforementioned execution entity can first perform word segmentation processing on the aforementioned private channel content push information set to obtain a push word segmentation set. Then, it can match the aforementioned push word segmentation set with preset channel prohibited words to obtain a prohibited word matching result set. The preset channel prohibited words can be prohibited words defined by the private channel. Afterwards, it can remove at least one push word representing a successful prohibited word match from the aforementioned push word segmentation set to obtain a target push word segmentation set. Finally, it can use a DFA (Deterministic Finite Automaton) to identify sensitive words in the target push word segmentation set to obtain the push content detection result set.

[0085] Step 107: Based on the push content detection result set and private domain channel feature information, perform traffic task scheduling processing on the private domain channel content push information set to obtain private domain channel push scheduling information.

[0086] In some embodiments, the executing entity can perform traffic task scheduling processing on the private channel content push information set based on the push content detection result set and the private channel feature information to obtain private channel push scheduling information. The private channel push scheduling information may include information on the push time, the private channel, and the target users for the push of the private channel content push information set. In practice, the executing entity can use a round-robin method to perform traffic task scheduling processing on the private channel content push information set based on the push content detection result set and the private channel feature information to obtain private channel push scheduling information.

[0087] Step 108: Based on the private domain channel push scheduling information, push and transmit the private domain channel content push information set to the user terminal set corresponding to the user identification information set.

[0088] In some embodiments, the aforementioned execution entity may, based on the aforementioned private channel push scheduling information, push and transmit the aforementioned private channel content push information set to the user terminal set corresponding to the aforementioned user identification information set.

[0089] Step 109: Dynamically adjust the inventory of goods based on the set of goods circulation information corresponding to the set of user terminals obtained.

[0090] In some embodiments, the aforementioned executing entity can dynamically adjust the inventory of goods based on the acquired set of goods circulation information corresponding to the user terminal set. This dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed replenishment. The goods circulation information can be the goods circulation information of items purchased by users through a private channel push notification. The goods inventory can be information on the quantity and category of goods stored by the enterprise. Dynamic adjustment of the goods inventory can include, but is not limited to, at least one of the following: discounting items to increase sales or increasing the quantity of goods in stock. In practice, the aforementioned executing entity can first input the goods circulation information into a goods demand forecasting model to obtain goods forecast information. This goods demand forecasting model can be a gray neural network model that predicts the demand for goods in a future time period based on the input goods circulation information and outputs the goods forecast information. The goods demand forecasting model can be a model trained using forward propagation and backpropagation methods. Then, the difference between the current goods inventory and the goods inventory in the goods forecast information is determined. Finally, when the inventory difference is less than the preset safety stock, the inventory is replenished; when the inventory difference is greater than the preset safety stock, the inventory is promoted.

[0091] The above embodiments of this disclosure have the following beneficial effects: The item inventory scheduling method based on private domain channels in some embodiments of this disclosure can improve the accuracy of push notifications, thereby accurately predicting changes in item inventory quantities, reducing item waste, and reducing the occurrence of item shortages or stockpiles. Specifically, the reasons for the inability to predict item inventory quantities in a timely and accurate manner, resulting in item shortages or stockpiles, are as follows: The data is too one-dimensional and incomplete because only user behavior information from a single private domain channel is obtained; the RFM model only relies on user consumption information when performing user value segmentation, without considering other data affecting user value information, resulting in low accuracy in user value segmentation; at the same time, the large language model suffers from model illusion problems, and the content push information generated for users at the same level is too general and does not consider user personalization information, resulting in low quality of content push information and low user push conversion rates, leading to the inability to predict item inventory quantities in a timely and accurate manner, resulting in item shortages or stockpiles. Based on this, some embodiments of the item inventory scheduling method based on private domain channels disclosed herein can first, in response to detecting a shortage or backlog of item inventory, obtain a set of user private domain channel behavior information from multiple private domain channels. This set of user private domain channel behavior information includes at least one of the following: data source information, user identification information, user item acquisition information, and user browsing behavior information. Obtaining user private domain channel behavior information from multiple private domain channels improves data comprehensiveness, avoids data silos, and facilitates subsequent user value stratification processing. Secondly, the user private domain channel behavior information set is aggregated to obtain a user behavior information set. This improves the quality of user behavior information and reduces the occurrence and volume of conflicting data. Thirdly, the user behavior information set undergoes anomaly processing to obtain a target user behavior information set. Anomaly processing can automatically identify malicious behaviors such as fraudulent transactions, facilitating improved accuracy in subsequent user value stratification. Next, based on the target user behavior information set, the user identification information set is processed for user value stratification to obtain a user value hierarchy information set. Here, user value stratification effectively classifies users into different levels, facilitating the generation of personalized private channel content push notifications. Subsequently, based on private channel feature information and the aforementioned user value stratification information set, a private channel content push notification set is generated. This personalization ensures the private channel content push notification set meets user preferences and private channel needs, enabling subsequent content detection. Afterward, content detection processing is performed on the aforementioned private channel content push notification set to obtain a push content detection result set. This content detection process automatically identifies content in the private channel content push notifications that does not conform to the private channel's rules, improving the quality of the private channel content push notifications and the success rate of subsequent pushes.Then, based on the aforementioned push content detection result set and the aforementioned private domain channel feature information, traffic task scheduling processing is performed on the aforementioned private domain channel content push information set to obtain private domain channel push scheduling information. Here, traffic task scheduling processing can reasonably follow the inference timing and quantity of the private domain channel content push information set, increasing the accuracy and precision of scheduling while meeting the requirements of the private domain channel. Next, based on the aforementioned private domain channel push scheduling information, the aforementioned private domain channel content push information set is pushed and transmitted to the user terminal set corresponding to the aforementioned user identification information set. Here, it is pushed to each user terminal for subsequent observation of user feedback information and conversion rates. Finally, based on the obtained item flow information set corresponding to the aforementioned user terminal set, the item inventory is dynamically adjusted, whereby the aforementioned dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed item replenishment. Here, by analyzing the user private domain channel behavior information set after the push, the user conversion rate can be obtained, which can improve the accuracy of predicting item inventory changes, reduce stockouts or backlogs, increase inventory turnover, and reduce item waste. Therefore, this private domain channel-based item inventory scheduling method can improve push accuracy, thereby accurately predicting changes in item inventory quantity, reducing item waste, and minimizing item shortages or stockpiles.

[0092] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of an item inventory scheduling device based on private domain channels. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this private domain channel-based item inventory scheduling device can be specifically applied to various electronic devices.

[0093] like Figure 2As shown, a private domain channel-based item inventory scheduling device 200 includes: an acquisition unit 201, an aggregation unit 202, an anomaly handling unit 203, a user value stratification unit 204, a generation unit 205, a content detection unit 206, a traffic task scheduling unit 207, a push transmission unit 208, and an adjustment unit 209. The acquisition unit 201 is configured to: in response to detecting a shortage or overstock of item inventory, acquire a set of user private domain channel behavior information from multiple private domain channels, wherein the user private domain channel behavior information set includes at least one of the following: data source information, user identification information, user item acquisition information, and user browsing behavior information. The aggregation unit 202 is configured to: aggregate the user private domain channel behavior information set to obtain a user behavior information set. The anomaly handling unit 203 is configured to: perform anomaly processing on the user behavior information set to obtain a target user behavior information set. The user value stratification unit 204 is configured to: perform user value stratification processing on the user identification information set according to the target user behavior information set to obtain a user value hierarchy information set. The generation unit 205 is configured to generate a private domain channel content push information set based on the private domain channel feature information and the aforementioned user value hierarchy information set. The content detection unit 206 is configured to perform content detection processing on the aforementioned private domain channel content push information set to obtain a push content detection result set. The traffic task scheduling unit 207 is configured to perform traffic task scheduling processing on the aforementioned private domain channel content push information set based on the aforementioned push content detection result set and the aforementioned private domain channel feature information to obtain private domain channel push scheduling information. The push transmission unit 208 is configured to push and transmit the aforementioned private domain channel content push information set to the user terminal set corresponding to the aforementioned user identification information set based on the aforementioned private domain channel push scheduling information. The adjustment unit 209 is configured to dynamically adjust the item inventory based on the obtained item flow information set corresponding to the aforementioned user terminal set, wherein the aforementioned dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed item replenishment.

[0094] It is understandable that the units recorded in the private domain channel-based item inventory scheduling device 200 are related to the reference... Figure 1 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method also apply to the item inventory scheduling device 200 based on private domain channels and the units contained therein, and will not be repeated here.

[0095] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device (e.g., an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0096] like Figure 3 As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0097] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0098] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0099] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0100] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0101] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: in response to detecting a shortage or overstock of items, acquire a set of user private domain channel behavior information from multiple private domain channels, wherein the set of user private domain channel behavior information includes at least one of the following: data source information, user identification information, user item acquisition information, and user browsing behavior information; aggregate the set of user private domain channel behavior information to obtain a user behavior information set; perform anomaly processing on the set of user behavior information to obtain a target user behavior information set; perform user value stratification processing on the set of user identification information based on the target user behavior information set to obtain a user value hierarchy information set; and perform private domain channel... Based on the channel feature information and the aforementioned user value hierarchy information set, a private domain channel content push information set is generated; content detection processing is performed on the aforementioned private domain channel content push information set to obtain a push content detection result set; based on the aforementioned push content detection result set and the aforementioned private domain channel feature information, traffic task scheduling processing is performed on the aforementioned private domain channel content push information set to obtain private domain channel push scheduling information; based on the aforementioned private domain channel push scheduling information, the aforementioned private domain channel content push information set is pushed and transmitted to the user terminal set corresponding to the aforementioned user identification information set; based on the obtained item flow information set corresponding to the aforementioned user terminal set, the item inventory is dynamically adjusted, wherein the aforementioned dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed item replenishment.

[0102] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0103] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0104] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an acquisition unit, an aggregation unit, an exception handling unit, a user value stratification unit, a generation unit, a content detection unit, a traffic task scheduling unit, a push transmission unit, and an adjustment unit. The names of these units do not necessarily limit the unit itself; for example, the acquisition unit may also be described as "a unit that, in response to detecting a shortage or overstock of item inventory, acquires a set of user private domain channel behavior information from multiple private domain channels."

[0105] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0106] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for scheduling item inventory based on private domain channels, comprising: In response to the detection of a shortage or backlog of items in inventory, a set of user private domain channel behavior information is obtained from multiple private domain channels, wherein the set of user private domain channel behavior information includes at least one of the following: a set of data source information, a set of user identification information, a set of user item acquisition information, and a set of user browsing behavior information; The user private channel behavior information set is aggregated to obtain the user behavior information set; Anomaly processing is performed on the user behavior information set to obtain the target user behavior information set; Based on the target user behavior information set, the user identification information set is processed to perform user value stratification to obtain a user value stratification information set. Based on the private domain channel feature information and the user value hierarchy information set, a private domain channel content push information set is generated; The private channel content push information set is subjected to content detection processing to obtain the push content detection result set; Based on the push content detection result set and the private domain channel feature information, traffic task scheduling processing is performed on the private domain channel content push information set to obtain private domain channel push scheduling information. According to the private domain channel push scheduling information, the private domain channel content push information set is pushed and transmitted to the user terminal set corresponding to the user identification information set; Based on the acquired set of item flow information corresponding to the user terminal set, the item inventory is dynamically adjusted, wherein the dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed item replenishment.

2. The method according to claim 1, wherein, The user's private channel behavior information set is obtained through the following steps: For each user's private channel behavior information in the user's private channel behavior information set, perform the following transmission steps: The user's private channel behavior information is preprocessed locally to obtain preprocessed channel behavior information; Determine the receiver address information of the data receiver corresponding to the preprocessed channel behavior information; In response to confirming a successful communication connection with the data receiving end, an end-to-end covert handshake message is sent to the data receiving end using the receiving end address information. The end-to-end covert handshake message includes: the size of the data transmission byte stream, the data integrity check value, the covert channel version information, and the covert channel type information. In response to receiving the end-to-end covert handshake message information sent by the data receiving end, and the covert channel type information is a stored covert channel, a first covert channel symmetric key is generated. Based on the first covert channel symmetric key, the preprocessed channel behavior information is cyclically encrypted to obtain an encrypted user behavior byte stream set; The encrypted user behavior byte stream set and the synchronization transmission identifier information are embedded into the target field set in the end-to-end covert handshake message information to obtain the handshake embedding protocol message information. Traffic anomaly detection is performed on the handshake embedded protocol message information to obtain traffic anomaly detection result information; In response to determining that the traffic anomaly detection result information indicates no anomaly, the number of data transmissions of the handshake embedded protocol message information is determined based on the encrypted user behavior byte stream set; The handshake embedded protocol message information is divided into several data transmission messages and transmitted separately, so that the data receiving end can decrypt and receive the user's private channel behavior information.

3. The method according to claim 2, wherein, The step of performing byte-stream cyclic encryption on the preprocessed channel behavior information based on the first covert channel symmetric key to obtain an encrypted user behavior byte-stream set includes: Based on the first covert channel symmetric key and preprocessed channel behavior information, the following encryption steps are performed: Based on the first covert channel symmetric key, the transmitted byte stream located at the beginning position in the preprocessed channel behavior information is encrypted to obtain the first encrypted user behavior byte stream; Based on the first encrypted user behavior byte stream, the target transmission byte stream is symmetrically encrypted to obtain the second encrypted user behavior byte stream, wherein the target transmission byte stream is the byte stream located after the starting position in the preprocessed channel behavior information; In response to determining that the target transmission byte stream is a transmission byte stream located at the end position of the preprocessed channel behavior information, a set of encrypted user behavior byte streams is determined based on the first encrypted user behavior byte stream and the second encrypted user behavior byte stream; In response to determining that the target transmission byte stream is not a transmission byte stream located at the termination position, the accumulated second encrypted user behavior byte stream is determined as the first covert channel symmetric key, and the transmission byte streams corresponding to the accumulated first encrypted user behavior byte stream and the second encrypted user behavior byte stream are removed and determined as the remaining transmission byte stream set, so as to perform the encryption step again.

4. The method according to claim 1, wherein, The aggregation processing of the user's private channel behavior information set to obtain the user behavior information set includes: Based on the user private channel behavior information set, construct an undirected graph of user channel identifiers; The user channel identifier undirected graph is subjected to connected component partitioning to obtain a user identifier connected component graph set; Based on the user identifier connected component graph set, user identifier matching processing is performed on the user private domain channel behavior information set to obtain the target user identifier matching information set. Based on the target user identifier matching information set, information extraction processing is performed on the user private channel behavior information set to obtain user attribute information set and user behavior information set; The user attribute information set is subjected to voting conflict disambiguation processing to obtain the disambiguated user attribute information set; The user behavior information set is subjected to indicator conflict disambiguation processing to obtain the disambiguated user behavior information set; Based on the target user identifier matching information set, the disambiguated user attribute information set and the disambiguated user behavior information set are aggregated to obtain the user behavior information set.

5. The method according to claim 1, wherein, The step of performing user value stratification processing on the user identification information set based on the target user behavior information set to obtain a user value hierarchy information set includes: Heterogeneous nodes and user behavior information are extracted from the target user behavior information set to obtain a heterogeneous node set and a user behavior information set, wherein the heterogeneous node set includes at least one of the following: target user node set, item node set, private domain content node set, and private domain channel push node set; Feature extraction is performed on the node attribute information set corresponding to the heterogeneous node set to obtain the node feature information set; Determine the edge weight information of the user behavior information set, and use it as the behavior edge weight set; The heterogeneous node set, the user behavior information set, the node feature information set, and the behavior edge weight set are input into the graph database to obtain a heterogeneous user behavior graph. Meta-path construction is performed on the user behavior heterogeneous graph to obtain a user behavior path heterogeneous graph; Behavioral interaction information is extracted from the target user behavior information set to obtain a behavioral interaction time sequence information set, wherein the behavioral interaction time sequence information is a time sequence triple consisting of user behavior, item information / private domain content and interaction time; Sequence feature encoding is performed on the behavioral interaction time sequence information set to obtain a behavioral interaction time sequence feature information set; The behavioral interaction time sequence feature information set and the user behavior path heterogeneous graph are fused and compared to obtain a fused interaction feature information set. Based on the fused interaction feature information set, user value is divided into the time series graph corresponding to the heterogeneous graph of user behavior path and the time series graph of behavior interaction time series information set, so as to obtain the user value hierarchy information set.

6. The method according to claim 5, wherein, The step involves classifying user value based on the fused interaction feature information set, the heterogeneous user behavior path graph, and the time series graph corresponding to the behavior interaction time series information set, to obtain a user value hierarchy information set, including: The fused interactive feature information set is input into the user value probability prediction network to obtain the user lifetime value prediction value set and the value classification probability set. Based on the fused interaction feature information set, the user graph is reconstructed by the time series graph corresponding to the heterogeneous graph of the user behavior path and the time series graph of the behavior interaction time series information set, resulting in a user isomorphic weighted graph. The connection edge type in the user isomorphic weighted graph includes at least one of the following: friend relationship, item purchase relationship, and same interest relationship. The user isomorphic weighted graph is partitioned into a temporal community graph to obtain a set of user community clusters; Based on the user community cluster set, the value classification probability set and the user lifecycle value prediction value set are adjusted to obtain the adjusted value level information set, and thus the user value level information set.

7. A private domain channel-based item inventory scheduling device, comprising: The acquisition unit is configured to acquire a set of user private domain channel behavior information from multiple private domain channels in response to detecting a shortage or backlog of item inventory. The set of user private domain channel behavior information includes at least one of the following: a data source information set, a user identification information set, a user item acquisition information set, and a user browsing behavior information set. The aggregation unit is configured to aggregate the user private channel behavior information set to obtain a user behavior information set. An exception handling unit is configured to perform exception handling on the user behavior information set to obtain a target user behavior information set. The user value stratification unit is configured to perform user value stratification processing on the user identification information set based on the target user behavior information set to obtain a user value stratification information set. The generation unit is configured to generate a private domain channel content push information set based on the private domain channel feature information and the user value hierarchy information set; The content detection unit is configured to perform content detection processing on the private channel content push information set to obtain a push content detection result set; The traffic task scheduling unit is configured to perform traffic task scheduling processing on the private channel content push information set based on the push content detection result set and the private channel feature information to obtain private channel push scheduling information. The push transmission unit is configured to push and transmit the private domain channel content push information set to the user terminal corresponding to the user identification information set according to the private domain channel push scheduling information. The adjustment unit is configured to dynamically adjust the inventory of goods based on the acquired set of goods flow information corresponding to the set of user terminals. The dynamic adjustment includes at least one of the following: multi-warehouse inventory transfer, inventory circuit breaker / freeze processing, and delayed replenishment of goods.

8. An electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

9. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

Images