Off-network user prediction method and device, computer equipment and storage medium

Abstract

The invention relates to an off-network user prediction method and device, computer equipment and a storage medium. Comprising the following steps: in a process that a target user uses an Internet television service by means of an Internet television receiving terminal, obtaining target operation data corresponding to the Internet television receiving terminal; determining target watching behavior data corresponding to the target user based on the target operation data; and inputting the target watching behavior data into a pre-trained off-network behavior prediction model to obtain an off-network probability corresponding to the target user, and when the off-network probability is greater than or equal to a preset threshold, determining the target user as an off-network user. On the basis, the method and the device have the advantages of advanced technologies such as machine learning, the influence of multiple factors on the user off-network behavior can be comprehensively considered, and the accuracy and the reliability of predicting the user off-network behavior are further improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of communication, and particularly relates to an off-network user prediction method and device, a computer device and a storage medium. BACKGROUND

[0002] Internet TV service is an innovative result of deep integration of traditional TV media and Internet technology, which enables users to order programs at any time and obtain customized content recommendation based on personal preferences.

[0003] In the related art, the way of determining whether a user is off-network only according to the shutdown duration of an Internet TV receiving terminal cannot accurately reflect the actual use of Internet TV service by the user, thus easily causing misjudgment and failing to meet the demand for accurate analysis of user behavior in Internet TV service operation. SUMMARY

[0004] Therefore, the embodiments of the present disclosure provide an off-network user prediction method and device, a computer device and a storage medium to solve the problems in the related art.

[0005] In a first aspect, the embodiments of the present disclosure provide an off-network user prediction method, which comprises: obtaining target running data corresponding to an Internet TV receiving terminal in a process in which a target user uses Internet TV service by means of the Internet TV receiving terminal; determining target viewing behavior data corresponding to the target user based on the target running data; inputting the target viewing behavior data into a pre-trained off-network behavior prediction model to obtain an off-network probability corresponding to the target user; and determining that the target user is an off-network user in a case where the off-network probability is greater than or equal to a preset threshold.

[0006] In a second aspect, the embodiments of the present disclosure provide an off-network user prediction device applied to the off-network user prediction method of the first aspect, which comprises: an obtaining module configured to obtain target running data corresponding to an Internet TV receiving terminal in a process in which a target user uses Internet TV service by means of the Internet TV receiving terminal; a first determining module configured to determine target viewing behavior data corresponding to the target user based on the target running data; an obtaining module configured to input the target viewing behavior data into a pre-trained off-network behavior prediction model to obtain an off-network probability corresponding to the target user; and a second determining module configured to determine that the target user is an off-network user in a case where the off-network probability is greater than or equal to a preset threshold.

[0007] In a third aspect, the embodiments of the present disclosure provide a computer device comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the off-network user prediction method.

[0008] In a fourth aspect, the present disclosure provides a computer readable storage medium having stored thereon computer programs / instructions which, when executed by a processor, implement the steps of the off-net user prediction method described above.

[0009] In a fifth aspect, the present disclosure provides a computer program product having computer programs / instructions which, when executed by a processor, implement the steps of the off-net user prediction method described above.

[0010] The above at least one technical solution adopted by the embodiments of the present disclosure can achieve the following beneficial effects: In the embodiments of the present disclosure, target running data corresponding to an Internet TV receiving terminal is obtained during the use of the Internet TV service by a target user via the Internet TV receiving terminal; then target viewing behavior data corresponding to the target user is determined based on the target running data; the target viewing behavior data is then input into a pre-trained off-net behavior prediction model to obtain an off-net probability corresponding to the target user; and finally, the target user is determined to be an off-net user when the off-net probability is greater than or equal to a preset threshold. In this process, by obtaining the target running data of the Internet TV receiving terminal and determining the target viewing behavior data therefrom, the actual situation of the target user using the Internet TV service can be comprehensively and meticulously understood; and finally, the target viewing behavior data is input into the pre-trained off-net behavior prediction model to obtain the off-net probability corresponding to the target user, and then whether the target user is an off-net user can be determined according to the off-net probability corresponding to the target user. Based on this, the embodiments of the present disclosure take advantage of advanced technologies such as machine learning, can comprehensively consider the influence of multiple factors on the off-net behavior of the user, and thus improve the accuracy and reliability of the prediction of the off-net behavior of the user. BRIEF DESCRIPTION OF DRAWINGS

[0011] The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which:

[0012] Figure 1 A schematic diagram of a service architecture of an Internet TV service provided by an embodiment of the present disclosure;

[0013] Figure 2 A flowchart of a user off-net behavior prediction method in related technologies provided by an embodiment of the present disclosure;

[0014] Figure 3 A flowchart of an off-net user prediction method provided by an embodiment of the present disclosure;

[0015] Figure 4 A schematic diagram of a user behavior association graph provided for an embodiment of the present disclosure;

[0016] Figure 5 A flowchart of another off-grid user prediction method provided for an embodiment of the present disclosure;

[0017] Figure 6 A structural diagram of an off-grid user prediction apparatus provided for an embodiment of the present disclosure;

[0018] Figure 7 A structural diagram of an electronic device provided for an embodiment of the present disclosure;

[0019] Figure 8 A structural diagram of a computer system provided for an embodiment of the present disclosure;

[0020] Figure 9 A schematic diagram of a computer program product provided for an embodiment of the present disclosure. DETAILED DESCRIPTION

[0021] Embodiments of the present disclosure will be described in more detail with reference to the drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein, but rather the embodiments are provided so that the present disclosure can be more thoroughly and completely understood. It should be understood that the drawings of the present disclosure are for exemplary purposes only and are not intended to limit the scope of protection of the present disclosure.

[0022] It should be understood that each step recited in the method embodiments of the present disclosure can be executed in different orders and / or in parallel. In addition, the method embodiments can include additional steps and / or omit the execution of the steps shown. The scope of the present disclosure is not limited in this respect.

[0023] The term “comprising” and variations thereof as used herein are open-ended, that is, “including but not limited to”. The term “based on” is “based, at least in part, on”. The term “one embodiment” means “at least one embodiment”; the term “another embodiment” means “at least one additional embodiment”; the term “some embodiments” means “at least some embodiments”. Related definitions are given below. It should be noted that the concepts mentioned in the present disclosure are merely used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0024] It should be noted that the modification of "one" and "multiple" mentioned in the present disclosure is illustrative rather than restrictive, and those skilled in the art should understand that "one or more" should be understood unless the context clearly indicates otherwise.

[0025] The names of the messages or information exchanged between the plurality of devices in the embodiments of the present disclosure are only for illustrative purposes, and are not used to limit the scope of the messages or information.

[0026] Internet TV service is the product of the deep integration of traditional TV media and Internet technology. It relies on Internet infrastructure such as broadband and wireless networks, and uses devices such as smart TVs and Internet TV receiving terminals to deliver rich and diverse multimedia content such as video, audio, and text to user TV screens. Compared with traditional TV, Internet TV service breaks the time and space limit, so that users can not only watch their favorite programs on demand, but also customize content recommendations according to their own preferences.

[0027] Figure 1 A schematic diagram of a service architecture of an Internet TV service according to an embodiment of the present disclosure is provided. As shown in Figure 1 Generally, the broadcast program source 101 includes various types of program resources, and the program resources usually need to be audited by the Internet TV license broadcaster 102 for program content. Only the audited program resources can be transmitted to the source server 103 for storage. In this way, when a user needs to obtain a target program resource, the cache server 104 can obtain the target program resource from the source server 130, and then transmit it to the Internet TV receiving terminal 106 through the IP metropolitan area network 105. The Internet TV receiving terminal 106 can receive the target program resource with the cooperation of the home gateway and other devices and display it on the display device.

[0028] The user can select the program resource to be watched on the display device through the remote control device and play it. In this process, in order to analyze the user's use behavior of the Internet TV service, a soft probe can be implanted in the Internet TV receiving terminal, and then the running data generated by the Internet TV receiving terminal during operation can be collected through the probe. The behavior data generated by the user when using the Internet TV receiving terminal can be determined through the running data, and then the running data can be converted into report data in a specified format. Based on this, the user's use behavior can be analyzed through the report data.

[0029] Figure 2 A flowchart of a user off-network behavior prediction method in related technologies according to an embodiment of the present disclosure is provided. As shown in Figure 2 Specifically, it includes:

[0030] S201, obtaining the shutdown duration of the Internet TV receiving terminal.

[0031] S202 dispatches on-site or telephone care orders when the shutdown duration is greater than or equal to the preset shutdown duration.

[0032] S203 determines whether a user is an offline user based on the results of on-site dispatch or telephone follow-up.

[0033] It is evident that current technologies for determining user churn in internet TV services lack comprehensive and refined data analysis. Simply relying on the length of time the internet TV receiver is switched off fails to accurately reflect a user's specific usage of the service. For example, when a user is away from home for an extended period due to business trips, their frequently used internet TV receiver may remain switched off for extended periods. In such cases, the prolonged shutdown is due to objective reasons and does not necessarily indicate a lack of internet TV service usage; the user may be using other internet TV receivers.

[0034] In recent years, after several years of rapid development, internet TV, especially with the user growth of operator TV services reaching its peak, has shifted its focus from "quantity" to "quality" and from "maintenance" to "operation." The current priority is reducing user churn and increasing user stickiness and service satisfaction. However, judging user behavior solely based on the duration of time the internet TV receiver is off is too lagging. Furthermore, subsequent actions such as dispatching work orders or making phone calls to users with long off-time periods may disrupt their daily lives, further contributing to churn.

[0035] To address the aforementioned problems, this disclosure provides a method for predicting churned users. The method involves acquiring target operational data corresponding to the internet TV receiving terminal during a target user's use of internet TV services via an internet TV receiving terminal; then determining target viewing behavior data corresponding to the target user based on the target operational data; finally, inputting the target viewing behavior data into a pre-trained churned user prediction model to obtain the churn probability corresponding to the target user; and finally, determining the target user as a churned user if the churn probability is greater than or equal to a preset threshold. In this process, by acquiring the target operational data of the internet TV receiving terminal and determining the target viewing behavior data from it, a comprehensive and detailed understanding of the target user's actual use of internet TV services can be obtained. Finally, the target viewing behavior data is input into the pre-trained churned user prediction model to obtain the churn probability corresponding to the target user, thereby determining whether the target user is a churned user based on the churn probability. Based on this, this disclosure leverages the advantages of advanced technologies such as machine learning, comprehensively considering the influence of multiple factors on user churn behavior, thus improving the accuracy and reliability of predicting user churn behavior.

[0036] Figure 3 This is a flowchart illustrating a method for predicting offline users according to an embodiment of this disclosure. Figure 3 As shown, it specifically includes:

[0037] S301, acquire multiple operational data generated during the process of multiple users using Internet TV services through Internet TV receiving terminals. It should be understood that when a user uses Internet TV services, a user can often only use a single Internet TV receiving terminal at a given time.

[0038] In some embodiments, monitoring software can be installed on the Internet TV receiver terminal, which can then monitor various operational data of the Internet TV receiver terminal in real time. This operational data can then be analyzed and processed to obtain user viewing behavior data. It should be understood that in practical applications, the operational status of the Internet TV receiver terminal can also be monitored through other methods, which are not specifically limited here.

[0039] S302, preprocess the operational data corresponding to multiple Internet TV receiving terminals to obtain multiple initial viewing behavior data. The following is a detailed explanation of the preprocessing process, using the operational data generated when user A uses Internet TV service through Internet TV receiving terminal A as an example:

[0040] In some embodiments, during the process of user A using Internet TV receiving terminal A to access Internet TV services, operational data generated by Internet TV receiving terminal A during operation can be collected. This operational data can then be preprocessed to obtain initial viewing behavior data. The preprocessing process may include two parts: data organization and normalization.

[0041] First, the operational data can be cleaned up. Specifically, non-viewing behavior data can be removed from the operational data, that is, data other than user behavior data, service quality data, and user feedback data can be removed.

[0042] The aforementioned user behavior data may include the power-on time of User A's most recent internet TV receiver, as well as the average daily power-on count, average daily power-on duration, average daily playback count, and average daily playback duration within a certain period. It should be understood that this "certain period" can be a week, a month, or a year, whichever is determined based on the actual situation. The average daily playback count and average daily playback duration refer to the number of times the user plays the same or different programs and the corresponding playback duration. The average daily playback duration is the total playback time for the same or different programs throughout the entire day, and the same applies to the average daily playback count. It should be understood that since the internet TV receiver's operational data usually does not directly display the average daily playback count, it needs to be obtained through background calculation. The specific calculation method is as follows: using a week as the statistical period, first, summarize the total number of playbacks within that period, and then divide the total number of playbacks by the number of days in a week to obtain the average daily playback count for that week. The calculation method for other metrics such as the average daily power-on count, average daily power-on duration, and average daily playback duration is the same as for the average daily playback count.

[0043] The aforementioned service quality data may include the EPG request success rate, EPG request response latency, program playback success rate, number of playback stutters, and percentage of playback stutter duration corresponding to the Electronic Program Guide (EPG). Specifically, an EPG request success rate greater than 95% is considered ideal, an EPG request response latency less than 300ms is considered ideal, a program playback success rate greater than 95% is considered ideal, and the number of playback stutters and the percentage of playback stutter duration are considered ideal. Under these ideal conditions, the internet TV service can operate efficiently and stably, providing users with a good experience.

[0044] The aforementioned user feedback data may include the number of complaints and the frequency of complaints. The number of complaints can be the cumulative number of complaints since the user started using the Internet TV service, while the frequency of complaints can be the number of complaints made by the user within a period.

[0045] After removing all data except user behavior data, service quality data, and user feedback data from the runtime data, punctuation marks can be removed from the remaining user behavior data, service quality data, and user feedback data. Then, word segmentation and part-of-speech tagging can be performed on the punctuation-removed user behavior data, service quality data, and user feedback data. Here, the Jieba library in Python can be used for word segmentation, as it offers good results and is easy to use. For part-of-speech tagging after word segmentation, the jieba.posseg module in Python can be used. It should be understood that other methods can also be used for word segmentation and tagging; no specific limitations are made here.

[0046] In some embodiments, Graph Neural Networks (GNNs) possess unique advantages in reflecting relationships between data. Therefore, GNNs can be used to process processed user behavior data, service quality data, and user feedback data, thereby enabling better prediction of user churn behavior for internet TV. Based on this, after processing the operational data, it can also be normalized to ensure that the operational data meets the input requirements of GNNs.

[0047] While the program playback success rate and EPG request success rate in Quality of Service (QoS) data cannot usually be directly obtained from operational data, they can be determined through analysis of the operational data. For example, to determine the program playback success rate, the total number of program playbacks within a period can be counted, and the success or failure of each playback can be recorded. The program playback success rate can be obtained by dividing the number of successful playbacks by the total number of program playbacks. The calculation method for the EPG request success rate can refer to the calculation process for the program playback success rate.

[0048] In some embodiments, user behavior data, service quality data, and user feedback data derived directly from operational data may contain non-numerical data. This non-numerical data can be converted into specific numerical values ​​to facilitate subsequent data processing. For example, the calculation of average daily power-on time is the average power-on time over a period of time. However, in operational data, the data corresponding to a single power-on duration is not a specific numerical value, but rather the start and end times of operation of the internet TV receiving terminal, representing a time period. This type of non-numerical data needs to be converted into a specific numerical value, based on the time difference between the start and end times of operation within a day.

[0049] In summary, the running data can be processed. However, since running data typically encompasses multiple data types, the numerical ranges of these different data types often differ significantly after processing. Therefore, to facilitate subsequent neural network operations, the processed running data can be normalized. For example, a min-max scaler can be used to normalize the processed running data.

[0050] Specifically, MinMaxScaler can scale processed running data to a given minimum and maximum value, typically between [0,1]. Its formula is as follows:

[0051]

[0052] The operational data includes various data types, such as average daily power-on count, average daily power-on duration, average daily playback count, or average daily playback duration. Scaling the operational data requires scaling each data point individually. Taking the average daily power-on count over a period as an example, the characters in the formula are explained as follows: X represents the average daily power-on count; X' represents the scaled average daily power-on count; Xmin represents the minimum average daily power-on count; Xmax represents the maximum average daily power-on count; Rmin represents the minimum target scaling range, usually 0; and Rmax represents the maximum target scaling range, usually 1.

[0053] As mentioned above, each type of data may include multiple data points. For example, for the average number of daily system startups per week within a month, a month usually includes four weeks, and each week corresponds to an average number of daily system startups, that is, a month corresponds to 4 sets of data. For each set of data, the corresponding values ​​need to be mapped to the range between Rmin and Rmax according to the above formula. Generally, Rmin is 0 and Rmax is 1, that is, the values ​​corresponding to the normalized data are all between [0,1].

[0054] Based on this, multiple initial viewing behavior data can be obtained by preprocessing the runtime data.

[0055] S303, construct a user behavior association graph based on multiple initial viewing behavior data.

[0056] In some embodiments, the user behavior graph can be a heterogeneous graph. Nodes in the heterogeneous graph can include user nodes, program nodes, or Internet TV receiving terminal nodes, and the connections between nodes can be used to represent user viewing behavior. For example, when a user's viewing behavior is turning on the Internet TV receiving terminal, the corresponding initial viewing behavior data can be the power-on time when the Internet TV receiving terminal is turned on, etc. Or, when a user's viewing behavior is a complaint behavior, the corresponding initial viewing behavior data can be the number of complaints, etc.

[0057] Figure 4 This is a schematic diagram of a user behavior association graph provided in one embodiment of this disclosure. For example... Figure 4 As shown, taking user A as an example, the corresponding meta-path can be "User A - (Watch) - Program A". Nodes can include user A nodes and program A nodes, and the connections between nodes, i.e., edges, represent the watching behavior. Here, a meta-path can include multiple edges or just one edge. Figure 4 As shown, when a metapath includes multiple edges, namely L1, L2, and L3, L1 can represent the viewing time of user A for program A, L2 can represent user A's complaint behavior while watching program A, and L3 can represent user A's satisfaction while watching program A. Satisfaction can be determined based on user behavior data, service quality data, and user feedback data. For example, if user behavior data shows that user A frequently uses the internet TV service to watch program A, each time watching for more than 8 hours, and there are no playback interruptions or complaints during viewing (i.e., zero complaints), then user satisfaction with watching program A can be considered high. It should be understood that a metapath can be defined as a single behavioral trajectory of user A watching program A. When user A watches program A again, a new metapath is generated, recording this viewing behavior.

[0058] In some embodiments, when determining satisfaction based on user behavior data, service quality data, and user feedback data, the following example illustrates the process: user behavior data includes the power-on time corresponding to the most recent time the internet TV receiving terminal was turned on; service quality data includes the EPG request success rate; and service quality data includes the number of complaints. Specifically, when the time difference between the power-on time corresponding to the most recent time user A turned on the internet TV receiving terminal and the time of obtaining the running data is less than 7 days, the user is determined to be an online user, i.e., a user who is normally using the internet TV service, which is 100%. At this time, the user behavior data can be assigned 1. The EPG request success rate is 96%, and the number of complaints is 0. When the number of complaints is 0, user A can be considered to be very satisfied with the current service, which is 100%. At this time, the service quality data is assigned 1. The satisfaction rate is 100% + 96% + 100% = 296%. Since only three data items are considered, the total satisfaction rate is 300%. The difference between 296% and 300% is very small. At this time, user A can be considered to be very satisfied with the current service.

[0059] In some embodiments, there can be multiple edges between two adjacent nodes, meaning there are multiple viewing behaviors. Simultaneously, a node can have multiple adjacent nodes, and each pair of adjacent nodes can form a subgraph. Directly connected nodes are defined as first-order nodes, nodes separated by one node are called second-order nodes, and so on. Based on this, the node v corresponding to user A can be generated according to the meta-path. i X-order subgraph g p i,k The set of nodes of multiple subgraphs is V. p i,k The edge set is E p i,k Furthermore, different subgraphs g can be generated through different metapaths. p1 i,k g p2 i,k , ..., g pl i,k Where p represents the metapath identifier, I is the number of metapaths, and the corresponding meaning is the total number of times the user watched the program.

[0060] User A corresponds to user node v i X-order neighbor subgraph g based on metapath p p i,k Node set The specific value is:

[0061]

[0062] Where i represents the identifier of the data segment from which user A registered to join the network, k represents the identifier of the data segment from which user A stopped watching, R represents the program playback success rate, R1 represents the number of complaints from user A, L represents the weight of user satisfaction, and Vall represents the user satisfaction value.

[0063] In some embodiments, E i,k Let represent the set of edges, symbolizing sticky viewing power as represented by internet TV programs, package value, and viewing time. The specific calculation method is as follows:

[0064]

[0065] Wherein, for program A, Y is a D×N matrix representing the content of program A watched by N users, Z is the viewing behavior matrix corresponding to user A, δ is the regularization parameter, K represents the popularity of program A among equivalent subscription packages, ||||F represents the Frobenius norm, and Tr represents the trace of the matrix. Let T represent the Laplace matrix, which represents the similarity between users, and T is the length of the time period.

[0066] This part represents the difference between the original data matrix Y and the data matrix YZ reconstructed from the latent matrix Z. The goal is to minimize this difference so that the reconstructed data matrix is ​​as close as possible to the original data matrix.

[0067] This part is a regularization term used to penalize the elements in matrix Z, making them as close as possible to the Laplace matrix. The elements in the matrix. Laplace matrix. This regularization term represents the similarity between users, so it helps to maintain the similarity between users.

[0068] Based on this, by minimizing the objective function, we can obtain a viewing behavior matrix Z for user A, thereby better understanding and predicting user A's viewing behavior.

[0069] The following formula can be used to merge all subgraphs corresponding to user A, resulting in a set of subgraphs corresponding to user A:

[0070] V i,k =(V p1 i,k ∪V p2 i,k ∪…∪V pl i,k )*e1

[0071] E i,k =(E p1 i,k∪E p2 i,k ∪…∪E pl i,k )*e2

[0072] In the formula, V p1 i,k V represents the set of nodes in the metapath p1; p2 i,k V represents the set of nodes in the metapath p2; pI i,k E represents the set of nodes in the metapath pI; p1 i,k E represents the set of nodes in the metapath p1; p2 i,k E represents the set of nodes in the metapath p2; pI i,k V represents the set of nodes in the metapath pI; i,k E represents the set of nodes in all metapaths; i,k This represents the set of edges in all metapaths. e1 and e2 represent correction deviations, which have a self-adjusting function. The specific identification method is as follows:

[0073]

[0074] Where Ke represents the characteristic parameters of different user groups, such as the product of average viewing time and the coefficient of preferred program type, which can be used to distinguish different user groups; ρ is the standard deviation of the Gaussian distribution, which controls the degree of matching between energy e and user feature Ke. If ρ is small, it means that more attention is paid to the energy value that closely matches the user feature, and if ρ is large, it means that a wider range of energy values ​​can be invested; s represents user behavior characteristics, corresponding to user behavior data, service quality data, and user feedback data; cv represents cross-validation parameters or context information; e represents energy, which is used to measure the indicator in the process of user interaction with Internet TV service; and K represents the identifier used to identify different user groups.

[0075] In some embodiments, while generating a high-order neighbor subgraph for the target user, it is necessary to calculate the importance of each node in the subgraph, i.e., the weight corresponding to different subgraphs. This weight can be determined based on V. i,k The frequency of a node's appearance is determined by its relative importance; a node is considered important if it appears frequently, and unimportant if it appears infrequently. For example, a node can be a program node; when it appears in V... i,k If the number of occurrences of the node corresponding to program A accounts for 60% of the total number of nodes, then program A is considered important, and the weight corresponding to node A can be set higher. When V i,kIf the number of times the node corresponding to program A appears accounts for 5% of the total number of nodes, then program A is considered unimportant, and the weight corresponding to node A can be set to a lower value.

[0076] For example, when node v j Only in V p1 i,k If it appears once, then v j 1 point; if v j In V p1 i,k V p2 i,k , ..., V pI i,k If it has appeared in both, then v j To get I points, use sem(v) j ) to represent node v j If the semantic representation is:

[0077]

[0078] At this point, s′ represents the neighbor node identifier. When s′ = 1, the corresponding node is determined to be a neighbor node; when s′ = 0, the corresponding node is determined not to be a neighbor node. j This indicates the node identified as j; if a node has a high semantic representativeness score and its neighbors also have high semantic representativeness scores, then the node contains richer information and is more important.

[0079] Specifically, node v in the X-order neighbor subgraph j Importance score s j for:

[0080]

[0081] In the formula, N(v) j ) is node v j direct neighbor node, m(v t The calculation formula for ) is as follows:

[0082]

[0083] In the formula, x t It is node v t The input features, i.e., the feature matrix X i,k The t-th row, m∈R d It is a learnable vector that represents x. t Projected onto m(v) t ).

[0084] Based on this, an importance score s can be selected according to a preset importance score threshold. jNodes with importance scores greater than or equal to the importance score threshold are considered as neighboring nodes of user A's corresponding node.

[0085] S304, Construct an churn behavior prediction model based on the user behavior association graph.

[0086] In some embodiments, for a given user A corresponding to node A, a graph convolutional layer can aggregate information from the neighboring nodes of node A to update the feature threshold of node A, generating an updated feature threshold for node A. The formula for the graph convolutional layer H is as follows:

[0087]

[0088] in, This represents the feature update threshold for node i at layer r, where r represents the layer number, i.e., the current iteration number or time step. This represents the updated feature threshold of node j at layer r; c represents the set of neighboring nodes of node i; ij The normalization coefficient is typically taken as the square root of the node's value to ensure a balance of contributions from different nodes. The normalization coefficient represents some contextual information, weight, or coefficient related to the node identified as i and its neighboring nodes identified as j; W (r) Let represent the weight matrix of the r-th layer, and let represent how this matrix changes with the number of iterations. Let σ represent the feature threshold of neighbor node j at layer r, and σ represent the activation function (such as ReLU). This is a residual connection.

[0089] In some embodiments, the updated feature threshold corresponding to a node is the local feature. Then, an aggregation function can be used to aggregate the local features of multiple nodes into a global feature. Commonly used aggregation functions include mean, sum, and max. Here, the mean aggregation function is selected to calculate the global feature of a node:

[0090] The formula for the average aggregation function is as follows:

[0091]

[0092] in, Represents the global features of node i; H represents the local feature of neighbor node j at the k-th layer; mean(·) represents the mean operation, which takes the average of the features in the set; j This represents the update feature threshold for node j.

[0093] In some embodiments, the output layer of the off-grid behavior prediction model may include a fully connected layer and a softmax classifier. The aggregated global features are passed through a fully connected layer and then fed into the softmax classifier to obtain the predicted off-grid probability. The aforementioned fully connected layer refers to a dense layer or linear layer to perform a linear transformation on the input data. The fully connected layer applies a weight to each global feature of the input, sums these weighted feature values, and adds a bias term to obtain an output value. This output value is a weighted combination of the input features and can capture important information from the input data.

[0094] The formula for a fully connected layer is as follows:

[0095]

[0096] Where z represents the output value of the fully connected layer; b represents the weight matrix corresponding to the nodes in the fully connected layer. fc This represents the bias threshold of the fully connected layer; σ represents the global features of a node; σ represents the activation function (such as Sigmoid, ReLU).

[0097] After obtaining the output value of the fully connected layer, the output value can also be input into the softmax classifier to convert the output value of the off-grid behavior prediction model into the prediction probability, i.e., the off-grid probability.

[0098] Based on this, when constructing a user behavior association graph, this application can indirectly reflect the weight of viewing behavior data by calculating the importance of nodes. For example, calculating node v j Semantic representativeness score sem(v j If a node appears in multiple subgraphs generated based on different metapaths, it scores higher, indicating that the node contains richer information and is more important. Meanwhile, node v in the X-order neighbor subgraph... j Importance score s j The calculation using the sigmoid function involves the semantic representativeness scores of neighboring nodes and input features. This series of calculations reflects the weights corresponding to different behavioral data. In the graph convolutional layer, the weight matrix W(k) changes with the number of iterations k. When aggregating neighboring node information to update the target node feature threshold, it weights the feature information of different neighboring nodes, affecting the contribution of each behavioral data to the node feature update, and thus determining the weight of the viewing behavior data.

[0099] Furthermore, in determining the churn probability, the churn behavior prediction model can update node feature thresholds through graph convolutional layers, then obtain global features using aggregation functions. These global features are then processed by fully connected layers and a softmax classifier to output the churn probability. Throughout this process, the previously determined viewing behavior data weights are integrated into each stage of node feature updating, global feature calculation, and final probability output, thus allowing for the comprehensive determination of the user's churn probability based on the churn probability and its weights.

[0100] In some embodiments, weights can be assigned to user behavior data, service quality data, and user feedback data. Specifically, the weights can be determined based on the impact of different data on user churn behavior. For example, taking the number of complaints in user feedback data as an example, if a user has accumulated 0-3 complaints since starting to use the internet TV service, the user is considered not to be churning, and the corresponding weight can be 0; if the accumulated number of complaints is 4-7, the user is considered a potential churner, and the corresponding weight can be 1; if the accumulated number of complaints is 8-10, the user is considered to have a high churn trend, and the corresponding weight can be 2. Based on this, multiple data points included in user behavior data, service quality data, and user feedback data can be categorized and weighted. Furthermore, by comprehensively analyzing multiple data points, the probability of a user churning can be determined.

[0101] Based on this, the embodiments of this disclosure can acquire target operational data corresponding to the Internet TV receiving terminal during the process of a target user using Internet TV services through an Internet TV receiving terminal; then, determine target viewing behavior data corresponding to the target user based on the target operational data; then, input the target viewing behavior data into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user; finally, if the churn probability is greater than or equal to a preset threshold, the target user is determined to be a churn user. In this process, by acquiring the target operational data of the Internet TV receiving terminal and determining the target viewing behavior data from it, a comprehensive and detailed understanding of the actual situation of the target user's use of Internet TV services can be obtained. Finally, the target viewing behavior data is input into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user, and thus, whether the target user is a churn user can be determined based on the churn probability. Based on this, the embodiments of this disclosure leverage the advantages of advanced technologies such as machine learning, and can comprehensively consider the influence of multiple factors on user churn behavior, thereby improving the accuracy and reliability of predicting user churn behavior.

[0102] The off-grid user prediction method provided in this disclosure can be executed by a terminal or by a chip applied to the terminal.

[0103] For example, the aforementioned terminals may include one or more of the following: mobile phones, tablets, wearable devices, in-vehicle devices, laptops, ultra-mobile personal computers (UMPCs), netbooks, handheld computers (PDAs), and wearable devices based on augmented reality (AR) and / or virtual reality (VR) technologies. They may also include, but are not limited to, remote control devices, wearable devices, streetlights, home appliances, and other smart terminals. This disclosure does not impose specific limitations on these aspects.

[0104] Figure 5 This is a flowchart illustrating another method for predicting offline users provided in an embodiment of this disclosure. Figure 5 As shown, it specifically includes:

[0105] S501: During the process of a target user using an Internet TV service through an Internet TV receiving terminal, obtain the target operation data corresponding to the Internet TV receiving terminal.

[0106] In some embodiments, monitoring software can be installed on the Internet TV receiving terminal. This software can then monitor various target operating data of the Internet TV receiving terminal in real time. The target operating data can then be analyzed and processed to obtain user viewing behavior data. It should be understood that in practical applications, the operating status of the Internet TV receiving terminal can also be monitored through other methods, which are not specifically limited here.

[0107] S502, determine the target user's target viewing behavior data based on the target operation data.

[0108] In some embodiments, the target operational data includes not only viewing behavior data, such as user behavior data, service quality data, and user feedback data, but also non-viewing behavior data, such as device operating status and data related to business operations. Therefore, after obtaining the target operational data, it is necessary to filter the content of the target operational data, that is, to remove non-viewing behavior data from the target operational data and obtain target viewing behavior data that is only related to the viewing behavior of the target user.

[0109] Specifically, target viewing behavior data can include target user behavior data, target service quality data, and target user feedback data. Target user behavior data can include the power-on time of the target user's most recent internet TV receiver, as well as the average daily power-on frequency, average daily power-on duration, average daily playback frequency, and average daily playback duration over a certain period. Target service quality data can include the EPG request success rate, EPG request response latency, program playback success rate, number of playback stutters, and percentage of playback stutter durations corresponding to the Electronic Program Guide (EPG). Target user feedback data can include the number of complaints and complaint frequency.

[0110] S503 inputs the target viewing behavior data into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user.

[0111] In some embodiments, inputting target viewing behavior data into a pre-trained churn prediction model to obtain the churn probability of the target user allows for more detailed analysis of the target viewing behavior data, thereby improving prediction accuracy. Furthermore, this helps operators accurately locate potential churn users, rationally allocate human and material resources, avoid ineffective door-to-door visits and telephone follow-ups, and reduce operating costs. Simultaneously, early intervention based on churn probability can effectively resolve user issues, improve user experience, enhance user stickiness, and reduce user churn. In addition, churn probability can provide a basis for business optimization, helping operators adjust strategies, promote sustainable business development, and gain a competitive advantage.

[0112] S504: If the probability of leaving the network is greater than or equal to a preset threshold, the target user is identified as a user who has left the network.

[0113] In some embodiments, a preset threshold can be set. After obtaining the churn probability of a target user, the churn probability can be compared with the preset threshold. When the churn probability of a target user is greater than or equal to the preset threshold, the target user can be determined to be a churned user. Here, since the preset threshold is only used to determine whether a user is already a churned user, to further analyze the user's churn status, the preset threshold can be considered a churn preset threshold. A potential churn preset threshold can also be set, which is less than the churn preset threshold. Therefore, when the user's corresponding churn probability falls between the potential churn preset threshold and the churn preset threshold, the user can be determined to be a potential churned user. A potential churned user can be a user with a high probability of ceasing to use the internet TV service in the future.

[0114] The following explanation uses target viewing behavior data, including the power-on time, playback stutters, and complaint counts corresponding to the most recent time the internet TV receiving terminal was turned on, as an example to illustrate the calculation of the probability of churn.

[0115] Table 1. Comparison of Off-Grid Probability

[0116] Off-net probability threshold Time since last boot Number of playback stalls Number of complaints 35％ < 7 days < 5 times < 3 times 70％ 7 days < 15 days 5 times < 10 times 3 times < 6 times 100％ > 15 days > 10 times > 6 times

[0117] Referring to Table 1, when the churn probability ∈ (0, 35%), the user is considered a normal user with no risk of churn; when the churn probability ∈ (35%, 70%), the user is considered a user with mild churn behavior, with a low risk of churn, and no further action is needed for the time being; when the churn probability ∈ (70%, 100%), the user is considered a potential churner, meaning there is a high probability of churn; when the churn probability ∈ (100%, +∞), the user is considered a churner, meaning churn has already occurred. It should be understood that this analysis of user churn behavior only considers three data points. The overall user churn probability should be the sum of the probabilities corresponding to the three data points. Therefore, the corresponding churn probability threshold should also be the sum of the churn probabilities corresponding to the three data points. For example, the churn probability threshold could be 300%, and the potential churn probability threshold could be 210%. Furthermore, the specific values ​​in Table 1 are only examples and not specific limitations, and can be modified according to actual circumstances.

[0118] For example, referring to Table 1, when the time difference between the user's most recent power-on time and the time of data acquisition is 14 days, the corresponding churn probability is 70%; when the number of playback stutters is 7, the corresponding churn probability is 80%; and when the number of complaints is 5, the corresponding churn probability is 90%. In this case, the total churn probability is 70% + 80% + 90% = 240%. This 240% is greater than the potential churn probability threshold but less than the actual churn probability threshold, therefore the user can be considered a potential churner. It should be understood that the churn probability corresponding to each data point can be determined based on the actual situation. For example, for the churn probability based on the number of complaints, 6 complaints are considered the limit; that is, when there are 6 complaints, the user's churn probability is considered to be 100%. The churn probability corresponding to 5 complaints is 84%, and the probability corresponding to 4 complaints is 68%, and so on.

[0119] This embodiment of the disclosure acquires target operational data corresponding to the Internet TV receiving terminal during the process of a target user using Internet TV services via an Internet TV receiving terminal; then, based on the target operational data, it determines the target viewing behavior data corresponding to the target user; next, it inputs the target viewing behavior data into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user; finally, if the churn probability is greater than or equal to a preset threshold, the target user is determined to be a churn user. Based on this, by acquiring the target operational data of the Internet TV receiving terminal and determining the target viewing behavior data from it, a comprehensive and detailed understanding of the actual situation of the target user's use of Internet TV services can be obtained. Finally, the target viewing behavior data is input into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user, and the churn probability is used to determine whether the target user is a churn user. In this process, this embodiment of the disclosure leverages the advantages of advanced technologies such as machine learning, which can comprehensively consider the influence of multiple factors on user churn behavior, thereby improving the accuracy and reliability of predicting user churn behavior.

[0120] In some embodiments, inputting target viewing behavior data into a pre-trained churn behavior prediction model includes: constructing a viewing behavior database, wherein the viewing behavior database includes multiple viewing behavior data generated by multiple users using Internet TV services through multiple Internet TV receiving terminals; constructing a user behavior association graph based on the viewing behavior database; training a preset model using the user behavior association graph to obtain a churn behavior prediction model; and inputting target viewing behavior data into the churn behavior prediction model to obtain the churn probability corresponding to the target user.

[0121] In some embodiments, viewing behavior data generated by multiple users using Internet TV services through different Internet TV receiving terminals can be collected extensively through various methods. After collection, the data is organized, classified, and stored to construct a comprehensive and orderly viewing behavior database. Then, the data in the viewing behavior database is transformed into a heterogeneous graph format.

[0122] Specifically, users and programs can be treated as nodes, and viewing behavior as edges. Viewing behavior includes various actions, such as average daily play count and number of complaints. Then, using the node corresponding to the target user as the center, a user behavior association graph is constructed based on the meta-path (e.g., user-(viewing)-program). This means that the node association path corresponding to each node can be determined from multiple association paths. A user behavior association subgraph is generated based on the node association path corresponding to each node. Finally, the user behavior association graph is constructed based on the user behavior association subgraph corresponding to each node.

[0123] For example, for any given user node, we can start with that user node and determine the node association paths corresponding to it from multiple association paths. These are the association paths containing the user node. Finally, we can integrate these multiple association paths to obtain the user behavior association subgraph corresponding to that user node. The overall user behavior association graph can be obtained by integrating the user association subgraphs corresponding to different users.

[0124] Based on this, a database containing a large amount of user viewing behavior data is constructed, which can comprehensively cover the behavioral characteristics of different users in various scenarios, providing rich materials for subsequent analysis. The user behavior correlation graph constructed based on this database can effectively capture complex behavioral relationships between users, such as similar viewing preferences and similar feedback patterns on service issues. Finally, the churn behavior prediction model obtained by training a preset model using this graph can deeply learn the potential connection between user behavior and churn probability, greatly improving the accuracy of prediction. This further helps operators to identify potential churn users in advance, formulate personalized strategies for users with different risk levels, such as providing exclusive discounts and optimizing service experience for high-risk users, improving user retention rates, rationally allocating resources, reducing operating costs, enhancing market competitiveness, and promoting the sustainable development of the Internet TV business.

[0125] In some embodiments, generating a user behavior association subgraph corresponding to each node based on the node association path corresponding to each node includes: for any node, if there are multiple association paths for any node, determining candidate nodes other than the node itself on each of the node association paths; determining the candidate node as a target node when the number of node association paths to which the candidate node belongs is greater than or equal to a preset number; and generating the user behavior association graph corresponding to the node based on the node association paths containing the node and the target node. It should be understood that the association path here is the same as the meta-path mentioned above.

[0126] Specifically, multiple related paths can be identified first, with a preset number set. This preset number can be determined based on actual business needs and data analysis experience. Based on this, for any given node, other candidate nodes besides the node itself can be identified on one or more related paths. When the number of related paths belonging to a candidate node is greater than or equal to the preset number, that candidate node is designated as the target node. For example, with a preset number of 3, if a program node has 3 or more related paths, then that program node will be identified as the target node. Therefore, this filtering method can highlight nodes that have a high influence and relevance in user behavior relationships.

[0127] Finally, from the numerous associated paths, we can select those that contain both user nodes and target nodes. Based on these selected paths, we integrate and draw the nodes and edges connecting them to form a user behavior association graph. In this graph, user nodes and target nodes serve as key nodes, and the connections between them intuitively demonstrate the important relationships between user behaviors, providing a clear structural basis for subsequent analysis and model training.

[0128] In some embodiments, the method for determining target viewing behavior data corresponding to a target user based on target running data further includes: organizing the target running data; and normalizing the organized target running data to obtain target viewing behavior data.

[0129] Specifically, this involves removing non-viewing behavior data from the target operational data, i.e., removing all data except user behavior data, service quality data, and user feedback data. After removing this data, punctuation marks can be removed from the remaining user behavior data, service quality data, and user feedback data. Then, the remaining user behavior data, service quality data, and user feedback data can be processed with word segmentation and part-of-speech tagging. Here, the Jieba library in Python can be used for word segmentation, as it offers good results and is easy to use. For part-of-speech tagging, the jieba.posseg module in Python can be used. It should be understood that other methods can also be used for word segmentation and tagging; no specific limitations are specified here.

[0130] In practical applications, target runtime data typically encompasses multiple data types, and after data processing, the numerical ranges of these different data types often exhibit significant differences. Therefore, to facilitate subsequent neural network operations, the processed target runtime data can be normalized. For example, a min-max scaler can be used to normalize the processed target runtime data.

[0131] Specifically, MinMaxScaler can scale the cleaned target running data to a given minimum and maximum value, typically between [0,1]. Its formula is as follows:

[0132]

[0133] Since the target operational data includes various data types, such as average daily power-on count, average daily power-on duration, average daily playback count, or average daily playback duration, scaling the target operational data requires scaling each individual data point. Taking the average daily power-on count within a period as an example, X represents the average daily power-on count; X' represents the scaled average daily power-on count; Xmin represents the minimum average daily power-on count; Xmax represents the maximum average daily power-on count; Rmin represents the minimum value of the target scaling range, typically 0; and Rmax represents the maximum value of the target scaling range, typically 1.

[0134] As mentioned above, each type of data may include multiple data points. For example, for the average number of daily system startups per week within a month, a month usually includes four weeks, and each week corresponds to an average number of daily system startups, that is, a month corresponds to 4 sets of data. For each set of data, the corresponding values ​​need to be mapped to the range between Rmin and Rmax according to the above formula. Generally, Rmin is 0 and Rmax is 1, that is, the values ​​corresponding to the normalized data are all between [0,1].

[0135] Based on this, target viewing behavior data can be obtained by preprocessing the target's operational data.

[0136] As can be seen, this embodiment of the disclosure acquires target operational data corresponding to the Internet TV receiving terminal during the process of a target user using the Internet TV service via an Internet TV receiving terminal; then, based on the target operational data, it determines the target viewing behavior data corresponding to the target user; next, it inputs the target viewing behavior data into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user; finally, if the churn probability is greater than or equal to a preset threshold, the target user is determined to be a churn user. In this process, by acquiring the target operational data of the Internet TV receiving terminal and determining the target viewing behavior data from it, a comprehensive and detailed understanding of the actual situation of the target user's use of the Internet TV service can be obtained. Finally, the target viewing behavior data is input into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user, and thus, the churn probability can be used to determine whether the target user is a churn user. Based on this, this embodiment of the disclosure leverages the advantages of advanced technologies such as machine learning, and can comprehensively consider the influence of multiple factors on user churn behavior, thereby improving the accuracy and reliability of predicting user churn behavior.

[0137] The foregoing primarily describes the solutions provided by the embodiments of this disclosure from the perspective of the server. It is understood that, in order to implement the above functions, the server includes the corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this disclosure can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0138] This disclosure embodiment can divide the server into functional units according to the above method example. For example, it can divide each function into separate functional modules, or it can integrate two or more functions into one management module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this disclosure embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0139] By dividing each functional module according to its corresponding function, an exemplary embodiment of this disclosure provides an off-grid user prediction device, which can be a server or a chip applied to a server. Figure 6 This is a schematic diagram of an off-grid user prediction device provided in one embodiment of the present disclosure. Figure 6 As shown, the churn prediction device 600 includes:

[0140] The acquisition module 601 is used to acquire target operation data corresponding to the Internet TV receiving terminal during the process of a target user using Internet TV services through an Internet TV receiving terminal.

[0141] The first determining module 602 is used to determine the target viewing behavior data corresponding to the target user based on the target running data;

[0142] The module 603 is used to input the target viewing behavior data into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user.

[0143] The second determining module 604 is used to determine the target user as an offline user when the offline probability is greater than or equal to a preset threshold.

[0144] In one alternative approach, the target viewing behavior data includes target user behavior data, target service quality data, and target user feedback data.

[0145] In an optional embodiment, the obtaining module 603 is further configured to construct a viewing behavior database, wherein the viewing behavior database includes multiple viewing behavior data generated by multiple users using Internet TV services through multiple Internet TV receiving terminals; construct a user behavior association graph based on the viewing behavior database; train a preset model using the user behavior association graph to obtain the churn behavior prediction model; and input the target viewing behavior data into the churn behavior prediction model to obtain the churn probability corresponding to the target user.

[0146] In an optional manner, the obtaining module 603 is further configured to determine the node association path corresponding to each node from multiple association paths; generate a user behavior association subgraph corresponding to each node based on the node association path corresponding to each node; and construct the user behavior association graph based on the user behavior association subgraph corresponding to each node.

[0147] In an optional manner, the obtaining module 603 is further configured to, for any given node, if there are multiple node association paths corresponding to the given node, determine candidate nodes other than the given node on each of the node association paths; when the number of node association paths to which the candidate node belongs is greater than or equal to a preset number, determine the candidate node as a target node; and generate the user behavior association graph corresponding to the given node based on the node association paths containing the given node and the target node.

[0148] In one alternative embodiment, the off-grid user prediction device 600 further includes a processing module 605, which is used to organize the target operation data and normalize the organized target operation data to obtain the target viewing behavior data.

[0149] This disclosure also provides an electronic device, including: at least one processor; a memory for storing at least one processor-executable instruction; wherein the at least one processor is used to execute the instruction to implement the steps of the method disclosed in this disclosure.

[0150] Figure 7 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present disclosure. Figure 7 As shown, the electronic device 700 includes at least one processor 701 and a memory 702 coupled to the processor 701. The processor 701 can perform the corresponding steps in the methods disclosed in the embodiments of this disclosure.

[0151] The processor 701 described above can also be referred to as a Central Processing Unit (CPU), which can be an integrated circuit chip with signal processing capabilities. Each step in the method disclosed in this embodiment can be implemented by the integrated logic circuitry in the processor 701's hardware or by software instructions. The processor 701 can be a general-purpose processor, a digital signal processor (DSP), an ASIC, a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this embodiment can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can be located in the memory 702, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor 701 reads information from the memory 702 and, in conjunction with its hardware, completes the steps of the above method.

[0152] Furthermore, various operations / processes according to this disclosure, implemented via software and / or firmware, can be transmitted from a storage medium or network to a computer system with a dedicated hardware architecture, for example, Figure 8 The computer system 800 shown is equipped with the programs that constitute the software. When various programs are installed, the computer system is able to perform various functions, including those mentioned above. Figure 8 This is a schematic diagram of the structure of a computer system provided in an embodiment of the present disclosure.

[0153] Computer system 800 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0154] like Figure 8As shown, the computer system 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. The RAM 803 may also store various programs and data required for the operation of the computer system 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0155] Multiple components in the computer system 800 are connected to the I / O interface 805, including: an input unit 806, an output unit 807, a storage unit 808, and a communication unit 809. The input unit 806 can be any type of device capable of inputting information into the computer system 800. The input unit 806 can receive input numerical or character information and generate key signal inputs related to user settings and / or function control of the electronic device. The output unit 807 can be any type of device capable of presenting information and may include, but is not limited to, a monitor, speaker, video / audio output terminal, vibrator, and / or printer. The storage unit 808 may include, but is not limited to, a hard disk and an optical disk. The communication unit 809 allows the computer system 800 to exchange information / data with other devices via a network such as the Internet, and may include, but is not limited to, a modem, network card, infrared communication device, wireless communication transceiver, and / or chipset, such as Bluetooth™ device, WiFi device, WiMax device, cellular communication device, and / or the like.

[0156] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above. For example, in some embodiments, the methods disclosed in this disclosure can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on an electronic device via ROM 802 and / or communication unit 809. In some embodiments, the computing unit 801 can be configured to perform the methods disclosed in this disclosure by any other suitable means (e.g., by means of firmware).

[0157] This disclosure also provides a computer-readable storage medium, wherein when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform the methods disclosed in this disclosure.

[0158] The computer-readable storage medium in this disclosure can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. The aforementioned computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specifically, the aforementioned computer-readable storage medium may include electrical connections based on 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 devices, magnetic storage devices, or any suitable combination of the foregoing.

[0159] 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.

[0160] Figure 9 This is a schematic diagram of a computer program product provided according to an embodiment of the present disclosure. Figure 9 As shown, the computer program product 900 includes a computer program 901, which, when executed by a processor, implements the methods disclosed in the embodiments of this disclosure.

[0161] In embodiments of this disclosure, computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as 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 it can be connected to an external computer.

[0162] 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.

[0163] The modules, components, or units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the modules, components, or units do not necessarily constitute a limitation on the module, component, or unit itself.

[0164] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary hardware logic components that can be used 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.

[0165] The above description is merely an embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope 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 concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0166] While specific embodiments of this disclosure have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A method for predicting churned users, characterized in that, include: During the process of a target user using an Internet TV receiving terminal to access Internet TV services, target operation data corresponding to the Internet TV receiving terminal is obtained. Based on the target operational data, determine the target user's target viewing behavior data; The target viewing behavior data is input into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user. If the probability of leaving the network is greater than or equal to a preset threshold, the target user is determined to be a user who has left the network.

2. The method according to claim 1, characterized in that, The target viewing behavior data includes target user behavior data, target service quality data, and target user feedback data.

3. The method according to claim 1, characterized in that, The step of inputting the target viewing behavior data into a pre-trained offline behavior prediction model includes: Construct a viewing behavior database, wherein the viewing behavior database includes multiple viewing behavior data generated by multiple users using Internet TV services through multiple Internet TV receiving terminals; Construct a user behavior association graph based on the viewing behavior database; The preset model is trained using the user behavior association graph to obtain the churn behavior prediction model; The target viewing behavior data is input into the churn behavior prediction model to obtain the churn probability corresponding to the target user.

4. The method according to claim 3, characterized in that, The user behavior association graph includes multiple nodes and multiple association paths, and each association path contains at least two nodes. Constructing the user behavior association graph based on the viewing behavior database includes: Determine the node association path corresponding to each node from multiple association paths; Generate a user behavior association subgraph corresponding to each node based on the node association path corresponding to each node; The user behavior association graph is constructed based on the user behavior association subgraph corresponding to each node.

5. The method according to claim 4, characterized in that, The step of generating a user behavior association subgraph corresponding to each node based on the node association path corresponding to each node includes: For any given node, if there are multiple associated paths for that node, determine the candidate nodes other than the given node on each of the associated paths. When the number of associated paths of the candidate node is greater than or equal to a preset number, the candidate node is determined as the target node. Based on the node association path containing either node and the target node, generate the user behavior association graph corresponding to either node.

6. The method according to claim 1, characterized in that, The method for determining target viewing behavior data corresponding to a target user based on the target operational data further includes: The target operational data is processed; The target operation data after data processing is normalized to obtain the target viewing behavior data.

7. An off-grid user prediction device, characterized in that, include: The acquisition module is used to acquire target operating data corresponding to the Internet TV receiving terminal during the process of a target user using Internet TV services through an Internet TV receiving terminal; The first determining module is used to determine the target viewing behavior data corresponding to the target user based on the target running data; The acquisition module is used to input the target viewing behavior data into a pre-trained churn behavior prediction model to obtain the churn probability corresponding to the target user. The second determining module is used to determine that the target user is an off-network user when the off-network probability is greater than or equal to a preset threshold.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 6.

Images