Information cascade sequence prediction method and device, equipment and medium

Abstract

The invention relates to the technical field of information processing, in particular to an information cascade sequence prediction method and device, equipment and a medium, and the method comprises the steps: constructing multi-level data association through extracting a propagation track of target information, a historical behavior track of a user and a diffusion track of associated information; then, the steps of updating the information influence and updating the user influence are executed through iteration, so that the information spreading value and the user spreading capability are mutually enhanced in a loop: in each round of iteration, the information spreading potential is re-evaluated based on the updated user influence, and meanwhile, the user spreading efficiency is re-evaluated based on the updated information influence; and a continuously optimized feedback loop is formed. The problem of information loss of traditional one-way modeling is effectively solved, dynamic interaction of propagation content features, propagator attribute features and network structure features can be captured at the same time, and the information propagation prediction effect is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, and in particular to an information cascade sequence prediction method, device, equipment and medium. BACKGROUND

[0002] Information cascade sequence prediction refers to predicting the popularity of an information (such as a microblog, news, or academic paper) in a social network or a citation network. This technology is the core of public opinion monitoring, popular content discovery, and influence evaluation, and accurate prediction can provide strong decision-making basis for relevant departments. Each timestamp corresponds to an information propagation event, reflecting the dynamic interaction process of system behavior in the time dimension.

[0003] Existing cascade sequence prediction models mainly include point process models and dynamic graph models. The point process model is based on modeling the intensity of event occurrence, and representative models include TREND and CASPER. TREND designs a message passing network to estimate the Hawkes intensity function, which has the defect of only focusing on the interaction features and intensity between node variables, ignoring the diffusion process of information. CASPER proposes a prediction model with an analytical solution, which uses historical timestamps and time-dependent node numbers to estimate the mean and variance of the information cascade diffusion scale, but at the cost of oversimplifying the complex interactions in the real world, compressing the rich diffusion process composed of heterogeneous variables coupled into a single scale growth statistical model, resulting in serious information loss. SUMMARY

[0004] The present application aims to provide an information cascade sequence prediction method, device, equipment and medium, which solves the problems in the prior art.

[0005] The present application is implemented by the following technical solutions:

[0006] In a first aspect, the present application provides an information cascade sequence prediction method, comprising:

[0007] Obtaining an initial forwarding user set of a target information before a prediction time, and a forwarding behavior trajectory of each initial forwarding user before the prediction time, wherein the forwarding behavior trajectory represents data records of all forwarding information forwarded by the initial forwarding user, and the forwarding behavior trajectory includes forwarding of the target information;

[0008] For each forwarding information in each forwarding behavior trajectory, obtaining a forwarding information diffusion trajectory of the forwarding information before the prediction time, the forwarding information diffusion trajectory being used to represent data records of the forwarding information forwarded by all forwarding diffusion users, the all forwarding diffusion users including the initial forwarding user;

[0009] For each forwarded message, based on the forwarding information diffusion trajectory of the forwarded message and the initial forwarding user influence of each forwarding diffusion user, the forwarding information influence of the forwarded message is updated to obtain the updated forwarding information influence of each forwarded message.

[0010] For each initial forwarding user, based on the forwarding behavior trajectory of the initial forwarding user and the forwarding information influence of each forwarding information update, the forwarding user influence of the initial forwarding user is updated to obtain the updated forwarding user influence of each initial forwarding user.

[0011] The forwarding user influence updated by each initial forwarding user is taken as the initial forwarding user influence of each initial forwarding user among all forwarding diffusion users. Asynchronous loop optimization is performed according to the steps of updating the forwarding information influence and updating the user influence until the convergence condition is met. The forwarding information influence of the target information is finally updated as the prediction result.

[0012] Preferably, obtaining the initial set of forwarding users of the target information before the prediction time, and the forwarding behavior trajectory of each initial forwarding user before the prediction time, includes:

[0013] Extract all user identifiers that have forwarded the target information from the historical forwarding records of the target information and whose forwarding time is before the predicted time, and use the set of user identifiers as the initial forwarding user set;

[0014] For each user in the initial set of forwarding users, extract all forwarding records of the user before the predicted time, and each forwarding record contains at least the information identifier of the forwarded information and the forwarding timestamp;

[0015] The forwarding records are arranged in chronological order according to their forwarding timestamps to form a forwarding behavior trajectory.

[0016] Preferably, obtaining the forwarding information diffusion trajectory before the predicted time includes:

[0017] Extract the historical records of the forwarding information that were forwarded by all users before the predicted time, with each historical record containing at least the forwarding user identifier and the forwarding timestamp;

[0018] The historical records are arranged in chronological order of forwarding timestamps to form a trajectory of the spread of forwarded information.

[0019] Preferably, the initialized forwarding user influence of each forwarding diffusion user is obtained according to the following method:

[0020] According to the static attribute data and / or dynamic behavior statistical data of each forwarding diffusion user, the initialized forwarding user influence is calculated;

[0021] The static attribute data includes at least one of account permission level, fan quantity, attention quantity, and account registration duration.

[0022] The dynamic behavior statistical data includes at least one of historical total forwarding times, historical average forwarding interval, and average popularity of historical forwarding information.

[0023] Preferably, the forwarding influence of the forwarding information influence of the forwarding information is updated according to the forwarding information diffusion track of the forwarding information and the initialized forwarding user influence of each forwarding diffusion user, and the updated forwarding information influence of each forwarding information is obtained, including:

[0024] The diffusion mode feature of the forwarding information itself is extracted by encoding the forwarding information diffusion track.

[0025] The initialized forwarding user influence of each forwarding diffusion user is aggregated to generate a user influence correlation feature.

[0026] The diffusion mode feature and the user influence correlation feature are fused, and the updated forwarding information influence of the forwarding information is output by a preset first updating model calculation.

[0027] Preferably, the forwarding user influence of the initial forwarding user is updated according to the forwarding behavior track of the initial forwarding user and the updated forwarding information influence of each forwarding information, and the updated forwarding user influence of each initial forwarding user is obtained, including:

[0028] The behavior mode feature of the initial forwarding user is extracted by encoding the forwarding behavior track of the initial forwarding user.

[0029] The forwarding information influence correlation feature is generated by aggregating the updated forwarding information influence corresponding to each forwarding information in the forwarding behavior track.

[0030] The behavior mode feature and the forwarding information influence correlation feature are fused, and the updated forwarding user influence of the initial forwarding user is output by a preset second updating model calculation.

[0031] Preferably, the convergence condition includes at least one of:

[0032] In the two consecutive asynchronous loop optimizations, the change amount of the forwarding information influence of the target information is less than a first threshold.

[0033] The asynchronous loop optimization number reaches a preset maximum round.

[0034] In a second aspect, an embodiment of the present application provides an information cascade sequence prediction device, comprising:

[0035] The first obtaining module is configured to obtain an initial forwarding user set of the target information before a prediction time and a forwarding behavior track of each initial forwarding user before the prediction time, wherein the forwarding behavior track represents data records of all forwarding information forwarded by the initial forwarding user, and the forwarding behavior track includes forwarding of the target information.

[0036] The second obtaining module is configured to obtain, for each forwarding information in each forwarding behavior track, a forwarding information diffusion track of the forwarding information before the prediction time, wherein the forwarding information diffusion track represents data records of forwarding of the forwarding information by all forwarding diffusion users, and the all forwarding diffusion users include the initial forwarding user.

[0037] The first updating module is configured to, for each forwarding information, perform forwarding influence updating on forwarding information influence of the forwarding information according to the forwarding information diffusion track of the forwarding information and the initialized forwarding user influence of each forwarding diffusion user, to obtain updated forwarding information influence of each forwarding information.

[0038] The second updating module is configured to, for each initial forwarding user, perform user influence updating on forwarding user influence of the initial forwarding user according to the forwarding behavior track of the initial forwarding user and the updated forwarding information influence of each forwarding information, to obtain updated forwarding user influence of each initial forwarding user.

[0039] The output module is configured to take the updated forwarding user influence of each initial forwarding user as the initialized forwarding user influence of each initial forwarding user in the all forwarding diffusion users, perform asynchronous loop optimization according to the forwarding information influence updating and the user influence updating, until a convergence condition is met, and output the final updated forwarding information influence of the target information as a prediction result.

[0040] In a third aspect, an embodiment of the present application provides an electronic device, comprising at least one processor, at least one memory, and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, the method of the first aspect in the above embodiment is implemented.

[0041] In a fourth aspect, an embodiment of the present application provides a storage medium having computer program instructions stored thereon, when the computer program instructions are executed by a processor, the method of the first aspect in the above embodiment is implemented.

[0042] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0043] The method realizes dynamic collaborative modeling of the information propagation process by constructing a double-layer diffusion network of target information, initial forwarding users and their associated information, and designing an asynchronous cyclic optimization mechanism of alternately updating information influence and user influence. Specifically, the system first extracts the propagation trajectory of the target information, the historical behavior trajectory of the users, and the diffusion trajectory of the associated information, and constructs a multi-level data association. Then, by iteratively performing the steps of information influence update and user influence update, the information propagation value and the user propagation ability are mutually enhanced in the cycle: in each iteration, the information propagation potential is re-evaluated based on the updated user influence, and the user propagation efficiency is re-evaluated based on the updated information influence, forming a continuous optimization feedback loop. This mechanism effectively overcomes the information loss problem of traditional one-way modeling, and can simultaneously capture the dynamic interaction of propagation content features, propagator attribute features and network structure features. By setting the convergence condition to control the iteration depth, the system reaches a stable state within a limited number of iterations, and finally outputs the information influence after multiple rounds of collaborative optimization as the prediction result, thereby achieving more comprehensive and robust prediction performance in social network information diffusion, news propagation scale prediction and other scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the example embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be considered as limiting the scope. For those skilled in the art, other related drawings can also be obtained without creative labor. In the drawings:

[0045] Figure 1 The flowchart of the information cascade sequence prediction method provided by the present application is shown in the figure;

[0046] Figure 2 The schematic diagram of the initial state in the case provided by the present application is shown in the figure;

[0047] Figure 3 The result schematic diagram after the first round of iteration in the case provided by the present application is shown in the figure;

[0048] Figure 4 The result schematic diagram after the second round of iteration in the case provided by the present application is shown in the figure;

[0049] Figure 5 The structure schematic diagram of the information cascade sequence prediction device provided by the present application is shown in the figure;

[0050] Figure 6A structural schematic diagram of an electronic device provided by the present application. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present application clearer, further detailed description will be given to the present application in combination with embodiments and drawings, the illustrative embodiments of the present application and the description thereof are only used to explain the present application, and do not limit the present application.

[0052] It should be noted that, in this document, the terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms “include”, “contain” or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the elements defined by the statement “include” do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0053] It should be noted that all actions of acquiring signals, information or data in the present application are carried out in compliance with the corresponding data protection regulations and policies of the place, and with the authorization given by the owner of the corresponding device.

[0054] Embodiment 1

[0055] Please refer to Figure 1 The embodiment of the present application provides an information cascade sequence prediction method, which comprises the following steps:

[0056] S1, acquiring an initial forwarding user set of target information before a prediction time, and a forwarding behavior track of each initial forwarding user before the prediction time, wherein the forwarding behavior track represents data records of all forwarding information forwarded by the initial forwarding user, and the forwarding behavior track includes forwarding of the target information;

[0057] Specifically, this step involves the systematic collection and structuring of the initial state of a specific information dissemination process. The target information refers to the digital content entity whose future dissemination scale is to be predicted, such as a tweet with a specific hashtag on a social media platform, a news article on a news client, or an original video on a short video platform. The prediction time is a set point in time; the system needs to estimate the cumulative dissemination scale of the information at the current time at the prediction time. The process of obtaining the initial set of forwarding users is as follows: using the unique content identifier of the target information as the search key, the system queries the platform's historical behavior database for all user account identifiers whose forwarding timestamps are earlier than the prediction time. The set of these identifiers is defined as the initial set of forwarding users. For example, when the target information is a news report about technology trends, the initial set of forwarding users includes all reader accounts that forwarded the article before the prediction time. For each user in this set, the system further extracts their complete behavioral records within the same time frame, arranging them in ascending chronological order to form a forwarding behavior trajectory. This trajectory is an ordered sequence of (forwarded information identifier, forwarding timestamp) tuples; for example, a user might have forwarded sports news, entertainment videos, and the target technology article sequentially over time. This structured representation not only records the key behaviors of users forwarding target information, but more importantly, it captures the user's content selection preferences and behavioral rhythm patterns throughout the observation period, providing a multi-dimensional data foundation for subsequent analysis of user influence.

[0058] S2. For each forwarding information in each forwarding behavior trajectory, obtain the forwarding information diffusion trajectory before the predicted time. The forwarding information diffusion trajectory is used to characterize the data records of all forwarding diffusion users forwarding the forwarding information. The initial forwarding user is included among all forwarding diffusion users.

[0059] Specifically, this step establishes a two-way mapping relationship from the user behavior dimension to the information propagation dimension. Each forwarding behavior trajectory obtained in step S1 is analyzed, and each forwarding information identifier in the trajectory represents an independent digital content that the user has propagated in the past. For example, a user in the technology field may have a forwarding behavior trajectory that contains product release information from multiple different technology companies. For each forwarding information in the trajectory, the system performs an independent diffusion network reconstruction: using the unique identifier of the information as the search key, it queries the platform history database for all user accounts whose forwarding time is earlier than the predicted time and their forwarding timestamps, and arranges them in chronological order to form the complete propagation chain of the information. For example, for a smartphone review video in a user's trajectory, its diffusion trajectory records the time sequence propagation path formed by multiple layers of user forwarding starting from the initial publisher. In particular, since the initial forwarding user's forwarding behavior trajectory contains a forwarding record of this information, the initial forwarding user's forwarding node must be found in the retrieved diffusion trajectory, and this design ensures the closure of the user-information association network. This two-way trajectory acquisition mechanism constructs the cross-network between user behavior and information propagation, enabling the forwarding behavior of a single user to be associated with the overall diffusion pattern of the information it propagates, and the diffusion trajectory of a single information to reflect the participation characteristics of different propagators, providing structured data support for subsequent collaborative optimization.

[0060] S3, for each forwarding information, updating the forwarding information influence of the forwarding information according to the forwarding information diffusion trajectory of the forwarding information and the initialized forwarding user influence of each forwarding diffusion user, to obtain the updated forwarding information influence of each forwarding information;

[0061] Specifically, this step performs dynamic re-estimation of information influence. For each diffusion trajectory of the forwarding information obtained in step S2, the system conducts comprehensive analysis in combination with the initial influence state of each user participating in the information dissemination. For example, the diffusion trajectory of a certain science popularization video shows that it is transmitted through the forwarding of users in different fields, including science bloggers, ordinary audiences and other types of users. The system presets an initial forwarding user influence parameter for each user, which can be calculated based on user historical authority indicators (such as authentication level), propagation efficiency indicators (such as historical forwarding adoption rate) and other dimensions. The forwarding information influence updating process adopts a fusion calculation model, which takes the time sequence characteristics of the diffusion trajectory as input, and takes the user influence of each node in the propagation path as the weighted coefficient. Through the time sequence attention mechanism or dynamic weight distribution algorithm, the influence estimation value of the information in the current propagation stage is calculated. Specifically, when a piece of information is forwarded by a high-influence user earlier, its influence update value will gain greater gain; on the contrary, if the propagation mainly occurs in the low-influence user group, the influence update may be relatively conservative. This updating mechanism makes the evaluation of information influence not only consider the propagation breadth (diffusion trajectory length), but also pay more attention to the propagation quality (influence distribution of propagation nodes), forming a multi-dimensional quantitative evaluation of the potential of information dissemination.

[0062] S4, for each initial forwarding user, updating the forwarding user influence of the initial forwarding user according to the forwarding behavior trajectory of the initial forwarding user and the updated forwarding information influence of each forwarding information, to obtain the updated forwarding user influence of each initial forwarding user;

[0063] Specifically, this step realizes the iterative optimization of user influence. Based on the complete behavior trajectory of each initial forwarding user obtained in step S1, the system conducts a reverse evaluation in combination with the latest influence state (updated by step S3) of each information historically forwarded by the user. For example, a user has forwarded 20 pieces of information in the past three months, and after step S3 calculation, the updated influence evaluation is obtained respectively. The user influence updating process uses an association analysis model, which analyzes the content selection mode (such as preference for technology-related content) and time behavior characteristics (such as high-frequency forwarding period) reflected in the user behavior trajectory, and at the same time takes the influence evaluation of each information in the trajectory as an input feature, recalculates the influence parameter of the user at the current time through weighted aggregation or pattern matching algorithm. In particular, when the information historically forwarded by the user generally has a high updated influence value, the influence parameter of the user will be correspondingly improved, because such correlation indicates that the user has the ability to discover and disseminate high-potential information; on the contrary, if the information influence of the user is generally low, the influence parameter of the user may be adjusted accordingly. This design realizes a positive feedback mechanism from information quality to disseminator reputation, so that the evaluation of user influence is not only based on its static attributes, but also reflects the effectiveness of its dynamic behavior selection.

[0064] S5, updating the forwarding user influence of each initial forwarding user as the initialized forwarding user influence of each initial forwarding user in all forwarding diffusion users, performing asynchronous loop optimization according to the steps of the forwarding information influence updating and the user influence updating until a convergence condition is met, and outputting the final updated forwarding information influence of the target information as a prediction result.

[0065] Specifically, this step establishes a closed-loop iterative optimization system. The system reassigns the updated user influence parameter calculated in step S4 to the corresponding initial forwarding user, and these users will use the updated influence parameter for calculation when they act as forwarding diffusion nodes in the subsequent iteration of the information propagation network. For example, in the first round of calculation, the initial influence of user A is 0.6, which is updated to 0.72 after step S4; in the second round of calculation, when the diffusion trajectory of other information involves user A, 0.72 will be used as its influence parameter. The loop optimization process is achieved by alternately executing step S3 (information influence update) and step S4 (user influence update): in each iteration, first update all information influence values using the current user influence parameter, and then recompute the user influence parameter using the updated information influence value, forming an alternating optimization flow of information influence, user influence, and information influence. Convergence is monitored based on parameter change amplitude, for example, when the change rate of the influence of the target information in the last two iterations is less than a set threshold (such as 0.5%), or when the maximum number of iterations (such as 50 rounds) is reached, the loop is terminated. The final output of the target information forwarding information influence, after multiple rounds of mutual enhancement calculation, integrates the information propagation characteristics and propagation network quality factors, and is converted into a future propagation scale prediction value. This asynchronous loop structure simulates the dynamic process of mutual shaping of information value and propagator reputation in real propagation, and gradually approaches the system stable state through parameter alternation update.

[0066] In some embodiments, the obtaining of the initial forwarding user set of the target information before the prediction time, and the forwarding behavior trajectory of each initial forwarding user before the prediction time, comprises:

[0067] Extracting the user identifiers of all users who have forwarded the target information and whose forwarding time is before the prediction time from the historical forwarding records of the target information, and taking the set of user identifiers as the initial forwarding user set;

[0068] For each user in the initial forwarding user set, extracting all forwarding records of the user before the prediction time, each forwarding record containing at least the information identifier being forwarded and the forwarding timestamp;

[0069] Arranging the forwarding records in chronological order of the forwarding timestamps to form the forwarding behavior trajectory.

[0070] In some embodiments, the obtaining of the forwarding information diffusion trajectory of the forwarding information before the prediction time comprises:

[0071] Extracting the historical records of the forwarding information being forwarded by all users before the prediction time, each historical record containing at least the forwarding user identifier and the forwarding timestamp;

[0072] arranging the history records in chronological order of forwarding timestamps to form a forwarding information diffusion trajectory.

[0073] In some embodiments, the initialized forwarding user influence of each forwarding diffusion user is obtained according to the following manner:

[0074] According to the static attribute data and / or dynamic behavior statistical data of each forwarding diffusion user, the initialized forwarding user influence is calculated.

[0075] The static attribute data includes at least one of account permission level, number of fans, number of followings, and account registration duration.

[0076] The dynamic behavior statistical data includes at least one of total number of historical forwarding times, average forwarding interval, and average popularity of historical forwarding information.

[0077] Specifically, the system calculates the initialized forwarding user influence for each forwarding diffusion user (i.e., all users involved in the diffusion trajectory in step S2), which integrates two types of data features: static attribute data reflecting the platform identity characteristics of the user, including account permission level (such as ordinary user, authenticated user, official media, etc.), number of fans, number of followings, and account registration duration, etc. long-term stable attributes; dynamic behavior statistical data reflecting the historical behavior pattern of the user, including total number of historical forwarding times, average forwarding interval, and average popularity of historical forwarding information, etc. behavior derived indicators. The initial influence calculation can be implemented by using weighted linear combination, multilayer perceptron or rule-based scoring system, etc. model, for example, the number of fans is logarithmically normalized, the account permission level is mapped to a weight coefficient, and the average popularity of historical forwarding information is taken as a quality factor. Through feature engineering and model calculation, a standardized initial value of influence between 0 and 1 is generated. In particular, for the average popularity of historical forwarding information, the system needs to trace back all the information forwarded by the user in the past, and calculate the average number of forwarding of these information in the respective observation period as a reference benchmark. This initialization mechanism provides a reasonable starting point for subsequent iterative optimization, avoiding the instability that may be caused by completely random initialization, and retaining the space for further improving the accuracy of parameters through subsequent collaborative update.

[0078] In some embodiments, the forwarding information influence of the forwarding information is updated according to the forwarding information diffusion trajectory of the forwarding information and the initialized forwarding user influence of each forwarding diffusion user, to obtain the updated forwarding information influence of each forwarding information, including:

[0079] The forwarding information diffusion trajectory is encoded to extract the diffusion pattern features of the forwarding information itself.

[0080] aggregating the initialized forwarding user influence of each forwarding diffusion user to generate a user influence associated feature;

[0081] fusing the diffusion pattern feature and the user influence associated feature, and outputting the forwarding information influence of the forwarding information update through a preset first update model.

[0082] Specifically, first, the forwarding information diffusion trajectory is encoded, which is a time sequence composed of user identifiers and forwarding time stamps; the encoding process extracts multi-dimensional diffusion pattern features through time series modeling techniques, including the time distribution characteristics of forwarding events, the structural features of propagation paths, and the evolution law of forwarding intensity. At the same time, for each forwarding diffusion user involved in the diffusion trajectory, the initialized forwarding user influence parameter of the user is obtained, which is calculated based on the static attribute data of the user (such as account permission level, number of fans) and dynamic behavior statistical data (such as total number of historical forwarding times, average popularity of historical forwarding information); by aggregating these influence parameters, a user influence associated feature reflecting the overall quality and structural features of the propagation network is generated. Subsequently, the diffusion pattern feature and the user influence associated feature are fused in multiple dimensions, and the fused feature vector is input into a preset first update model; the model learns the interaction law between the diffusion features and the user influence features, and calculates and outputs the updated forwarding information influence value. This updating process makes the evaluation of information influence not only consider the time dynamic characteristics of the propagation process, but also combine the quality distribution of user nodes in the propagation network, forming a comprehensive quantitative representation of the information propagation potential.

[0083] The first update model can adopt a diffusion modeling architecture based on a point process. The model regards the forwarding information diffusion trajectory as a point process event sequence, and models the occurrence probability of forwarding events by defining a conditional intensity function , where Ht represents the set of historical events up to time t. The specific form of the intensity function can be designed as: , where represents the basic propagation rate, is the influence coefficient, is the forwarding time of the historical event, is a time decay kernel function (such as an exponential function or a power law function), is the initialized influence weight of the forwarding user . The user influence associated feature is incorporated into the model through the parameter, and the diffusion pattern feature is encoded through the parameterized form of the time decay kernel function. The model parameters can be optimized through maximum likelihood estimation, and finally the quantitative estimation value of information influence is obtained by integrating the intensity function: , where To observe the time window.

[0084] The first update model can adopt an attention mechanism-based time fusion network, which includes a diffusion feature encoding layer, an influence feature projection layer, and a cross-attention calculation layer. The diffusion feature encoding layer receives the time sequence feature vector obtained by encoding the diffusion trajectory, and captures the importance difference of different time segments through a time attention module; the influence feature projection layer maps the user influence association feature to a dimension space matching the diffusion feature; the cross-attention calculation layer establishes a dynamic weight association between the two types of features, and through a multi-head attention mechanism, respectively takes the diffusion feature as a query vector and the influence feature as a key value vector to calculate the weighted fused context representation. Finally, the scalar form of the influence update value is output through a multi-layer perception regression head.

[0085] In some embodiments, the user influence update of the initial forwarding user's forwarding user influence according to the forwarding behavior trajectory of the initial forwarding user and the forwarding information influence updated by each forwarding information update includes:

[0086] Encode the forwarding behavior trajectory of the initial forwarding user to extract the behavior pattern feature of the initial forwarding user;

[0087] Aggregate the forwarding information influence updated by each forwarding information in the forwarding behavior trajectory to generate a forwarding information influence association feature;

[0088] Fuse the behavior pattern feature and the forwarding information influence association feature, and output the initial forwarding user's updated forwarding user influence through a preset second update model calculation.

[0089] Specifically, this step realizes the iterative re-estimation of user influence. First, the forwarding behavior trajectory of the initial forwarding user is encoded, which is a time sequence composed of (forwarded information identifier, forwarding timestamp) tuples; the encoding process extracts multi-dimensional behavior pattern features through behavior modeling techniques, including the time distribution characteristics of the forwarded content, the change law of the forwarding frequency, the distribution characteristics of the forwarded information type, and the periodic pattern of the behavior activity. At the same time, for each forwarded information recorded in the forwarding behavior trajectory, the updated forwarding information influence value is obtained, forming a dynamically updated influence parameter set; by aggregating these influence parameters, the forwarding information influence correlation features reflecting the historical forwarding content quality distribution of the user are generated, which can specifically calculate the weighted average value, time series weighted sum, and influence growth trend of the forwarding information influence. Subsequently, the behavior pattern features and the forwarding information influence correlation features are fused in multiple dimensions, and the fused feature vector is input into the preset second update model; this model learns the correlation between user behavior features and the quality features of the forwarded content, and calculates the updated forwarding user influence value. This updating process enables the evaluation of user influence not only based on the time dynamic characteristics of the historical behavior pattern, but also combined with the quality evolution of the propagated content, forming a dynamic quantitative characterization of the user's propagation ability. The second update model can use the same or different machine learning architecture as the first update model, including but not limited to attention mechanism-based feature fusion network, hybrid model of graph neural network and behavior sequence, or gradient boosting decision tree integrated learning model.

[0090] The following describes the embodiment through a specific case.

[0091] I. Case background and data initialization

[0092] The diffusion scale of information m at future time T needs to be predicted, and the current time t < T. The system identifies the users who have forwarded information m before t to form the initial forwarding set {user1, user2}. Based on historical data, the user behavior trajectory is constructed: the trajectory of user1 is [(information m, t1), (information 1, t2)], and the trajectory of user2 is [(information m, t3), (information 2, t4)], where t1 to t4 are all time stamps less than t. Further, the diffusion trajectory of the associated information is extracted: the diffusion trajectory of information 1 is [(user1, t2), (user3, t5)], and the diffusion trajectory of information 2 is [(user2, t4), (user4, t6)].

[0093] II. Influence initialization and asynchronous loop optimization

[0094] All users are calculated initial influence by static attributes (e.g. the number of fans, authentication status) and dynamic behavior (e.g. the average popularity of historical retweets), the normalized initial values of user 1, 2, 3, 4 are 0.72, 0.35, 0.45, 0.28 respectively; the initial value of all information influence is set to 0.5. The system enters the asynchronous loop optimization, the initial state is shown in Figure 2 The first update model calculates the updated information influence of information m, information 1 and information 2 as 0.47, 0.68 and 0.42 respectively based on the diffusion trajectory and the initial influence of users in the first round of iteration. Then, the second update model calculates the updated user influence of user 1 and user 2 as 0.75 and 0.38 respectively based on the user behavior trajectory and the updated information influence, the first round of iteration results are shown in Figure 3 .

[0095] In the second round of iteration, the first update model recalculates the information influence using the updated user influence, the information influence of information m, information 1 and information 2 is adjusted to 0.48, 0.71 and 0.43 respectively. The second update model calculates the user influence of user 1 and user 2 as 0.77 and 0.41 respectively using the updated information influence in this round of iteration, the second round of iteration results are shown in Figure 4 . This process is continued using multiple rounds of iteration, the information influence and the user influence are updated alternately in each iteration. The user influence of user 3 and user 4 remains unchanged in each round, the system checks the convergence condition after each round of iteration, the information influence change rate of information m is set to be less than 1% in this case. After five rounds of iteration, the information influence of information m stabilizes at 0.47, the change rate is reduced to 0.8%, which meets the convergence condition. Finally, the information influence of information m is output as 0.47.

[0096] Embodiment 2

[0097] Please refer to Figure 5 , the embodiment of the present application provides an information cascade sequence prediction device, which comprises:

[0098] A first acquisition module is configured to acquire an initial retweet user set of target information before a prediction time and a retweet behavior trajectory of each initial retweet user before the prediction time, wherein the retweet behavior trajectory represents data records of all retweet information retweeted by the initial retweet user, and the retweet behavior trajectory includes the retweet of the target information;

[0099] A second acquisition module is configured to acquire, for each retweet information in each retweet behavior trajectory, a retweet information diffusion trajectory of the retweet information before the prediction time, wherein the retweet information diffusion trajectory is used to represent data records of all retweet diffusion users retweeting the retweet information, and the all retweet diffusion users include the initial retweet user.

[0100] The first updating module is configured to, for each forwarding information, update the forwarding information influence of the forwarding information according to a forwarding information diffusion track of the forwarding information and the forwarding user influence initialized by each forwarding diffusion user, to obtain an updated forwarding information influence of each forwarding information.

[0101] The second updating module is configured to, for each initial forwarding user, update the forwarding user influence of the initial forwarding user according to a forwarding behavior track of the initial forwarding user and the updated forwarding information influence of each forwarding information, to obtain an updated forwarding user influence of each initial forwarding user.

[0102] The output module is configured to, as the forwarding user influence initialized by each initial forwarding user in the all forwarding diffusion users, perform asynchronous loop optimization according to the forwarding information influence updating and the user influence updating until a convergence condition is met, and output the final updated forwarding information influence of the target information as a prediction result.

[0103] It should be noted that the modules and units in the information cascade sequence prediction device in this embodiment one-to-one correspond to the steps in the information cascade sequence prediction method in the foregoing embodiments, and therefore the specific implementation of this embodiment can refer to the implementation of the foregoing information cascade sequence prediction method, which will not be described here again.

[0104] Embodiment 3

[0105] Please refer to Figure 6 The embodiment provides an electronic device, including: at least one processor 601 and memory 602. Optionally, the device further includes a communication component 603. Wherein, the processor 601, the memory 602 and the communication component 603 are connected through the bus 604.

[0106] In the specific implementation process, the at least one processor 601 executes the computer execution instruction stored in the memory 602, so that the at least one processor 601 executes the method described above.

[0107] The specific implementation process of the processor 601 can refer to the method embodiments described above, which has similar implementation principles and technical effects, and will not be described here again in this embodiment.

[0108] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the application can be directly embodied as hardware processor execution, or executed by a combination of hardware and software modules in the processor.

[0109] The memory can include a random access memory (RAM), and can also include a non-volatile memory (NVM), such as at least one disk memory.

[0110] The bus can be an industry standard architecture (ISA) bus, a peripheral component (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application does not limit to only one bus or one type of bus.

[0111] The present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the above method.

[0112] The present application also provides a computer readable storage medium, which stores computer execution instructions, and when a processor executes the computer execution instructions, the above method is implemented.

[0113] The above readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0114] An example readable storage medium is coupled to the processor such that the processor can read information from the readable storage medium and can write information to the readable storage medium. Of course, the readable storage medium can also be a part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in the device.

[0115] The division of units is only a logical functional division, and in actual implementation, there can be another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0116] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0117] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit.

[0118] If the functions are realized in the form of software functional units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0119] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The foregoing program can be stored in a computer readable storage medium. The program executes to perform the steps of the above-mentioned method embodiments; and the foregoing storage medium includes various storage media that can store program codes, such as ROM, RAM, magnetic disk or optical disk.

[0120] The above detailed description of the specific implementation of the present application further explains the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific implementation of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method of information cascading sequence prediction, the method comprising: The method comprises the following steps: obtaining an initial forwarding user set of target information before a prediction time, and a forwarding behavior track of each initial forwarding user before the prediction time, wherein the forwarding behavior track represents data records of all forwarding information forwarded by the initial forwarding user, and the forwarding behavior track comprises forwarding of the target information; for each forwarding information in each forwarding behavior track, obtaining a forwarding information diffusion track of the forwarding information before the prediction time, wherein the forwarding information diffusion track represents data records of all forwarding diffusion users forwarding the forwarding information, and the all forwarding diffusion users comprise the initial forwarding user; for each forwarding information, updating forwarding information influence of the forwarding information according to the forwarding information diffusion track of the forwarding information and the initialized forwarding user influence of each forwarding diffusion user, to obtain updated forwarding information influence of each forwarding information; for each initial forwarding user, updating forwarding user influence of the initial forwarding user according to the forwarding behavior track of the initial forwarding user and the updated forwarding information influence of each forwarding information, to obtain updated forwarding user influence of each initial forwarding user; performing asynchronous loop optimization according to the forwarding information influence updating and the user influence updating until a convergence condition is met, and outputting the final updated forwarding information influence of the target information as a prediction result.

2. The information cascading sequence prediction method of claim 1, wherein, The method comprises the following steps: extracting user identifiers of all users forwarding the target information before the prediction time from historical forwarding records of the target information, and taking a set of the user identifiers as the initial forwarding user set; for each user in the initial forwarding user set, extracting all forwarding records of the user before the prediction time, wherein each forwarding record comprises at least an information identifier being forwarded and a forwarding timestamp; arranging the forwarding records in chronological order of the forwarding timestamps to form a forwarding behavior track.

3. The information cascading sequence prediction method of claim 1, wherein, The method comprises the following steps: extracting historical records of the forwarding information being forwarded by all users before the prediction time, wherein each historical record comprises at least a forwarding user identifier and a forwarding timestamp; arranging the historical records in chronological order of the forwarding timestamps to form a forwarding information diffusion track.

4. The information cascading sequence prediction method of claim 1, wherein, The initialized forwarding user influence of each forwarding diffusion user is obtained in the following manner: calculating the initialized forwarding user influence according to static attribute data and / or dynamic behavior statistical data of each forwarding diffusion user; wherein the static attribute data comprises at least one of account permission level, number of fans, number of followers, and account registration duration. The dynamic behavior statistics include at least one of a historical total forwarding times, a historical average forwarding interval, and a historical average popularity of the forwarding information.

5. The information cascading sequence prediction method of claim 1, wherein, The forwarding influence updating of the forwarding information influence of the forwarding information is performed according to the forwarding information diffusion track of the forwarding information and the forwarding user influence initialized by each forwarding diffusion user, to obtain the updated forwarding information influence of each forwarding information, including: The diffusion mode feature of the forwarding information itself is extracted by encoding the forwarding information diffusion track; The user influence association feature is generated by aggregating the forwarding user influence initialized by each forwarding diffusion user; The updated forwarding information influence of the forwarding information is output by fusing the diffusion mode feature and the user influence association feature through preset first updating model calculation.

6. The information cascading sequence prediction method of claim 1, wherein, The user influence updating of the forwarding user influence of the initial forwarding user is performed according to the forwarding behavior track of the initial forwarding user and the updated forwarding information influence of each forwarding information, to obtain the updated forwarding user influence of each initial forwarding user, including: The behavior mode feature of the initial forwarding user is extracted by encoding the forwarding behavior track of the initial forwarding user; The forwarding information influence association feature is generated by aggregating the updated forwarding information influence corresponding to each forwarding information in the forwarding behavior track; The updated forwarding user influence of the initial forwarding user is output by fusing the behavior mode feature and the forwarding information influence association feature through preset second updating model calculation.

7. The information cascading sequence prediction method of claim 1, wherein, The convergence condition includes at least one of the following: In two continuous asynchronous loop optimizations, the change amount of the forwarding information influence of the target information is less than a first threshold value; The number of asynchronous loop optimizations reaches a preset maximum round.

8. An information concatenation sequence prediction apparatus characterized by comprising: The method comprises: The first acquisition module is configured to acquire a set of initial forwarding users of target information before a prediction time, and a forwarding behavior track of each initial forwarding user before the prediction time, wherein the forwarding behavior track represents data records of all forwarding information forwarded by the initial forwarding user, and the forwarding behavior track includes forwarding of the target information; The second acquisition module is configured to acquire, for each forwarding information in each forwarding behavior track, a forwarding information diffusion track of the forwarding information before the prediction time, wherein the forwarding information diffusion track represents data records of forwarding of the forwarding information by all forwarding diffusion users, and the all forwarding diffusion users include the initial forwarding user; The first updating module is configured to perform, for each forwarding information, forwarding influence updating of a forwarding information influence of the forwarding information according to the forwarding information diffusion track of the forwarding information and the forwarding user influence initialized by each forwarding diffusion user, to obtain an updated forwarding information influence of each forwarding information; a second updating module, configured to perform user influence updating on the influence of the initial forwarding user on the forwarding user according to the forwarding behavior track of the initial forwarding user and the forwarding information influence of each forwarding information update, to obtain the updated forwarding user influence of each initial forwarding user; an output module, configured to take the updated forwarding user influence of each initial forwarding user as the initialized forwarding user influence of each initial forwarding user in the all forwarding diffusion users, to perform asynchronous loop optimization according to the forwarding information influence updating and the user influence updating until a convergence condition is met, and to output the final updated forwarding information influence of the target information as a prediction result.

9. An electronic device, comprising: comprising: at least one processor, at least one memory, and computer program instructions stored in the memory that, when executed by the processor, implement the method of any one of claims 1-7.

10. A computer-readable storage medium having stored thereon computer program instructions, wherein, when the computer program instructions are executed by the processor, implement the method of any one of claims 1-7.

Images