Event heat prediction method and system based on multi-view neural hox process

Abstract

The application discloses an event heat prediction method and system based on a multi-view neural Hawkes process, relates to the technical field of data mining, and comprises the following steps: receiving a streaming event heat prediction data set, wherein the streaming event heat prediction data set is constructed based on event-related data, the event-related data comprises post information, comment information and heat information, the streaming event heat prediction data set is extracted to obtain a training set, the training set is input into a pre-established streaming event heat prediction model based on a multi-view neural Hawkes process, the streaming event heat prediction model based on the multi-view neural Hawkes process is trained based on a negative log maximum likelihood loss function, and a trained streaming event heat prediction model based on the multi-view neural Hawkes process is obtained; and the streaming event heat prediction data set is input into the trained streaming event heat prediction model based on the multi-view neural Hawkes process, and an event heat prediction result is output, so that the prediction accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data mining, in particular to an event heat prediction method and system based on a multi-view neural Hawkes process. BACKGROUND

[0002] Social media platforms have fundamentally changed the way information is disseminated. Platforms such as Weibo and TikTok allow users to quickly access breaking news and discuss social events and share their opinions on products. The extensive interaction of users on these platforms reflects the collective behavior of different groups. In this context, event heat prediction has begun to attract widespread attention from researchers because of its important influence on content distribution and digital marketing.

[0003] Early research mainly treated popularity prediction as a classification task (i.e., predicting whether a post will be popular) or a regression task (i.e., predicting the maximum popularity of a post) at the post level and used statistical methods for prediction. With the rise of deep learning, existing methods have begun to use neural networks for automatic feature extraction and popularity prediction. From the existing methods, they can be roughly divided into two categories: (1) methods based on propagation paths: they predict which users are likely to receive information next and estimate popularity by counting the number of users who receive information. (2) methods based on content understanding: they predict popularity based on the semantic information of the content, which assumes that similar content tends to exhibit similar popularity. Although these methods have achieved preliminary success, they have not fully modeled the influence of different information related to events on heat changes, and the prediction results have high uncertainty, making it difficult to ensure reasonable prediction results. SUMMARY

[0004] To solve the problems mentioned in the background, the purpose of the present application is to provide an event heat prediction method and system based on a multi-view neural Hawkes process.

[0005] In a first aspect, the purpose of the present application can be achieved by the following technical solution: an event heat prediction method based on a multi-view neural Hawkes process, the method comprising the following steps:

[0006] Receiving a streaming event heat prediction dataset, the streaming event heat prediction dataset being constructed based on event-related data, wherein the event-related data includes post information, comment information, and heat information;

[0007] Extracting the streaming event heat prediction dataset to obtain a training set, inputting the training set into a pre-established streaming event heat prediction model based on a multi-view neural Hawkes process, training based on a negative log-likelihood loss function, and obtaining a trained streaming event heat prediction model based on a multi-view neural Hawkes process;

[0008] The stream event hotness prediction dataset is input into the trained multi-view neural Hawkes process-based stream event hotness prediction model, and an event hotness prediction result is output.

[0009] In combination with the first aspect, in some implementations of the first aspect, the method further includes that the pre-established multi-view neural Hawkes process-based stream event hotness prediction model includes a multi-view feature encoder and a Hawkes process driven predictor.

[0010] The multi-view feature encoder includes a time series view encoder, a post view encoder, and a comment view encoder; the time series view encoder takes a historical hotness sequence of the hotness information as input and outputs time series features; the post view encoder takes text of the post information as input and outputs post features; and the comment view encoder takes text of the comment information as input and outputs comment features.

[0011] The Hawkes process driven predictor takes the time series features, the post features, and the comment features as input and outputs predicted future event hotness of the next day, where t is a positive integer. t is a positive integer.

[0012] In combination with the first aspect, in some implementations of the first aspect, the method further includes that the process of the time series view encoder includes:

[0013] A sliding window strategy is adopted to obtain historical hotness values of T days, a sequence longer than T days is truncated, and a sequence shorter than T days is reversed and zero-padded.

[0014] The change amount of the hotness of each day in T days is calculated to obtain a hotness change sequence with a length of T-1.

[0015] The hotness sequence is standardized by logarithmic change:

[0016]

[0017] wherein, is the change amount of the hotness of the tth day, is the standardized hotness change amount, sign(·) is a sign function, and |·| is an absolute value function.

[0018] Each standardized hotness change amount in the hotness sequence is mapped to a learnable hotness representation by an embedding layer, where d is the dimension of the representation.

[0019] A learnable position encoding is added to each hotness representation in the hotness sequence:

[0020]

[0021] ​wherein, represents the learnable position encoding for the heat change amount of the t-th day, represents the value of the i-th position in the feature, the value of the i-th position in the feature, the value of the i-th position in the feature, the value of the i-th position in the feature, represents the heat feature after adding the position encoding;

[0022] fuses the heat feature after adding the position encoding in each heat sequence:

[0023]

[0024] wherein, is the feature encoded by the final time sequence perspective encoding module, and Transform (·) is a Transformer encoder function.

[0025] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the post perspective encoding process includes:

[0026] randomly sampling n posts from the posts related to the event collected on the day, wherein, n is a positive integer;

[0027] encoding the features of each post using a pre-trained text semantic feature encoder and a text sentiment feature encoder, for the i-th post, the semantic feature is , and the sentiment feature is ;

[0028] using a mutual attention mechanism to enhance the semantic feature and the sentiment feature:

[0029]

[0030] wherein, , is the enhanced semantic feature and the sentiment feature; , , is the semantic feature query feature, index feature and value feature obtained by linear mapping; , , is the sentiment feature query feature, index feature and value feature obtained by linear mapping; and softmax (·) is a softmax function.

[0031] fusing the features of each post:

[0032] ​

[0033] wherein, is the feature encoded by the final post perspective encoding module, is the enhanced semantic feature of the nth post, is the enhanced sentiment feature of the nth post.

[0034] In combination with the first aspect, in some implementations of the first aspect, the method further comprises: the process of the comment perspective encoding comprises:

[0035] randomly sampling comments from the comments collected on the day related to the event, wherein, is a positive integer;

[0036] encoding the features of each comment using a pre-trained text semantic feature encoder and a text sentiment feature encoder, for the ith comment, the semantic feature is , and the sentiment feature is ;

[0037] enhancing the semantic feature and the sentiment feature using a mutual attention mechanism:

[0038]

[0039] wherein, , is the enhanced semantic feature and the sentiment feature; , , is the semantic feature of the query feature, the index feature and the value feature obtained by linear mapping; , , is the sentiment feature of the query feature, the index feature and the value feature obtained by linear mapping; softmax(·) is a softmax function;

[0040] fusing the features of each comment:

[0041]

[0042] wherein, is the feature encoded by the final comment perspective encoding module, is the enhanced semantic feature of the nth comment, is the enhanced sentiment feature of the nth comment.

[0043] In combination with the first aspect, in some implementations of the first aspect, the method further comprises: the Hawkes process driven predictor is as follows:

[0044] According to the timing characteristics, the change amount of the heat under the timing perspective influence is calculated:

[0045]

[0046] wherein, , is a weight mapping matrix, , is a bias matrix; sign(·) is a sign function; |·| is an absolute value function; is the observed heat value of the event at t moment; is the predicted change amount of the heat under the timing perspective influence at t+1 moment;

[0047] According to the post characteristics, the change amount of the heat under the post perspective influence is calculated:

[0048]

[0049] wherein, , is a weight mapping matrix, , is a bias matrix; is a learnable timing feature; is the predicted change amount of the heat under the post perspective influence at t+1 moment;

[0050] The change amount of the heat under the comment perspective influence is calculated in the same way as the change amount of the heat under the post perspective influence ;

[0051] The influences of various perspectives are integrated to predict the heat of the event of the next day:

[0052]

[0053] wherein, is the predicted heat of the event of the next day; according to the formula, a step-by-step cycle prediction is performed to obtain the heat related to the event in the future day, is a positive integer.

[0054] In the second aspect, in order to achieve the above object, the application discloses an event heat prediction system based on a multi-perspective neural Hox process, comprising:

[0055] A data acquisition module is used to receive a streaming event heat prediction data set, and the streaming event heat prediction data set is constructed based on event related data, wherein the event related data includes post information, comment information and heat information;

[0056] The model training module is configured to extract the streaming event hotness prediction dataset to obtain a training set, input the training set into a pre-established streaming event hotness prediction model based on a multi-view neural Hawkes process, train based on a negative log maximum likelihood loss function, and obtain the trained streaming event hotness prediction model based on the multi-view neural Hawkes process.

[0057] The event hotness prediction module is configured to input the streaming event hotness prediction dataset into the trained streaming event hotness prediction model based on the multi-view neural Hawkes process, and output an event hotness prediction result.

[0058] In another aspect of the present application, in order to achieve the above-mentioned purpose, a terminal device is disclosed, which comprises a memory, a processor, and a computer program stored in the memory and capable of running on the processor, the memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, the event hotness prediction method based on the multi-view neural Hawkes process is adopted.

[0059] In another aspect of the present application, in order to achieve the above-mentioned purpose, a computer readable storage medium is disclosed, which stores a computer program, and when the processor loads and executes the computer program, the event hotness prediction method based on the multi-view neural Hawkes process is adopted.

[0060] The present application has the following advantages:

[0061] The present application aims to solve the problem of incomplete hidden association mining, and the multi-view feature encoder is used for combined coding to capture the hidden association between different information and hotness changes under different perspectives, thereby enhancing the representation accuracy of the features. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without creative labor;

[0063] Figure 1 is a schematic diagram of the method of the present application;

[0064] Figure 2 is a structural diagram of the streaming event hotness prediction model based on the multi-view neural Hawkes process of the present application;

[0065] Figure 3 is a schematic diagram of the system structure of the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0067] Embodiment one:

[0068] As shown in the figure, the event heat prediction method based on the multi-view neural Hawkes process comprises the following steps: Figure 1

[0069] S101: receiving a streaming event heat prediction dataset, the streaming event heat prediction dataset being constructed based on event related data, wherein the event related data comprises post information, comment information and heat information;

[0070] Specifically, 464 events, 9.15 million posts and 1.03 million comment information are collected;

[0071] Part of the data in the dataset is divided into a training set; in this embodiment, 376 events are divided into a training set;

[0072] S102: extracting the streaming event heat prediction dataset to obtain a training set, inputting the training set into a pre-established streaming event heat prediction model based on the multi-view neural Hawkes process, training based on a negative log maximum likelihood loss function, and obtaining a trained streaming event heat prediction model based on the multi-view neural Hawkes process;

[0073] The pre-established streaming event heat prediction model based on the multi-view neural Hawkes process comprises a multi-view feature encoder and a Hawkes process driven predictor;

[0074] The multi-view feature encoder comprises a time series view encoding, a post view encoding and a comment view encoding; the time series view encoding takes a historical heat sequence of the heat information as input and outputs time series features; the post view encoding takes text of the post information as input and outputs post features; the comment view encoding takes text of the comment information as input and outputs comment features;

[0075] The Hawkes process driven predictor takes the time series features, the post features and the comment features as input and outputs predicted event heat in the future days, wherein is a positive integer.​

[0076] The specific steps of the time sequence view coding module in the multi-view feature encoder are as follows:

[0077] Step 1: The sliding window strategy is used to obtain the heat values of the past T days. For a sequence longer than T days, the sequence is truncated, and for a sequence shorter than T days, the sequence is reversed and zero-padded. In this embodiment, T is 7.

[0078] Step 2: The change amount of heat per day in T days is calculated to obtain a heat change sequence with a length of T-1.

[0079] Step 3:

[0080] The heat sequence is standardized by logarithmic change:

[0081]

[0082] wherein, is the change amount of heat on the tth day, is the standardized heat change amount, sign(·) is a sign function, and |·| is an absolute value function.

[0083] Step 4: Each standardized heat change amount in the heat sequence is mapped to a learnable heat representation through an embedding layer , d is the dimension of the representation; in this embodiment, d is 64.

[0084] Step 5:

[0085] A learnable position encoding is added to each heat representation in the heat sequence:

[0086]

[0087] wherein, represents the learnable position encoding for the heat change amount on the tth day, represents the value of the first position in the feature, the value of the first position in the feature, represents the heat feature after the position encoding is added; Step 6:

[0088] Each heat feature after the position encoding is added in the heat sequence is fused:

[0089]

[0090]

[0091] wherein, ​​The feature obtained by encoding the final time sequence perspective coding module is denoted as

[0092] The specific steps of the post perspective coding module in the multi-perspective feature encoder are as follows:

[0093] Step 1: Randomly sampling posts from the posts collected on the same day and related to the event, wherein is a positive integer; in this embodiment, is 100;

[0094] Step 2: Encoding the features of each post by using a pre-trained text semantic feature encoder and a text sentiment feature encoder, wherein the semantic feature of the i-th post is denoted as and the sentiment feature is denoted as ;

[0095] Step 3:

[0096] The mutual attention mechanism is used to enhance the semantic feature and the sentiment feature:

[0097]

[0098] wherein , is the enhanced semantic feature and the enhanced sentiment feature; , , is the semantic feature obtained by linear mapping of the query feature, the index feature and the value feature; , , is the sentiment feature obtained by linear mapping of the query feature, the index feature and the value feature; and softmax(·) is a softmax function.

[0099] Step 4:

[0100] Fusing the features of each post:

[0101]

[0102] wherein is the feature obtained by encoding the final post perspective coding module, is the enhanced semantic feature of the n-th post, is the enhanced sentiment feature of the n-th post.

[0103] The specific steps of the comment perspective coding module in the multi-perspective feature encoder are as follows:

[0104] Step one: randomly sampling one comment from the event-related comments collected on the same day, where, is a positive integer; in this embodiment, is 50;

[0105] Step two: encoding the features of each comment using a pre-trained text semantic feature encoder and a text sentiment feature encoder, where the semantic feature of the ith comment is and the sentiment feature is ;

[0106] Step three:

[0107] Using a mutual attention mechanism to enhance the semantic feature and the sentiment feature:

[0108]

[0109] where, , is the enhanced semantic feature and the sentiment feature; , , is the semantic feature of the query feature, the index feature, and the value feature obtained by linear mapping; , , is the sentiment feature of the query feature, the index feature, and the value feature obtained by linear mapping; softmax(·) is a softmax function;

[0110] Step four:

[0111] Fusing the features of each comment:

[0112]

[0113] where, is the feature encoded by the final comment perspective encoding module, is the enhanced semantic feature of the nth comment, is the enhanced sentiment feature of the nth comment. The Hawkes process-driven predictor takes the time series features, post features, and comment features as input and outputs the predicted event heat in the next days, where is a positive integer; in this embodiment, is 3 days; the specific steps are:

[0114] Step one:

[0115] According to the time series features, calculate the change amount of the heat under the influence of the time series perspective: ​

[0116]

[0117] wherein, , is a weight mapping matrix, , is a bias matrix; sign(·) is a sign function; |·| is an absolute value function; is an observed heat value of an event at time t; is a predicted change amount of heat under the influence of a time series perspective at time t+1;

[0118] Step two:

[0119] According to the post features, the change amount of heat under the influence of a post perspective is calculated:

[0120]

[0121] wherein, , is a weight mapping matrix, , is a bias matrix; is a learnable time series feature; is a predicted change amount of heat under the influence of a post perspective at time t+1;

[0122] Step three: in the same way as step two, the change amount of heat under the influence of a comment perspective is calculated ;

[0123] Step four:

[0124] Comprehensive influence of various perspectives, predict the heat of the event of the next day:

[0125]

[0126] wherein, is the predicted heat of the event of the next day; according to the formula, step-by-step cycle prediction is performed to obtain the heat related to the event in the future day, is a positive integer.

[0127] S103: input the streaming event heat prediction dataset into the trained multi-perspective neural Hawkes process-based streaming event heat prediction model, and output to obtain the event heat prediction result.

[0128] Specifically, the following embodiments are used to further illustrate the present application:

[0129] In this embodiment, the performance of the event heat prediction method of the present application and the event heat prediction method in the prior art is compared, and the process of the comparison experiment is as follows:

[0130] The present embodiment reproduces the existing mainstream event heat prediction method based on the Pytorch framework, and develops the event heat prediction method based on the multi-view neural Hawkes process proposed in the present application. The parameters in all comparative methods follow the best setting of the reported parameters, and the Adam optimizer is used to train the event heat prediction model. The event heat prediction test process proposed in the present application is as described in Embodiment 1.

[0131] The experimental data set is tested by the data not included in the training set in Embodiment 1.

[0132] All experiments are performed on an NVIDIA Tesla V100 GPU.

[0133] The comparative test results are shown in Table 1, and it can be seen that:

[0134] The performance of the present method is obviously better than that of the existing method on multiple public data sets. Through the combination coding of the multi-view feature encoder, the hidden correlation between different information and heat changes under different perspectives can be captured, which can effectively enhance the representation accuracy of the features. At the same time, the predictor driven by the Hawkes process introduces the prior propagation dynamics knowledge, ensures that the prediction result conforms to the prior law, and further enhances the prediction accuracy.

[0135] Table 1 Comparative test results

[0136]

[0137] Figure 2The detection method is more intuitive. For the event to be detected, first, through the multi-view feature encoder, the features of the event are modeled from the time sequence view, the post view and the comment view; the time sequence view takes the historical heat sequence (the current heat is 4152, and the future heat to be predicted is empty) as the input, passes through the time sequence encoder, and outputs the time sequence feature; the post view takes the event-related post content as the input, passes through the post encoder, and outputs the post feature; the comment view takes the event-related comment content as the input, passes through the comment encoder, and outputs the comment feature; the obtained time sequence feature, post feature and comment feature are analyzed and synthesized by the Hox process driven predictor to output the step-by-step prediction result: first, the prediction value 8938 of the heat after 1 day is obtained, and the prediction value is updated to the heat sequence to obtain the updated heat sequence (the current heat is 4152, and the heat after 1 day is 8938); then, according to the updated heat sequence, the prediction process is repeated to obtain the prediction value 16129 of the heat after 2 days, and the prediction value is updated to the heat sequence to obtain the updated heat sequence (the current heat is 4152, the heat after 1 day is 8938, and the heat after 2 days is 16129); finally, according to the updated heat sequence, the prediction process is repeated to obtain the prediction value 6047 of the heat after 3 days; the prediction value of the heat after 1 day, the prediction value of the heat after 2 days and the prediction value of the heat after 3 days are spliced to obtain the complete heat sequence (the current heat is 4152, the heat after 1 day is 8938, the heat after 2 days is 16129, and the heat after 3 days is 6047).

[0138] In order to achieve the above purpose, as shown in Figure 3 based on the basis of example one, the application discloses an event heat prediction system based on multi-view neural Hox process, comprising:

[0139] The data acquisition module 11 is used for receiving a streaming event heat prediction data set, and the streaming event heat prediction data set is constructed based on event-related data, wherein the event-related data includes post information, comment information and heat information;

[0140] The model training module 12 is used for extracting the streaming event heat prediction data set to obtain a training set, inputting the training set into a pre-established streaming event heat prediction model based on multi-view neural Hox process, training based on a negative log maximum likelihood loss function, and obtaining a trained streaming event heat prediction model based on multi-view neural Hox process;

[0141] The event heat prediction module 13 is used for inputting the streaming event heat prediction data set into the trained streaming event heat prediction model based on multi-view neural Hox process, and outputting an event heat prediction result.

[0142] Based on the same inventive concept, the present application further provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the program comprises program instructions, and the processor is configured to execute the program instructions stored in the memory. 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), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are configured to implement one or more instructions, and are specifically configured to load and execute one or more instructions in the computer storage medium to implement the above method.

[0143] It needs to be further explained that, based on the same inventive concept, the present application further provides a computer storage medium, which stores a computer program, and the computer program is executed by the processor to perform the above method. The storage medium can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, be but is not limited to an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples (non-exhaustive list) of the computer readable storage medium include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component.

[0144] In the description of the present application, the description of the terms "one embodiment", "an example", "a specific example" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0145] The foregoing presents and describes the basic principles, main features and advantages of the present disclosure. It should be understood by those skilled in the art that the present disclosure is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only to illustrate the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, various changes and improvements can be made to the present disclosure, and all these changes and improvements fall within the scope of the present disclosure.

Claims

1. An event heat prediction method based on a multi-view neural Hawkes process, characterized in that, The method comprises the following steps: Receiving a streaming event heat prediction data set, which is constructed based on event-related data, wherein the event-related data comprises post information, comment information and heat information; Extracting the streaming event heat prediction data set to obtain a training set, inputting the training set into a pre-established streaming event heat prediction model based on a multi-view neural Hawkes process, training based on a negative log maximum likelihood loss function, and obtaining a trained streaming event heat prediction model based on the multi-view neural Hawkes process; The pre-established streaming event heat prediction model based on the multi-view neural Hawkes process comprises a multi-view feature encoder and a Hawkes process driven predictor; The multi-view feature encoder comprises a time sequence view encoding, a post view encoding and a comment view encoding; the time sequence view encoding takes a historical heat sequence of the heat information as input and outputs time sequence features; the post view encoding takes text of the post information as input and outputs post features; and the comment view encoding takes text of the comment information as input and outputs comment features; The Hox process-driven predictor takes time sequence features, post features, and comment features as inputs and outputs predicted future event heat of the day, wherein is a positive integer; Inputting the streaming event heat prediction data set into the trained streaming event heat prediction model based on the multi-view neural Hawkes process, and outputting an event heat prediction result; The Hawkes process driven predictor comprises: According to the time sequence features, calculating a change amount of the heat under the influence of the time sequence view: wherein, , is a weight mapping matrix, , is a bias matrix; sign(·) is a sign function; |·| is an absolute value function; is an observed heat value of an event at time t; is a predicted change in heat under the influence of a time-series perspective at time t+1. According to the post features, calculating a change amount of the heat under the influence of the post view: wherein, , is a weight mapping matrix, , is a bias matrix; is a learnable temporal feature; is a predicted change in heat under the influence of the post's perspective at time t+1. The change amount of the heat under the influence of the comment perspective is calculated in the same manner as the change amount of the heat under the influence of the post perspective ; Integrating the influences of various views to predict the heat of the event on the next day: wherein, is the predicted heat of the next day event; the stepwise loop prediction is obtained according to the formula, and the future heat of the event related to the day, is a positive integer. 2.The multi-view neural Hawkes process based event heat prediction method of claim 1, wherein, The process of the time sequence view encoding comprises: Using a sliding window strategy to obtain heat values of the last T days, truncating sequences longer than T days, and reversing and zero-padding sequences shorter than T days; Calculating a change amount of the heat of each day within T days to obtain a heat change sequence with a length of T-1; Standardizing the heat sequence through logarithmic change: wherein, is a change amount of the heat on the t-th day, is a standardized heat change amount, sign(·) is a sign function, and |·| is an absolute value function. each standardized heat change in the heat sequence is mapped to a learnable heat representation through an embedding layer d is the dimension of the representation; Adding learnable position encoding to each heat representation in the heat sequence: in, This represents the learnable location code for the change in heat on day t. express The first feature The value of the position, The first feature The value of the position, This indicates the popularity feature after adding location encoding; Fusing heat features of each heat representation in the heat sequence after adding position encoding: wherein, are features encoded by the final temporal view coding module, and Transform (·) is a Transformer encoder function. 3.The multi-view neural Hawkes process based event heat prediction method of claim 2, wherein, The process of the post view encoding comprises: randomly sampling from event-related posts collected on the day posts, wherein, is a positive integer; Each post is encoded with a pre-trained text semantic feature encoder and a text sentiment feature encoder. For the i-th post, its semantic feature is , and its sentiment feature is ; Using a mutual attention mechanism to enhance semantic features and emotional features: wherein, , is the enhanced semantic feature and the emotional feature; , , is the semantic feature the query feature, the index feature and the value feature obtained through linear mapping; , , is the emotional feature the query feature, the index feature and the value feature obtained through linear mapping; softmax(·) is a softmax function; Fusing features of each post: wherein, is the feature encoded by the final post perspective encoding module, is the enhanced semantic feature of the n-th post, is the enhanced sentiment feature of the n-th post. 4.The multi-view neural Hawkes process based event heat prediction method of claim 3, wherein, The encoding process of the comment view comprises: randomly sampling from event-related comments collected on the day comments, wherein, is a positive integer; Each review is encoded with a pre-trained text semantic feature encoder and a text sentiment feature encoder. For the i-th review, its semantic feature is , and its sentiment feature is ; Using a mutual attention mechanism to enhance semantic features and emotional features: wherein, , are enhanced semantic features and emotional features; , , are semantic features query features, index features and value features obtained through linear mapping; softmax(·) is a softmax function; , , are emotional features query features, index features and value features obtained through linear mapping; softmax(·) is a softmax function; Fusing features of each comment: wherein, is the feature encoded by the final review perspective encoding module, is the enhanced semantic feature of the n-th review, is the enhanced sentiment feature of the n-th review.

5. An event heat prediction system based on multi-view neural Hawkes process, adopting the event heat prediction method based on multi-view neural Hawkes process in any one of claims 1 to 4, characterized in that, Comprise: A data acquisition module is configured to receive a streaming event heat prediction data set, which is constructed based on event-related data, wherein the event-related data comprises post information, comment information and heat information; A model training module is configured to extract the streaming event heat prediction data set to obtain a training set, input the training set into a pre-established streaming event heat prediction model based on a multi-view neural Hawkes process, train based on a negative log maximum likelihood loss function, and obtain a trained streaming event heat prediction model based on the multi-view neural Hawkes process. An event heat prediction module is configured to input a streaming event heat prediction dataset into the trained multi-view neural Hawkes process-based streaming event heat prediction model, and output an event heat prediction result.

6. A terminal device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, The memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, the multi-view neural Hawkes process-based event heat prediction method in any one of claims 1 to 4 is adopted.

7. A computer-readable storage medium having stored therein a computer program, characterized in that, The computer program is loaded and executed by the processor, and the multi-view neural Hawkes process-based event heat prediction method in any one of claims 1 to 4 is adopted.

Images