Event popularity prediction method and system based on multi-view neuro-hokes process

Abstract

The invention discloses an event popularity prediction method and system based on a multi-view neural houkes process, and relates to the technical field of data mining, and the method comprises the steps: receiving a streaming event popularity prediction data set which is constructed based on event related data, the event related data comprises post information, comment information and popularity information; extracting the streaming event popularity prediction data set to obtain a training set, inputting the training set into a pre-established streaming event popularity prediction model based on the multi-view neural houkes process, and performing training based on a negative logarithm maximum likelihood loss function to obtain a streaming event popularity prediction model; obtaining a trained streaming event popularity prediction model based on the multi-view neural houkes process; and inputting the streaming event popularity prediction data set into the trained streaming event popularity prediction model based on the multi-view neural houkes process, and outputting to obtain an event popularity prediction result, so that the prediction precision is enhanced.

Description

Technical Field

[0001] This invention relates to the field of data mining technology, specifically to a method and system for predicting event heat based on multi-perspective neural Hawkes processes. Background Technology

[0002] Social media platforms have fundamentally changed the way information is disseminated. Platforms like Weibo and Douyin allow users to quickly access breaking news, discuss social events, and share their opinions on products. The widespread interaction among users on these platforms reflects the collective behavior of different groups. Against this backdrop, event trend prediction has begun to receive widespread attention from researchers because of its significant impact on content distribution and digital marketing.

[0003] Early research primarily viewed popularity prediction as a post-level classification task (i.e., predicting whether a post will become popular) or a regression task (i.e., predicting the maximum popularity of a post), and used statistical methods for prediction. With the rise of deep learning, existing methods have begun to utilize neural networks for automatic feature extraction and popularity prediction. From the perspective of existing methods, they can be roughly divided into two categories: (1) Methods based on propagation paths: These predict which users are likely to receive the information next and estimate the popularity by counting the number of users who receive the information. (2) Methods based on content understanding: These predict popularity based on the semantic information of the content, assuming that similar content tends to exhibit similar popularity. Although these methods have achieved initial success, they have not fully modeled the influence between different information related to the event and changes in popularity, and the prediction results have high uncertainty, making it difficult to guarantee reasonable prediction results. Summary of the Invention

[0004] To address the shortcomings mentioned in the background art, the present invention aims to provide an event heat prediction method and system based on multi-perspective neural Hawkes processes.

[0005] Firstly, the objective of this invention can be achieved through the following technical solution: an event heat prediction method based on multi-perspective neural Hawkes processes, the method comprising the following steps: Receive a streaming event popularity prediction dataset, which is constructed based on event-related data, including: post information, comment information, and popularity information; The streaming event popularity prediction dataset is extracted to obtain the training set. The training set is then input into the pre-established streaming event popularity prediction model based on the multi-view neural Hawkes process. The model is trained based on the negative log maximum likelihood loss function to obtain the trained streaming event popularity prediction model based on the multi-view neural Hawkes process. The streaming event popularity prediction dataset is input into the trained streaming event popularity prediction model based on multi-view neural Hawkes process, and the event popularity prediction results are output.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established streaming event heat prediction model based on multi-view neural Hawkes process includes a multi-view feature encoder and a Hawkes process-driven predictor. The multi-view feature encoder includes temporal view encoding, post view encoding, and comment view encoding; temporal view encoding takes the historical popularity sequence of popularity information as input and outputs temporal features; post view encoding takes the text of post information as input and outputs post features; comment view encoding takes the text of comment information as input and outputs comment features. The Hawkes process-driven predictor takes time-series features, post features, and comment features as input and outputs a predicted future. The popularity of the event that day, among which It is a positive integer.

[0007] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the temporal perspective encoding process, comprising: A sliding window strategy is used to obtain the historical heat value of T days. For sequences longer than T days, they are truncated, and for sequences shorter than T days, zeros are padded in reverse order. Calculate the daily change in heat over T days to obtain a heat change sequence of length T-1; The heat series is standardized by logarithmic transformation: in, Let be the change in heat on day t. For the standardized change in heat, sign(·) is the sign function, and |·| is the 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; Add a learnable positional 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; Each heat feature in the fused heat sequence after adding positional encoding: in, The features encoded by the final temporal view encoding module are represented by Transformer(), which is the Transformer encoder function.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of encoding the post perspective, including: Random sampling was conducted from posts related to the event collected that day. 10 posts, of which It is a positive integer; Each post's features are encoded using a pre-trained text semantic feature encoder and a text sentiment feature encoder. For the i-th post, its semantic features are: Emotional characteristics are ; Employing mutual attention mechanisms to enhance semantic and sentiment features: in, , For enhanced semantic and sentiment features; , , semantic features The query features, index features, and value features obtained through linear mapping; , , Emotional characteristics The query features, index features, and value features are obtained through linear mapping; softmax(·) is the softmax function; Integrate the characteristics of each post: in, Features encoded by the final post perspective encoding module The enhanced semantic features for the nth post, The enhanced sentiment features for the nth post.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of encoding the comment perspective, comprising: Random sampling was conducted from comments related to the event collected that day. 1 comment, of which, It is a positive integer; Each comment's features are encoded using a pre-trained text semantic feature encoder and a text sentiment feature encoder. For the i-th comment, its semantic features are: Emotional characteristics are ; Employing mutual attention mechanisms to enhance semantic and sentiment features: in, , For enhanced semantic and sentiment features; , , semantic features The query features, index features, and value features obtained through linear mapping; , , Emotional characteristics The query features, index features, and value features are obtained through linear mapping; softmax(·) is the softmax function; Integrate the characteristics of each comment: in, The features encoded by the final comment perspective encoding module For the enhanced semantic features of the nth comment, The enhanced sentiment profile for the nth comment.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the Hawkes process-driven predictor as follows: Based on the temporal characteristics, calculate the change in popularity under the influence of the temporal perspective: in, , It is a weight mapping matrix. , It is the bias matrix; sign(·) is the sign function; |·| is the absolute value function; It is the observed heat value of the event at time t; It is the change in heat under the influence of the time series perspective at the predicted time t+1. Based on post characteristics, calculate the change in popularity under the influence of post perspective: in, , It is a weight mapping matrix. , It is the bias matrix; It is a learnable temporal feature; It is the predicted change in popularity at time t+1 under the influence of the post's perspective; Using the same calculation method as for the change in popularity under the influence of the post's perspective, the change in popularity under the influence of the comment's perspective is calculated. ; Based on the influence of various perspectives, predict the popularity of the event next day: in, To predict the popularity of events for the next day; a step-by-step cyclical prediction is performed based on a formula to obtain future... The popularity related to the Tian incident It is a positive integer.

[0011] Secondly, in order to achieve the above objectives, this invention discloses an event heat prediction system based on multi-perspective neural Hawkes processes, comprising: The data acquisition module is used to receive a streaming event popularity prediction dataset, which is constructed based on event-related data, including: post information, comment information, and popularity information. The model training module is used to extract the streaming event popularity prediction dataset to obtain the training set. The training set is then input into the pre-established streaming event popularity prediction model based on the multi-view neural Hawkes process. The model is trained based on the negative log maximum likelihood loss function to obtain the trained streaming event popularity prediction model based on the multi-view neural Hawkes process. The event popularity prediction module is used to input the streaming event popularity prediction dataset into the trained streaming event popularity prediction model based on multi-view neural Hawkes process, and output the event popularity prediction results.

[0012] In another aspect of the present invention, in order to achieve the above-mentioned objective, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores the computer program capable of running on the processor. When the processor loads and executes the computer program, it employs the event heat prediction method based on the multi-perspective neural Hawkes process as described above.

[0013] In another aspect of the present invention, in order to achieve the above-mentioned objective, a computer-readable storage medium is disclosed, wherein a computer program is stored in the computer program, and when the computer program is loaded and executed by a processor, the event heat prediction method based on the multi-perspective neural Hawkes process described above is employed.

[0014] The beneficial effects of this invention are: This invention addresses the problem of incomplete discovery of hidden relationships by using a multi-view feature encoder to combine encodings and capture hidden relationships between different information and popularity changes from different perspectives, thereby enhancing the accuracy of feature representation. To address the problem of inaccurate prediction results, this invention introduces prior propagation dynamics knowledge through a Hawkes process-driven predictor, ensuring that the prediction results conform to prior laws and enhancing prediction accuracy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a structural diagram of the streaming event heat prediction model based on multi-perspective neural Hawkes processes of this invention; Figure 3 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1: like Figure 1 As shown, an event heat prediction method based on multi-perspective neural Hawkes processes includes the following steps: S101: Receive streaming event popularity prediction dataset, which is constructed based on event-related data, including: post information, comment information and popularity information; Specifically, a total of 464 events, 9.15 million posts, and 1.03 million comments were collected. A portion of the dataset is divided into a training set; in this embodiment, 376 events are divided into the training set. S102: Extract the streaming event popularity prediction dataset to obtain the training set, input the training set into the pre-established streaming event popularity prediction model based on the multi-view neural Hawkes process, train it based on the negative log maximum likelihood loss function, and obtain the trained streaming event popularity prediction model based on the multi-view neural Hawkes process. The pre-established streaming event heat prediction model based on multi-view neural Hawkes process includes a multi-view feature encoder and a Hawkes process-driven predictor. The multi-view feature encoder includes temporal view encoding, post view encoding, and comment view encoding; temporal view encoding takes the historical popularity sequence of popularity information as input and outputs temporal features; post view encoding takes the text of post information as input and outputs post features; comment view encoding takes the text of comment information as input and outputs comment features. The Hawkes process-driven predictor takes time-series features, post features, and comment features as input and outputs a predicted future. The popularity of the event that day, among which It is a positive integer.

[0018] The specific steps of the temporal perspective encoding module in the multi-view feature encoder are as follows: Step 1: Use a sliding window strategy to obtain the historical heat value for T days. Truncate sequences longer than T days and pad with zeros in reverse order for sequences shorter than T days. In this embodiment, T is 7. Step 2: Calculate the daily change in heat over T days to obtain a heat change sequence of length T-1; Step 3: The heat series is standardized by logarithmic transformation: in, Let be the change in heat on day t. For the standardized change in heat, sign(·) is the sign function, and |·| is the absolute value function; Step 4: Map each standardized change in popularity in the popularity sequence to a learnable popularity representation through an embedding layer. d is the dimension of the representation; in this embodiment, d is 64. Step 5: Add a learnable positional 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; Step Six: Each heat feature in the fused heat sequence after adding positional encoding: in, The features encoded by the final temporal view encoding module are represented by Transformer(), which is the Transformer encoder function.

[0019] The specific steps of the post perspective encoding module in the multi-view feature encoder are as follows: Step 1: Randomly sample posts related to the event collected that day. 10 posts, of which It is a positive integer; in this embodiment, It is 100; Step 2: Encode the features of each post using a pre-trained text semantic feature encoder and text sentiment feature encoder. For the i-th post, its semantic features are as follows: Emotional characteristics are ; Step 3: Employing mutual attention mechanisms to enhance semantic and sentiment features: in, , For enhanced semantic and sentiment features; , , semantic features The query features, index features, and value features obtained through linear mapping; , , Emotional characteristics The query features, index features, and value features are obtained through linear mapping; softmax(·) is the softmax function; Step Four: Integrate the characteristics of each post: in, Features encoded by the final post perspective encoding module The enhanced semantic features for the nth post, The enhanced sentiment features for the nth post.

[0020] The specific steps of the comment perspective encoding module in the multi-view feature encoder are as follows: Step 1: Randomly sample comments related to the event collected that day. 1 comment, of which, It is a positive integer; in this embodiment, It is 50; Step 2: Encode the features of each comment using a pre-trained text semantic feature encoder and text sentiment feature encoder. For the i-th comment, its semantic features are as follows: Emotional characteristics are ; Step 3: Employing mutual attention mechanisms to enhance semantic and sentiment features: in, , For enhanced semantic and sentiment features; , , semantic features The query features, index features, and value features obtained through linear mapping; , , Emotional characteristics The query features, index features, and value features are obtained through linear mapping; softmax(·) is the softmax function; Step Four: Integrate the characteristics of each comment: in, The features encoded by the final comment perspective encoding module For the enhanced semantic features of the nth comment, The enhanced sentiment feature for the nth comment. The Hawkes process-driven predictor takes temporal features, post features, and comment features as input and outputs a predicted future... The popularity of the event that day, among which It is a positive integer. In this embodiment, It takes 3 days; the specific steps are as follows: Step 1: Based on the temporal characteristics, calculate the change in popularity under the influence of the temporal perspective: in, , It is a weight mapping matrix. , It is the bias matrix; sign(·) is the sign function; |·| is the absolute value function; It is the observed heat value of the event at time t; It is the change in heat under the influence of the time series perspective at the predicted time t+1. Step Two: Based on post characteristics, calculate the change in popularity under the influence of post perspective: in, , It is a weight mapping matrix. , It is the bias matrix; It is a learnable temporal feature; It is the predicted change in popularity at time t+1 under the influence of the post's perspective; Step 3: Using the same method as in Step 2, calculate the change in popularity under the influence of the comment perspective. ; Step Four: Based on the influence of various perspectives, predict the popularity of the event next day: in, To predict the popularity of events for the next day; a step-by-step cyclical prediction is performed based on a formula to obtain future... The popularity related to the Tian incident It is a positive integer.

[0021] S103: Input the streaming event popularity prediction dataset into the trained streaming event popularity prediction model based on multi-view neural Hawkes process, and output the event popularity prediction results.

[0022] Specifically, the present invention will be further illustrated below through embodiments: In this embodiment, the performance of the event popularity prediction method of the present invention is compared with that of the prior art event popularity prediction method. The comparison experiment is as follows: This embodiment reproduces existing mainstream event popularity prediction methods based on the PyTorch framework and develops the event popularity prediction method based on multi-view neural Hawkes processes proposed in this invention. The parameters in all comparative methods follow the reported optimal settings, and the Adam optimizer is used to train the event popularity prediction model. The event popularity prediction testing process proposed in this invention is as described in Example 1.

[0023] The experimental dataset used for testing was data that was not included in the training set in Example 1.

[0024] All experiments were conducted on an NVIDIA Tesla V100 GPU.

[0025] The results of the comparative experiment are shown in Table 1. It can be seen that: Our method significantly outperforms existing methods on multiple public datasets. By combining encoding through a multi-view feature encoder, we capture the hidden correlations between different information and popularity changes from different perspectives, which can effectively enhance the accuracy of feature representation. At the same time, by introducing prior propagation dynamics knowledge through a Hawkes process-driven predictor, we ensure that the prediction results conform to prior laws, further enhancing the prediction accuracy.

[0026] Table 1 Comparative Experimental Results Figure 2 This provides a more intuitive demonstration of the detection method. For the event to be detected, a multi-view feature encoder is first used to model the event's features from the perspectives of time sequence, post, and comment. The time sequence perspective takes the historical popularity sequence (current popularity is 4152, and the predicted future popularity is empty) as input, and the time sequence encoder outputs time sequence features. The post perspective takes event-related post content (e.g., bad news: Black Monkey has no DLC...) as input, and the post encoder outputs post features. The comment perspective takes event-related comment content (e.g., ① What a pity...; ② Thematic overlap is a characteristic of 3A productions; ③ In the end, only a CG was released...; ④ To be able to calm down and devote oneself to a new work in the face of huge traffic and popularity, this mentality and thinking are truly those of people who achieve great things) as input, and the comment encoder outputs comment features. The obtained time sequence features, post features, and comment features are analyzed and synthesized by a Hawkes process-driven predictor to output a step-by-step feature. The prediction results are as follows: First, the predicted value of the popularity one day later is 8938, and this predicted value is updated to the popularity sequence to obtain the updated popularity sequence (current popularity is 4152, popularity one day later is 8938); then, based on the updated popularity sequence, the prediction process is repeated to obtain the predicted value of the popularity two days later is 16129, and this predicted value is updated to the popularity sequence to obtain the updated popularity sequence (current popularity is 4152, popularity one day later is 8938, popularity two days later is 16129); finally, based on the updated popularity sequence, the prediction process is repeated to obtain the predicted value of the popularity three days later is 6047; the predicted values ​​of the popularity one day later, two days later, and three days later are concatenated to obtain the complete popularity sequence (current popularity is 4152, popularity one day later is 8938, popularity two days later is 16129, popularity three days later is 6047).

[0027] Example 2: To achieve the above objective, such as Figure 3As shown, based on Embodiment 1, this invention discloses an event heat prediction system based on multi-perspective neural Hawkes processes, including: The data acquisition module 11 is used to receive a streaming event popularity prediction dataset, which is constructed based on event-related data, including post information, comment information, and popularity information. The model training module 12 is used to extract the streaming event heat prediction dataset to obtain the training set, input the training set into the pre-established streaming event heat prediction model based on the multi-view neural Hawkes process, and train it based on the negative log maximum likelihood loss function to obtain the trained streaming event heat prediction model based on the multi-view neural Hawkes process. The event popularity prediction module 13 is used to input the streaming event popularity prediction dataset into the trained streaming event popularity prediction model based on multi-view neural Hawkes process, and output the event popularity prediction results.

[0028] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0029] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be 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 can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, 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, apparatus, or device.

[0030] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0031] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

Claims

1. An event heat prediction method based on multi-perspective neural Hawkes processes, characterized in that, The method includes the following steps: Receive a streaming event popularity prediction dataset, which is constructed based on event-related data, including: post information, comment information, and popularity information; The streaming event popularity prediction dataset is extracted to obtain the training set. The training set is then input into the pre-established streaming event popularity prediction model based on the multi-view neural Hawkes process. The model is trained based on the negative log maximum likelihood loss function to obtain the trained streaming event popularity prediction model based on the multi-view neural Hawkes process. The streaming event popularity prediction dataset is input into the trained streaming event popularity prediction model based on multi-view neural Hawkes process, and the event popularity prediction results are output.

2. The event heat prediction method based on multi-perspective neural Hawkes processes according to claim 1, characterized in that, The pre-established streaming event heat prediction model based on multi-view neural Hawkes process includes a multi-view feature encoder and a Hawkes process-driven predictor. The multi-view feature encoder includes temporal view encoding, post view encoding, and comment view encoding; Temporal-view coding takes the historical popularity sequence of popularity information as input and outputs temporal features; Post perspective encoding takes the text of post information as input and outputs post features; Comment perspective encoding takes the text of comment information as input and outputs comment features; The Hawkes process-driven predictor takes time-series features, post features, and comment features as input and outputs a predicted future. The popularity of the event that day, among which It is a positive integer.

3. The event heat prediction method based on multi-perspective neural Hawkes processes according to claim 2, characterized in that, The temporal perspective encoding process includes: A sliding window strategy is used to obtain the historical heat value of T days. For sequences longer than T days, they are truncated, and for sequences shorter than T days, zeros are padded in reverse order. Calculate the daily change in heat over T days to obtain a heat change sequence of length T-1; The heat series is standardized by logarithmic transformation: in, Let be the change in heat on day t. For the standardized change in heat, sign(·) is the sign function, and |·| is the 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; Add a learnable positional 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; Each heat feature in the fused heat sequence after adding positional encoding: in, The features encoded by the final temporal view encoding module are represented by Transformer(), which is the Transformer encoder function.

4. The event heat prediction method based on multi-perspective neural Hawkes processes according to claim 3, characterized in that, The process of encoding the post's perspective includes: Random sampling was conducted from posts related to the event collected that day. 10 posts, of which It is a positive integer; Each post's features are encoded using a pre-trained text semantic feature encoder and a text sentiment feature encoder. For the i-th post, its semantic features are: Emotional characteristics are ; Employing mutual attention mechanisms to enhance semantic and sentiment features: in, , For enhanced semantic and sentiment features; , , semantic features The query features, index features, and value features obtained through linear mapping; , , Emotional characteristics The query features, index features, and value features are obtained through linear mapping; softmax(·) is the softmax function; Integrate the characteristics of each post: in, Features encoded by the final post perspective encoding module The enhanced semantic features for the nth post, The enhanced sentiment feature for the nth post.

5. The event heat prediction method based on multi-perspective neural Hawkes processes according to claim 4, characterized in that, The encoding process for the commentary perspective is as follows: Random sampling was conducted from comments related to the event collected that day. 1 comment, of which, It is a positive integer; Each comment's features are encoded using a pre-trained text semantic feature encoder and a text sentiment feature encoder. For the i-th comment, its semantic features are: Emotional characteristics are ; Employing mutual attention mechanisms to enhance semantic and sentiment features: in, , For enhanced semantic and sentiment features; , , semantic features The query features, index features, and value features obtained through linear mapping; , , Emotional characteristics The query features, index features, and value features are obtained through linear mapping; softmax(·) is the softmax function; Integrate the characteristics of each comment: in, The features encoded by the final comment perspective encoding module The enhanced semantic features for the nth comment. The enhanced sentiment profile for the nth comment.

6. The event heat prediction method based on multi-perspective neural Hawkes processes according to claim 5, characterized in that, The Hawkes process-driven predictor includes: Based on the temporal characteristics, calculate the change in popularity under the influence of the temporal perspective: in, , It is a weight mapping matrix. , It is the bias matrix; sign(·) is the sign function; |·| is the absolute value function; It is the observed heat value of the event at time t; It is the change in heat under the influence of the time series perspective at the predicted time t+1. Based on post characteristics, calculate the change in popularity under the influence of post perspective: in, , It is a weight mapping matrix. , It is the bias matrix; It is a learnable temporal feature; It is the predicted change in popularity at time t+1 under the influence of the post's perspective; Using the same calculation method as for the change in popularity under the influence of the post's perspective, the change in popularity under the influence of the comment's perspective is calculated. ; Based on the influence of various perspectives, predict the popularity of the event next day: in, To predict the popularity of events for the next day; a step-by-step cyclical prediction is performed based on a formula to obtain future... The popularity related to the Tian incident It is a positive integer.

7. An event heat prediction system based on multi-perspective neural Hawkes processes, employing the event heat prediction method based on multi-perspective neural Hawkes processes as described in any one of claims 1 to 6, characterized in that, include: The data acquisition module is used to receive a streaming event popularity prediction dataset, which is constructed based on event-related data, including: post information, comment information, and popularity information. The model training module is used to extract the streaming event popularity prediction dataset to obtain the training set. The training set is then input into the pre-established streaming event popularity prediction model based on the multi-view neural Hawkes process. The model is trained based on the negative log maximum likelihood loss function to obtain the trained streaming event popularity prediction model based on the multi-view neural Hawkes process. The event popularity prediction module is used to input the streaming event popularity prediction dataset into the trained streaming event popularity prediction model based on multi-view neural Hawkes process, and output the event popularity prediction results.

8. 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 that can run on a processor. When the processor loads and executes the computer program, it employs the event heat prediction method based on the multi-perspective neural Hawkes process as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is loaded and executed by the processor, it employs the event heat prediction method based on the multi-perspective neural Hawkes process as described in any one of claims 1 to 6.

Images