Public opinion early warning method, device, equipment and storage medium

Abstract

The present application relates to the technical field of commercial information systems, and in particular to a method, device, equipment and storage medium for warning of public opinion. The method includes the following steps: obtaining hot topics to be warned; obtaining comment data of hot topics to be warned; obtaining the first layer of related hot topics based on the comment data; obtaining the second layer of related hot topics according to the first layer of related hot topics, thereby iteratively forming a set of related hot topics; obtaining the peak heat of each related hot topic in the extended set of related hot topics, and the time interval between each related hot topic and the first appearance time of the hot topic to be warned, and constructing a first prediction time window matrix; inputting the first prediction time window matrix into the heat warning model, thereby predicting the future heat peak of the hot topic to be warned; when the future heat peak is greater than the preset second heat threshold, sending a warning message to the staff. The present application can warn of public opinion more accurately.

Description

Technical Field

[0001] The present application relates to the technical field of commercial information systems, and in particular to a public opinion early warning method, device, equipment and storage medium. Background Art

[0002] In the information age, the speed of spread and scope of influence of online public opinion have increased significantly, and effective monitoring and early warning have become key tasks for enterprises and other organizations. Traditional public opinion early warning methods are mainly based on direct data analysis of current topics, such as the number of comments, forwarding times, and likes. When these indicators reach the preset threshold, the system will trigger an early warning.

[0003] However, there are still some problems with existing technologies, which makes it impossible to provide more accurate public opinion warnings. Summary of the invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present application provides a public opinion warning method, device, equipment and storage medium, which can more accurately warn of public opinion.

[0005] In a first aspect, the present application provides a public opinion early warning method, the public opinion early warning method comprising the following steps:

[0006] Obtain topics whose heat index is greater than a preset first heat threshold as hot topics to be warned;

[0007] Obtaining comment data of the hot topic to be warned;

[0008] Based on the comment data, obtaining a first layer of associated hot topics associated with the hot topic to be warned;

[0009] According to the first layer of related hot topics, the related hot topics of the first layer of related hot topics are obtained to obtain the second layer of related hot topics, and the related hot topics of each layer of related hot topics are iteratively obtained until a preset number of layers is reached to form an extended set of related hot topics;

[0010] Obtaining the peak heat of each associated hot topic in the expanded associated hot topic set, and the time interval between each associated hot topic and the first appearance time of the hot topic to be warned;

[0011] Constructing a first prediction time window matrix, wherein the first prediction time window matrix is ​​composed of each associated hot topic in the expanded associated hot topic set and its corresponding peak heat and the time interval;

[0012] Inputting the first prediction time window matrix into a preset heat warning model to predict the future heat peak value of the hot topic to be warned;

[0013] When the future heat peak value is greater than a preset second heat threshold, an early warning message is sent to the staff.

[0014] Optionally, based on the comment data, obtaining the first-layer associated hot topics associated with the hot topics to be warned includes the following steps:

[0015] Obtain each hot topic whose occurrence time from the hot topic to be warned is within a first preset time interval and whose popularity is greater than a preset third popularity threshold as a presumed associated hot topic;

[0016] For each combination of the assumed associated hot topic and the hot topic to be warned, the following steps are performed to obtain a second prediction time window matrix after the hot topic to be warned is combined with each assumed associated hot topic:

[0017] Obtaining the associated hot topics of the assumed associated hot topics to obtain the second layer of assumed associated hot topics, and iteratively obtaining the associated hot topics of each layer of assumed associated hot topics until a preset number of layers is reached, thereby forming an extended set of assumed associated hot topics;

[0018] Obtaining the peak heat of each assumed associated hot topic in the expanded set of assumed associated hot topics, and the time interval between the first appearance time of the assumed associated hot topic and the hot topic to be warned;

[0019] Constructing a second prediction time window matrix, wherein the second prediction time window matrix is ​​composed of each assumed associated hot topic in the extended set of assumed associated hot topics and its corresponding peak heat and the time interval;

[0020] Input each second prediction time window matrix into a preset heat warning model to predict the future heat peak value of the hot topic to be warned after the hot topic to be warned is combined with each assumed associated hot topic;

[0021] According to the preset correlation analysis formula, the correlation threshold of each assumed associated hot topic is calculated by using the predicted future heat peak value after combining the hot topic to be warned with each assumed associated hot topic; wherein, when the assumed associated hot topic is determined to be the first-layer associated hot topic of the hot topic to be warned, the higher its future heat peak value is, the lower the corresponding correlation threshold is;

[0022] When obtaining the first-layer associated hot topics associated with the hot topics to be warned based on the comment data, the association threshold is adopted so that hot topics with lower association thresholds are more easily determined as the first-layer associated hot topics associated with the hot topics to be warned.

[0023] Optionally, the preset correlation analysis formula is:

[0024]

[0025] in,

[0026]

[0027] in, is the association threshold of the ith assumed associated hot topic, is the preset correlation threshold reference value, is the future peak value of heat of the ith assumed associated hot topic, μ is the average of the future peak values ​​of heat of all assumed associated hot topics, σ is the standard deviation of the future peak values ​​of heat of all assumed associated hot topics, and β and γ are preset empirical parameters.

[0028] Optionally, the preset heat warning model is trained by the following steps:

[0029] Collecting historical public opinion data, the historical public opinion data includes multiple hot topics, their associated hot topic sets, the peak heat corresponding to each associated hot topic in the associated hot topic set, and the time interval between each associated hot topic and the hot topic;

[0030] Constructing a training sample, wherein the training sample includes an expanded set of associated hot topics, a peak heat and time interval corresponding to each associated hot topic in the set, and a known future heat peak of the hot topic;

[0031] The convolutional neural network is used as the network framework of the heat warning model. Each associated hot topic in the associated hot topic set, the corresponding peak heat and time interval are used as the input of the heat warning model network framework. The known future heat peaks of hot topics are used as training labels to train the heat warning model.

[0032] In a second aspect, the present application provides a public opinion early warning device, comprising:

[0033] A data acquisition module, the data acquisition module is used to acquire topics whose heat index is greater than a preset first heat threshold value as hot topics to be warned;

[0034] Obtaining comment data of the hot topic to be warned;

[0035] Based on the comment data, obtaining a first layer of associated hot topics associated with the hot topic to be warned;

[0036] According to the first layer of related hot topics, the related hot topics of the first layer of related hot topics are obtained to obtain the second layer of related hot topics, and the related hot topics of each layer of related hot topics are iteratively obtained until a preset number of layers is reached to form an extended set of related hot topics;

[0037] Obtaining the time interval between each associated hot topic in the expanded associated hot topic set and the first appearance time of the hot topic to be warned;

[0038] A data analysis module, the data analysis module is used to construct a prediction time window matrix, the prediction time window matrix is ​​composed of each associated hot topic in the expanded associated hot topic set and the corresponding time interval;

[0039] Input the prediction time window matrix into a preset heat warning model to predict the future heat peak value of the hot topic to be warned;

[0040] The early warning module is used to send an early warning message to the staff when the future heat peak value is greater than a preset second heat threshold.

[0041] In a third aspect, the present application provides a public opinion warning device, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement a public opinion warning method as described in any one of the first aspects.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which, when executed, implements the public opinion warning method as described in any one of the first aspects.

[0043] Compared with the prior art, the technical solution provided by this application has the following advantages:

[0044] One of its beneficial effects is that the public opinion warning method provided by the present application can evaluate the popularity of the cultural core or values ​​represented by these topics at the current time point by capturing the popularity changes and time relationships of multiple current topics to be warned and their related topics, and can thus be used to evaluate the impact of the popularity of related topics on the future popularity of the current topic, and therefore can more accurately predict the future popularity trend of the current topic.

[0045] The second beneficial effect is that the public opinion warning method provided by this application assumes possible related hot topics in advance, predicts their impact on the future popularity of the warning topic, and dynamically adjusts the sensitivity of related hot topics according to the risk level. By dynamically adjusting the association threshold, early identification of high-risk related hot topics is achieved, enhancing the sensitivity and flexibility of the warning system. The warning system can issue warning information more promptly, giving staff more time to respond, and improving the initiative and accuracy of public opinion warnings.

[0046] The third beneficial effect is that the public opinion warning method provided by this application realizes the adaptive adjustment of the warning system to the changes in the public opinion environment by introducing the overall heat level and difference into the correlation analysis formula, ensuring stable and accurate warning performance in different time periods. This method effectively solves the problem of warning failure caused by fluctuations in the overall heat level and enhances the accuracy of public opinion warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flowchart of the public opinion warning method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] The technical solution in this application will be described below in conjunction with the accompanying drawings.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application can also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only part of the embodiments of the present application, not all of the embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0050] One of the application scenarios of the public opinion warning method proposed below in the embodiment of the present application can be used for warning of various hot spots in Weibo, where the topics referred to in the present application can be understood as hot spots in Weibo.

[0051] In a first aspect, an embodiment of the present application provides a public opinion early warning method, the public opinion early warning method comprising the following steps:

[0052] S101: Acquire topics whose popularity index is greater than a preset first popularity threshold as hot topics to be warned.

[0053] Specifically, the heat index here can be calculated by weighted addition based on the number of comments, the number of reposts and the number of likes, etc. In the embodiment of the present application, the heat index is directly extracted from the corresponding web page.

[0054] The first heat threshold is a manually selected value.

[0055] S102: Obtaining comment data of the hot topic to be warned;

[0056] Specifically, the comment data here can use a web crawler or an API interface to obtain all comment data under the hot topics to be warned, and the comment data at least includes comment content and user information.

[0057] S103: Based on the comment data, obtaining the first layer of associated hot topics associated with the hot topic to be warned, including the following steps:

[0058] S201: Obtain each hot topic whose occurrence time from the hot topic to be warned is within a first preset time interval and whose popularity is greater than a preset third popularity threshold as a presumed associated hot topic.

[0059] Specifically, the first preset time interval and the third heat threshold here are both manually set values.

[0060] S202: For each combination of the assumed associated hot topic and the hot topic to be warned, respectively, the following steps S301-S303 are executed to obtain a second prediction time window matrix of the combination of the hot topic to be warned and each assumed associated hot topic.

[0061] S301: Obtain associated hot topics of the assumed associated hot topics to obtain a second layer of assumed associated hot topics, and iteratively obtain associated hot topics of each layer of assumed associated hot topics until a preset number of layers is reached to form an extended set of assumed associated hot topics.

[0062] S302: Acquire the peak heat of each assumed associated hot topic in the expanded set of assumed associated hot topics, and the time interval between the first appearance time of the assumed associated hot topic and the hot topic to be warned.

[0063] S303: Construct a second prediction time window matrix, wherein the second prediction time window matrix is ​​composed of each assumed associated hot topic in the extended set of assumed associated hot topics and its corresponding peak heat and the time interval.

[0064] S203: Input each second prediction time window matrix into a preset heat warning model to predict the future heat peak value of the hot topic to be warned after the hot topic to be warned is combined with each assumed associated hot topic.

[0065] S204: According to a preset correlation analysis formula, the correlation threshold of each assumed associated hot topic is calculated using the predicted future heat peak value after combining the hot topic to be warned with each assumed associated hot topic; wherein, when the assumed associated hot topic is determined to be the first-level associated hot topic of the hot topic to be warned, the higher its future heat peak value is, the lower the corresponding correlation threshold is.

[0066] S205: When obtaining the first-layer associated hot topics associated with the hot topics to be warned based on the comment data, the associated threshold is adopted so that hot topics with lower associated thresholds are more easily determined as the first-layer associated hot topics associated with the hot topics to be warned.

[0067] Specifically, the preset correlation analysis formula is:

[0068]

[0069] in,

[0070]

[0071] in, is the association threshold of the ith assumed associated hot topic, It is the preset correlation threshold reference value, which is an artificially set value.

[0072] is the future peak value of heat of the ith assumed associated hot topic, μ is the average of the future peak values ​​of heat of all assumed associated hot topics, σ is the standard deviation of the future peak values ​​of heat of all assumed associated hot topics, and β and γ are artificially preset empirical parameters.

[0073] Specifically, the step of using the association threshold to determine the associated hot topics includes:

[0074] Get the identifiers of other topics mentioned in the comment data of the hot topics to be warned, as well as the corresponding number of mentions.

[0075] The embodiment of the present application goes through each comment content in the comment data one by one, extracts keywords in the comment content through existing natural language processing methods, such as the Jieba word segmentation algorithm, and then searches the Internet based on the keywords to determine the topic corresponding to the keyword, or matches it with a pre-defined keyword-topic matching list, thereby determining the remaining topics mentioned in the comment data.

[0076] The co-occurrence rate of the mentioned topics is calculated, which is the number of mentions of the topic divided by the total number of comments.

[0077] The co-occurrence rate of the mentioned topic is compared with the associated threshold corresponding to the mentioned topic. When the co-occurrence rate is greater than the associated threshold, the mentioned topic is determined to be a first-layer associated hot topic of the hot topic to be warned.

[0078] S104: According to the first layer of associated hot topics, the associated hot topics of the first layer are obtained to obtain the second layer of associated hot topics, and the associated hot topics of each layer of associated hot topics are iteratively obtained until a preset number of layers is reached to form an extended set of associated hot topics.

[0079] Specifically, the preset number of layers mentioned in the above steps S301 and S104 is an artificially set value, and is generally set to 4-6 layers in the embodiment of the present application.

[0080] S105: Acquire the peak popularity of each associated hot topic in the expanded associated hot topic set, and the time interval between each associated hot topic and the first appearance time of the hot topic to be warned.

[0081] S106: Constructing a first prediction time window matrix, wherein the first prediction time window matrix is ​​composed of each associated hot topic in the extended associated hot topic set and its corresponding peak heat and the time interval.

[0082] S107: Inputting the first prediction time window matrix into a preset heat warning model to predict the future heat peak value of the hot topic to be warned.

[0083] Specifically, the preset heat warning model in the above steps S203 and S107 is trained by the following steps:

[0084] Collecting historical public opinion data, the historical public opinion data includes multiple hot topics, their associated hot topic sets, the peak heat corresponding to each associated hot topic in the associated hot topic set, and the time interval between each associated hot topic and the hot topic;

[0085] Constructing a training sample, wherein the training sample includes an expanded set of associated hot topics, a peak heat and time interval corresponding to each associated hot topic in the set, and a known future heat peak of the hot topic;

[0086] The convolutional neural network is used as the network framework of the heat warning model. Each associated hot topic in the associated hot topic set, the corresponding peak heat and time interval are used as the input of the heat warning model network framework. The known future heat peaks of hot topics are used as training labels to train the heat warning model.

[0087] S108: When the future heat peak value is greater than a preset second heat threshold, a warning message is sent to the staff.

[0088] Specifically, the second heat threshold here is a manually set value.

[0089] In summary, the public opinion early warning method provided by the embodiment of the present application has one of the beneficial effects as follows:

[0090] It should be noted that the popularity of a topic not only depends on its own discussion, but is also affected by the cultural core or values ​​associated with it. The popularity of these cultural cores or values ​​at the current time point is reflected through a series of related topics, and their influence is related to the time interval between these related topics and the current topic, as well as the popularity of these related topics themselves.

[0091] The public opinion warning method provided in the present application first obtains the hot topics to be warned whose heat index is greater than the preset first heat threshold, and obtains their comment data. Based on these comment data, the first layer of related topics associated with the topic is extracted, and then each layer of related topics is iteratively obtained until the preset number of layers is reached to form an extended set of related topics. This set represents the propagation path and influence range of the cultural core or values ​​related to the topic to be warned along time.

[0092] For each related topic in the set, obtain its peak heat and the time interval between the first appearance of the topic to be warned and construct a prediction time window matrix. Input the matrix into the preset heat warning model to predict the future heat peak of the topic to be warned. When the predicted future heat peak exceeds the preset second heat threshold, send a warning message to the staff.

[0093] The beneficial effect is that the public opinion warning method provided in the embodiment of the present application can evaluate the popularity of the cultural core or values ​​represented by these topics at the current time point by capturing the popularity changes and time relationships of multiple current topics to be warned and their related topics, and can thus be used to evaluate the impact of the popularity of related topics on the future popularity of the current topic, and thus can more accurately predict the future popularity trend of the current topic.

[0094] The second beneficial effect is described as follows:

[0095] It should be noted that if we simply extract the related hot topics mentioned in the comment data of the current topic, and then analyze the future popularity of the current topic based on the change trend of these related hot topics, this method has a shortcoming: it needs to rely on a large amount of comment data to identify related topics. When the related topics are identified, public opinion may have begun to ferment, resulting in a short warning time window and the inability to provide warning information in a timely manner.

[0096] In order to solve this problem, the embodiment of the present application further proposes an improved method by assuming in advance that some hot topics at previous moments may be associated with the current topic to be warned. Specifically, within the first preset time interval before the occurrence time of the topic to be warned, the hot topics whose heat is greater than the preset third heat threshold are obtained as assumed related hot topics. Then, for each assumed related hot topic, it is combined with the topic to be warned, and its associated hot topics are iteratively obtained until the preset number of layers is reached, forming an extended set of assumed related hot topics. Next, the peak heat of these related topics and the time interval between the first appearance time of the topic to be warned are obtained to construct a second prediction time window matrix.

[0097] Input each second prediction time window matrix into the preset heat warning model to predict the future heat peak value of the topic to be warned when it is associated with the assumed associated hot topic. According to the predicted future heat threat level, the association threshold of each assumed associated hot topic is calculated using the association analysis formula. Among them, the higher the future heat peak value, the greater the potential public opinion risk, and the lower the corresponding association threshold. In this way, when obtaining associated hot topics based on comment data, the calculated association threshold is applied, so that for high-risk assumed associated hot topics, the warning system only needs to detect a small number of comment mentions to determine it as an associated hot topic.

[0098] The beneficial effect is that the embodiment of the present application assumes possible related hot topics in advance, predicts their impact on the future popularity of the warning topic, and dynamically adjusts the sensitivity of the related hot topics according to the risk level. By dynamically adjusting the association threshold, early identification of high-risk related hot topics is achieved, and the sensitivity and flexibility of the warning system are enhanced. The warning system can issue warning information more promptly, buy more response time for staff, and improve the initiative and effectiveness of public opinion warning.

[0099] The third beneficial effect is described as follows:

[0100] It should be noted that the overall level of public opinion heat may change significantly during different time periods, such as weekdays and holidays. During holidays, netizens are usually more active and discuss more topics, leading to an increase in the overall heat level. On the contrary, during weekdays, netizens may be less active and the overall heat level may drop.

[0101] If we only consider the correlation between the future peak value of the topic to be warned and the warning sensitivity, then when the overall popularity level rises, the warning system may issue warnings for a large number of normal high-popularity topics, increasing the false alarm rate. When the overall popularity level drops, the warning system may not be able to capture important public opinion changes in time, resulting in missed reports.

[0102] In order to solve the above problems, the embodiment of the present application introduces the overall heat level μ and the difference D through the correlation analysis formula. i The overall heat level μ reflects the average heat of the future heat peaks of all assumed related hot topics in the current period, and is an indicator to measure the overall public opinion activity. i Indicates the degree of deviation of the future peak value of the i-th assumed related hot topic from the overall average level. A positive difference indicates that the topic is hotter than the average level, and a negative difference indicates that it is hotter than the average level.

[0103] Through the formula and Two, the early warning system can dynamically adjust the relevance threshold based on the current overall popularity level and the relative popularity of each topic :

[0104] When the overall heat level increases (μ increases):

[0105] If the popularity of a topic Also increases, and the ratio with μ remains stable, then There will be no drastic changes and the sensitivity of the early warning system will remain stable.

[0106] If a topic's popularity increases more than the overall level ( >μ and >0), then The risk of hot topics will be reduced, and the early warning system will be more sensitive to the topic and capture abnormal high-heat topics in time.

[0107] When the overall heat level decreases (μ decreases):

[0108] If a topic is still popular, >μ, then It is still relatively low, and the early warning system is able to pay attention to this hot topic.

[0109] If the popularity of the topic is comparable to the overall level, The ratio with μ remains stable, then Maintain stability to avoid false positives due to a drop in overall heat.

[0110] The embodiment of the present application introduces the overall heat level and difference into the correlation analysis formula, so as to realize the adaptive adjustment of the early warning system to the changes in the public opinion environment, and ensure that the early warning performance can be maintained stably and accurately in different time periods. This method effectively solves the problem of early warning failure caused by fluctuations in the overall heat level, and enhances the accuracy of public opinion early warning.

[0111] In a second aspect, the present application provides a public opinion early warning device, comprising:

[0112] A data acquisition module, the data acquisition module is used to acquire topics whose heat index is greater than a preset first heat threshold value as hot topics to be warned;

[0113] Obtaining comment data of the hot topic to be warned;

[0114] Based on the comment data, obtaining a first layer of associated hot topics associated with the hot topic to be warned;

[0115] According to the first layer of related hot topics, the related hot topics of the first layer of related hot topics are obtained to obtain the second layer of related hot topics, and the related hot topics of each layer of related hot topics are iteratively obtained until a preset number of layers is reached to form an extended set of related hot topics;

[0116] Obtaining the time interval between each associated hot topic in the expanded associated hot topic set and the first appearance time of the hot topic to be warned;

[0117] A data analysis module, the data analysis module is used to construct a prediction time window matrix, the prediction time window matrix is ​​composed of each associated hot topic in the expanded associated hot topic set and the corresponding time interval;

[0118] Input the prediction time window matrix into a preset heat warning model to predict the future heat peak value of the hot topic to be warned;

[0119] The early warning module is used to send an early warning message to the staff when the future heat peak value is greater than a preset second heat threshold.

[0120] In a third aspect, the present application provides a public opinion warning device, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement a public opinion warning method as described in any one of the first aspects.

[0121] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which, when executed, implements the public opinion warning method as described in any one of the first aspects.

[0122] It should be noted that, in this article, relational 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 any such actual relationship or order between these entities or operations. In addition, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. Moreover, in the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can represent A or B; "and / or" in this article is only a kind of association relationship describing the associated objects, indicating that there can be three relationships, for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. Furthermore, in the description of the embodiments of the present application, “plurality” refers to two or more than two.

[0123] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A public opinion early warning method, characterized in that: The public opinion early warning method comprises the following steps: Obtain topics whose heat index is greater than a preset first heat threshold as hot topics to be warned; Obtaining comment data of the hot topic to be warned; Based on the comment data, obtaining a first layer of associated hot topics associated with the hot topic to be warned; According to the first layer of related hot topics, the related hot topics of the first layer of related hot topics are obtained to obtain the second layer of related hot topics, and the related hot topics of each layer of related hot topics are iteratively obtained until a preset number of layers is reached to form an extended set of related hot topics; Obtaining the peak heat of each associated hot topic in the expanded associated hot topic set, and the time interval between each associated hot topic and the first appearance time of the hot topic to be warned; Constructing a first prediction time window matrix, wherein the first prediction time window matrix is ​​composed of each associated hot topic in the expanded associated hot topic set and its corresponding peak heat and the time interval; Inputting the first prediction time window matrix into a preset heat warning model to predict the future heat peak value of the hot topic to be warned; When the future heat peak value is greater than a preset second heat threshold, a warning message is sent to the staff; Based on the comment data, obtaining the first-layer associated hot topics associated with the hot topics to be warned includes the following steps: Obtain each hot topic whose occurrence time from the hot topic to be warned is within a first preset time interval and whose popularity is greater than a preset third popularity threshold as a presumed associated hot topic; For each combination of the assumed associated hot topic and the hot topic to be warned, the following steps are performed to obtain a second prediction time window matrix after the hot topic to be warned is combined with each assumed associated hot topic: Obtaining the associated hot topics of the assumed associated hot topics to obtain the second layer of assumed associated hot topics, and iteratively obtaining the associated hot topics of each layer of assumed associated hot topics until a preset number of layers is reached, thereby forming an extended set of assumed associated hot topics; Obtaining the peak heat of each assumed associated hot topic in the expanded set of assumed associated hot topics, and the time interval between the first appearance time of the assumed associated hot topic and the hot topic to be warned; Constructing a second prediction time window matrix, wherein the second prediction time window matrix is ​​composed of each assumed associated hot topic in the extended set of assumed associated hot topics and its corresponding peak heat and the time interval; Input each second prediction time window matrix into a preset heat warning model to predict the future heat peak value of the hot topic to be warned after the hot topic to be warned is combined with each assumed associated hot topic; According to the preset correlation analysis formula, the correlation threshold of each assumed associated hot topic is calculated by using the predicted future heat peak value after combining the hot topic to be warned with each assumed associated hot topic; wherein, when the assumed associated hot topic is determined to be the first-layer associated hot topic of the hot topic to be warned, the higher its future heat peak value is, the lower the corresponding correlation threshold is; When obtaining the first-layer associated hot topics associated with the hot topics to be warned based on the comment data, the association threshold is adopted so that the hot topics with lower association thresholds are more likely to be determined as the first-layer associated hot topics associated with the hot topics to be warned; The preset correlation analysis formula is: in, in, is the association threshold of the ith assumed associated hot topic, is the preset correlation threshold reference value, is the future peak value of heat of the ith assumed associated hot topic, μ is the average of the future peak values ​​of heat of all assumed associated hot topics, σ is the standard deviation of the future peak values ​​of heat of all assumed associated hot topics, and β and γ are preset empirical parameters.

2. The public opinion early warning method according to claim 1, characterized in that: The preset heat warning model is trained by the following steps: Collecting historical public opinion data, the historical public opinion data includes multiple hot topics, their associated hot topic sets, the peak heat corresponding to each associated hot topic in the associated hot topic set, and the time interval between each associated hot topic and the hot topic; Constructing a training sample, wherein the training sample includes an expanded set of associated hot topics, a peak heat and time interval corresponding to each associated hot topic in the set, and a known future heat peak of the hot topic; The convolutional neural network is used as the network framework of the heat warning model. Each associated hot topic in the associated hot topic set, the corresponding peak heat and time interval are used as the input of the heat warning model network framework. The known future heat peaks of hot topics are used as training labels to train the heat warning model.

3. Public opinion early warning equipment, characterized in that: It includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the public opinion warning method as described in any one of claims 1-2.

4. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is executed, the public opinion early warning method as described in any one of claims 1-2 is implemented.

Images