Public Opinion Early Warning Method, Device, Readable Storage Medium and Electronic Device

Abstract

The present invention provides a public opinion early warning method, device, readable storage medium and electronic device thereof. The public opinion early warning method includes: obtaining public opinion texts, determining events included in the public opinion texts according to preset event tags, determining event subjects corresponding to the events in the public opinion texts, determining the sentiment polarity of the event subjects in the public opinion texts, and associating the public opinion texts with the event subjects and the event tags. By using this method, it is possible to obtain news with relatively high attention and relatively high risk from a large number of public opinion news, further accurately determine the subjects associated with the public opinion, avoid the occurrence of associating some unimportant public opinion news with subjects that have little relation to them, and at the same time, through this method, it is possible to accurately judge the subject corresponding to the sentiment expressed by the public opinion news and associate with it, so that users can obtain the early warning public opinion corresponding to the subject through the event tags and determine the risk information through the public opinion.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a public opinion early warning method, device, readable storage medium and electronic device. Background Art

[0002] In the public opinion section of the enterprise details page, users hope to learn about the recent events, news, etc. of an enterprise through this section. In order to enable users to timely understand the public opinion risks in the public opinion news corresponding to the enterprise, it is necessary to associate the public opinion with relevant risk information with the enterprise for users to intuitively understand the enterprise risks. However, in the Internet era, there are a large number of public opinion news, including positive and negative public opinions. In the prior art, the emotions of public opinion news are not screened, and only the subjects in the public opinion are simply identified by a model. This not only results in incorrect association between the enterprise and the public opinion, but also associates some unimportant public opinion news with subjects that have little to do with it. In addition, when there are multiple subjects in the public opinion news, the prior art cannot accurately determine which of the multiple subjects the emotion expressed by the public opinion news corresponds to. At the same time, in the prior art, only the emotion of the public opinion is simply identified by a model, and it cannot be accurately associated with its corresponding subject, resulting in low credibility of risk information and bringing many inconveniences to users. Therefore, how to accurately warn of public opinion with risk information has become an urgent technical problem to be solved. Summary of the Invention

[0003] The present invention provides a public opinion early warning method, device, readable storage medium and its electronic device to overcome or alleviate the above technical problems existing in the prior art.

[0004] According to one aspect of the present invention, a public opinion early warning method is provided, and the method includes:

[0005] Obtain public opinion text;

[0006] Determine the events included in the public opinion text according to the pre-set event tags;

[0007] Determine the event subject corresponding to the event in the public opinion text;

[0008] Determine the emotional polarity of the event subject in the public opinion text;

[0009] In response to determining that the public opinion text meets the early warning condition according to the emotional polarity corresponding to the event subject, associate the public opinion text with the event subject and the event tag so that users can obtain the risk information of the event subject through the event tag.

[0010] Optionally, before determining the events included in the public opinion text according to the preset event tags, the method further includes:

[0011] Using a collection of garbage corpus words, judge the quality of the public opinion text;

[0012] If it is determined that the public opinion text is garbage public opinion according to the quality of the public opinion text, perform filtering processing on the public opinion text;

[0013] If it is determined that the public opinion text is valid public opinion according to the quality of the public opinion text, perform the step of determining the events included in the public opinion text according to the preset event tags.

[0014] Optionally, the determining the event entity corresponding to the event in the public opinion text specifically includes:

[0015] Using an event entity dictionary tree, perform event entity mining on the public opinion text to obtain a first event entity in the public opinion text;

[0016] Through an event entity recognition model, perform event entity recognition on the public opinion text to obtain a second event entity in the public opinion text;

[0017] Based on the first event entity in the public opinion text and the second event entity in the public opinion text, determine the event entity in the public opinion text.

[0018] Optionally, the determining the event entity in the public opinion text specifically includes:

[0019] Determine the frequency of occurrence of the event entity in the public opinion text or the frequency of occurrence of the event entity from the first-person perspective in the public opinion text;

[0020] Based on the frequency of occurrence of the event entity in the public opinion text or the frequency of occurrence of the event entity from the first-person perspective in the public opinion text, determine the relevance between the event entity and the public opinion text;

[0021] Based on the relevance between the event entity and the public opinion text, determine the final event entity.

[0022] Optionally, the determining the event entity corresponding to the event in the public opinion text specifically includes:

[0023] Perform sentence splitting and word segmentation processing on the public opinion text to obtain multiple entities;

[0024] Determine the frequency of occurrence of each entity in the multiple entities from the first-person perspective in the public opinion text;

[0025] Determine the event entity corresponding to the event based on the frequency of each entity appearing from the first-person perspective in the public opinion text.

[0026] Optionally, the process of splitting the public opinion text into clauses and words to obtain multiple entities specifically includes:

[0027] In response to the public opinion text being a Chinese text, split the public opinion text into multiple sentences according to the final identifier;

[0028] In response to the public opinion text being an English text, split the public opinion text into multiple sentences according to the combined method of preset labels and capital letters;

[0029] For each sentence in the multiple sentences, perform word segmentation according to the sentence composition method to obtain the execution subject of each sentence, and use the execution subject as the multiple entities appearing in the public opinion text.

[0030] Optionally, the process of determining the events included in the public opinion text according to the preset event labels includes:

[0031] Perform corpus annotation on the text according to the preset event labels to obtain public opinion sample data;

[0032] Based on the public opinion sample data, perform machine learning on the model according to the multi-label classification model method to obtain a first recognition model;

[0033] Determine the events included in the public opinion text through the first recognition model.

[0034] Optionally, the process of determining the events included in the public opinion text according to the preset event labels includes:

[0035] Classify the preset event labels to obtain first-level event labels and second-level event labels;

[0036] Perform corpus annotation on the text according to the first-level event labels to obtain first public opinion sample data;

[0037] Based on the first public opinion sample data, perform machine learning on the model to obtain first semantic interaction data;

[0038] Determine the events included in the public opinion text corresponding to the first-level event labels through the first semantic interaction data;

[0039] Perform corpus annotation on the text according to the second-level event labels to obtain second public opinion sample data;

[0040] Based on the second public opinion sample data, perform machine learning on the model to obtain second semantic interaction data;

[0041] Determine the events corresponding to the second-level event tags included in the public opinion text based on the second semantic interaction data.

[0042] Optionally, determining the event entity corresponding to the event in the public opinion text includes:

[0043] Perform corpus annotation on the text according to the preset event tags and event entity tags to obtain public opinion sample data;

[0044] Perform machine learning on the model based on the public opinion sample data according to the binary classification model method to obtain a second recognition model;

[0045] Determine the event entity corresponding to the event in the public opinion text through the second recognition model.

[0046] Optionally, determining the sentiment polarity corresponding to the event entity in the public opinion text includes:

[0047] Based on a pre-configured sentiment question template, expand the event entity in the public opinion sample data to obtain the sentiment question text of the event entity in the public opinion sample data;

[0048] Based on the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, perform semantic interaction processing on the sentiment question text and the public opinion text to obtain the semantic interaction data between the sentiment question text and the public opinion text;

[0049] Through the event entity sentiment prediction model, based on the semantic interaction data between the sentiment question text and the public opinion text, determine the sentiment polarity of the event entity in the public opinion text.

[0050] Optionally, the performing semantic interaction processing on the sentiment question text and the public opinion text based on the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text to obtain the semantic interaction data between the sentiment question text and the public opinion text includes:

[0051] Determine the absolute value of the difference between the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, the product of the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, and the concatenated data of the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text;

[0052] Determine the semantic interaction data between the characters in the emotional problem text and the characters in the public opinion text based on the absolute value of the difference between the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text, the product of the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text, and the concatenated data of the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text;

[0053] Determine the semantic interaction data between the emotional problem text and the public opinion text based on the semantic interaction data between the characters in the emotional problem text and the characters in the public opinion text.

[0054] According to another aspect of the present invention, there is provided a public opinion early warning device, the device includes:

[0055] An acquisition module, configured to acquire public opinion text;

[0056] A first determination module, configured to determine the events included in the public opinion text according to a preset event label;

[0057] A second determination module, configured to determine the event entity corresponding to the event in the public opinion text;

[0058] A third determination module, configured to determine the emotional polarity of the event entity in the public opinion text;

[0059] The association module is configured to, in response to determining that the public opinion text meets the early warning condition according to the emotional polarity corresponding to the event entity, associate the public opinion text with the event entity and the event label, so that the user can obtain the risk information of the event entity through the event label.

[0060] According to still another aspect of the present invention, there is provided a computer-readable storage medium, on which a computer-executable program is stored, and the computer-executable program is run to implement any of the methods in the embodiments of the present invention.

[0061] According to still another aspect of the present invention, there is provided an electronic device, the electronic device includes a memory and a processor, the memory is used to store a computer-executable program, and the processor is used to run the computer-executable program to implement any of the methods in the embodiments of the present invention.

[0062] The present invention provides a public opinion early warning method. By using this method, news with high attention and great risk can be obtained from a large number of public opinion news. Through the screened public opinion news, the subject associated with the public opinion can be further accurately determined, avoiding the occurrence of associated events between some unimportant public opinion news and subjects that have little to do with them. At the same time, when there are multiple subjects in the public opinion news, this method can accurately judge the subject corresponding to the emotion expressed in the public opinion news and associate the public opinion warning with this subject, so that users can obtain the warning public opinion corresponding to the subject through the event label and determine the risk information through the public opinion. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a schematic flowchart of a public opinion early warning method according to an embodiment of the present invention;

[0064] Figure 2 It is a schematic flowchart of a public opinion early warning method according to an embodiment of the present invention;

[0065] Figure 3 It is a schematic flowchart of a public opinion early warning method according to an embodiment of the present invention;

[0066] Figure 4 It is a schematic structural diagram of a public opinion early warning device according to an embodiment of the present invention;

[0067] Figure 5 It is a schematic structural diagram of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0068] Next, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited by the exemplary embodiments described herein.

[0069] It should be noted that: unless otherwise specifically stated, the relative arrangements, numerical expressions, and numerical values of the components and steps described in these embodiments do not limit the scope of the present invention.

[0070] Those skilled in the art can understand that the terms "first", "second", etc. in the embodiments of the present invention are only used to distinguish different steps, devices, or modules, etc., and neither represent any specific technical meaning nor indicate an inevitable logical order between them.

[0071] It should also be understood that in the embodiments of the present invention, "a plurality of" may refer to two or more, and "at least one" may refer to one, two, or more.

[0072] It should also be understood that for any component, data, or structure mentioned in the embodiments of the present invention, in the absence of explicit definition or contrary indication in the context, it is generally understood as one or more.

[0073] In addition, the term "and / or" in the present invention is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects before and after are in an "or" relationship.

[0074] It should also be understood that the description of each embodiment of the present invention emphasizes the differences between the embodiments, and their similarities can be referred to each other. For the sake of brevity, they will not be elaborated one by one.

[0075] At the same time, it should be understood that for the convenience of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationship.

[0076] The following description of at least one exemplary embodiment is actually merely illustrative and in no way constitutes any limitation to the present invention and its application or use.

[0077] Techniques, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods, and devices should be regarded as part of the specification.

[0078] It should be noted that like reference numerals and letters indicate similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0079] The embodiments of the present invention can be applied to electronic devices such as terminal devices, computer systems, servers, etc., which can operate together with many other general or special computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments, and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, servers, etc. include, but are not limited to: personal computer systems, server computer systems, thin clients, fat clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, small computer systems, large computer systems, and distributed cloud computing technology environments including any of the above systems, and so on.

[0080] Electronic devices such as terminal devices, computer systems, and servers can be described in the general context of computer system-executable instructions (such as program modules) executed by a computer system. Generally, program modules can include routines, programs, target programs, components, logics, data structures, etc., which perform specific tasks or implement specific abstract data types. The computer system / server can be implemented in a distributed cloud computing environment where tasks are executed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules can be located on local or remote computing system storage media including storage devices.

[0081] Exemplary method

[0082] Figure 1 is a schematic flowchart of a public opinion early warning method provided by an exemplary embodiment of the present invention; as Figure 1 shown, the public opinion early warning method includes the following steps:

[0083] Step 101, obtain public opinion text;

[0084] In this embodiment, the public opinion text can be obtained from channels that produce public opinion such as various news media. For example, events or news issued by major news websites and news media accounts. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations in this regard.

[0085] In some optional embodiments, after obtaining the public opinion text, it further includes:

[0086] Use a collection of garbage corpus words to judge the quality of the public opinion text;

[0087] If it is determined that the public opinion text is garbage public opinion according to the quality of the public opinion text, perform filtering processing on the public opinion text;

[0088] If it is determined that the public opinion text is valid public opinion according to the quality of the public opinion text, perform the step of determining the events included in the public opinion text according to the preset event tags.

[0089] When it is determined that the public opinion text is garbage public opinion, performing filtering processing on the public opinion text can effectively ensure the quality of the public opinion text. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations in this regard.

[0090] In a specific example, the set of spam corpus words can be pre-configured. If a spam corpus word in the set of spam corpus words appears in the title or abstract of the public opinion text, it is determined that the quality of the public opinion text is low quality, and the public opinion text is determined to be spam public opinion, and then the public opinion text is filtered. For example, if the title of the public opinion text appears with spam words such as "giving out red envelopes", the public opinion text is not processed. If no spam corpus word in the set of spam corpus words appears in the title or abstract of the public opinion text, it is determined that the quality of the public opinion text is high quality, and the public opinion text is determined to be valid public opinion, and then step 102 is executed on the public opinion text. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations on this.

[0091] Step 102: Determine the events included in the public opinion text according to the pre-set event tags;

[0092] In this embodiment, before step 102, it also includes determining keywords related to the event subject according to historical news data, and constructing event tags according to the keywords; it should be noted that the keywords include words expressing positive meanings and words expressing negative meanings;

[0093] Specifically, the constructed event tags include: credit warning, breach of commitment, uncertainty in redemption and repayment, bond and debt default, China Bond implied rating, credit rating downgrade, negative rating outlook, included in rating observation, postponed rating, ordered to correct, information disclosure problem, financial warning, audit opinion, guarantee warning, capital risk, related to provision for bad debts management, senior management change, equity incentive, employee stock ownership plan, operation related, operating performance, strategic cooperation, merger and acquisition, equity pledge, capital increase and fundraising, investment and financing, bidding, asset restructuring, profit distribution, takeover and trusteeship, production capacity, related party transactions, product information, project signing, market related, increase and decrease in holdings, share repurchase, transfer of equity, new share issue, stock price decline, block trade, listing and delisting, backdoor listing and shell protection, suspension and resumption of trading, lifting of restricted share lock-up, order transaction, other related, credit business, general meeting of shareholders, rating information, honorary awards, policy impact, inspection and research, license, patent, public notice, meeting related, competition, blockchain, completion and commissioning, organization establishment, 5G, autonomous driving, force majeure;

[0094] Specifically, the event tags can also be classified into first-level event tags and second-level event tags; among them, there are 13 first-level event tags, specifically: credit warning, financial warning, management warning, operation warning, supervision warning, product warning, project warning, other warning, management related, operation related, market related, other related, force majeure, and the above-mentioned other event tags are second-level event tags.

[0095] In some alternative embodiments, determining the events included in the public opinion text according to the pre-set event tags includes:

[0096] The text is annotated according to the pre-set event labels to obtain the public opinion sample data;

[0097] Based on the public opinion sample data, the model is machine-learned according to the multi-label classification model method to obtain the first recognition model;

[0098] Events included in the public opinion text are determined through the first recognition model.

[0099] For example, in a specific application scenario, the sample text is: The fall of a giant! Company A suffered a quarterly loss of more than 7 billion for the first time since its listing, and U.S. stocks plunged 5% last night. According to the financial report data released yesterday, Company A's revenue in Q4 2021 was more than 100 billion yuan, a year-on-year increase of 64%; the operating profit loss was more than 7 billion yuan, while the operating profit in the same period last year was more than 7 billion yuan, a year-on-year plunge of 207.73%. The investigation found that the last time Company A suffered a single-quarter loss was in the second quarter of 2012, which is nearly ten years ago. Events that are annotated with event tags include operating performance, stock price declines, monopoly information, and thunderstorms.

[0100] In some optional embodiments, based on the public opinion sample data, machine learning is performed on the model according to a multi-label classification model method to obtain a first recognition model, specifically including:

[0101] Use the basic model to encode the labeled sample data to obtain a text representation vector;

[0102] The first recognition model is obtained by machine learning based on the loss function and the text representation vector.

[0103] Among them, the design of the loss function is aimed at the imbalance of multiple event samples. The use of the loss function can increase the difficulty of sample learning, thereby increasing the weight coefficient of difficult samples; specifically, in the process of obtaining the first recognition model through machine learning based on the loss function and the text representation vector, the weight coefficient of the event label and the corresponding word in the sample is calculated according to the following formula: Label = a1*trigger+a2*other_word+...+an*low_word.

[0104] Specifically, basic models such as BERT, ROBERT, and ERNIE are used to learn parameters on the model using labeled sample data. A multi-label classification model method is used to interactively learn the text and the above event labels to obtain the first recognition model, in which a single text is classified once for each event label using the sigmoid function, and a probability value greater than the preset value of 0.5 indicates the existence of the event.

[0105] In some alternative embodiments, determining the events included in the public opinion text according to pre-set event tags includes:

[0106] Classifying the pre-set event tags to obtain first-level event tags and second-level event tags;

[0107] Performing corpus annotation on the text according to the first-level event tags to obtain first public opinion sample data;

[0108] Performing machine learning on the model based on the first public opinion sample data to obtain first semantic interaction data;

[0109] Determining the events corresponding to the first-level event tags included in the public opinion text through the first semantic interaction data;

[0110] Performing corpus annotation on the text according to the second-level event tags to obtain second public opinion sample data;

[0111] Performing machine learning on the model based on the second public opinion sample data to obtain second semantic interaction data;

[0112] Determining the events corresponding to the second-level event tags included in the public opinion text through the second semantic interaction data.

[0113] Specifically, dividing the event tags into levels and then training different recognition models corresponding to different events through event tags at different levels can play a restrictive role and filter out the situations where the event relationships preset by the model are inaccurate.

[0114] Step 103: Determine the event subject corresponding to the event in the public opinion text;

[0115] In some alternative embodiments, as Figure 2 shown, determining the event subject corresponding to the event in the public opinion text specifically includes:

[0116] Step 201: Using an event subject dictionary tree, performing event subject mining on the public opinion text to obtain a first event subject in the public opinion text;

[0117] Step 202: Performing event subject recognition on the public opinion text through an event subject recognition model to obtain a second event subject in the public opinion text;

[0118] Step 203: Determining the event subject in the public opinion text based on the first event subject in the public opinion text and the second event subject in the public opinion text.

[0119] Thus, the first event entity in the public opinion text obtained through the event entity trie and the second event entity in the public opinion text obtained through the event entity recognition model can accurately determine the event entity in the public opinion text. When determining the event entity in the public opinion text, the intersection of the first event entity in the public opinion text and the second event entity in the public opinion text is determined as the event entity in the public opinion text. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations thereto.

[0120] In a specific example, the event entity trie can be understood as a trie that stores strings of event entities. The event entity recognition model can be understood as a neural network model for event entity recognition. The event entity recognition model can be any suitable neural network model that can implement feature extraction or target object detection, including but not limited to convolutional neural networks, reinforcement learning neural networks, generative networks in adversarial neural networks, etc. The specific structure settings in the neural network can be appropriately set by those skilled in the art according to actual needs, such as the number of convolutional layers, the size of the convolutional kernel, the number of channels, etc. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations thereto.

[0121] In a specific example, using the event entity trie for matching to obtain the first event entity in the public opinion text, that is, using string matching to mine event entities. However, this method lacks semantic information, but has a high recall rate. Using the event entity recognition method, semantic information can be introduced. By integrating the event entity trie and the event entity recognition model, the recall rate and accuracy can be improved, making the event entity more reliable and improving the accuracy of determining the subsequent sentiment polarity. For example, for the public opinion text "Apples have recently increased in price, 5 yuan per catty. / Apple has recently launched a new series of mobile phones, and the price has not increased but decreased instead.", the former refers to fruits and the latter refers to mobile phones. If only matching through the event entity trie, the event entity will introduce incorrect event entities. Eventually, regardless of whether the sentiment prediction is accurate, the result will be incorrect. If the event entity recognition model is added, ultimately the fruits will not be recognized, and the model only recognizes the event entity, such as Apple in Apple mobile phones representing Apple Inc. In this way, incorrect event entities can be reduced, and the sentiment polarity of the finally predicted event entity can be improved. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations thereto.

[0122] In some alternative embodiments, after determining the event entity in the public opinion text, it further includes:

[0123] Determining the frequency of occurrence of the event entity in the public opinion text or the frequency of the event entity appearing from the first-person perspective in the public opinion text;

[0124] Determine the relevance between the event entity and the public opinion text based on the frequency of the event entity's appearance in the public opinion text or the frequency of the event entity's appearance from the first-person perspective in the public opinion text;

[0125] Determine the final event entity based on the relevance between the event entity and the public opinion text.

[0126] Thereby, based on the frequency of the event entity's appearance in the public opinion text or the frequency of the event entity's appearance from the first-person perspective in the public opinion text, the relevance between the event entity and the public opinion text can be accurately determined. In addition, based on the relevance between the event entity and the public opinion text, the final event entity can be accurately determined. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations thereto.

[0127] In a specific example, the more frequently the event entity appears in the public opinion text, the stronger the relevance between the event entity and the public opinion text. The more frequently the event entity appears from the first-person perspective in the public opinion text, the stronger the relevance between the event entity and the public opinion text. Among them, the relevance can be characterized by relevance characterization data. When determining the final event entity, determine the event entity with the strongest relevance to the public opinion text as the final event entity. For example, the public opinion text is "Company B has a lawsuit with an enterprise regarding a certain game, and Company B wins". Company B appears from the first-person perspective and appears twice, so the public opinion text has a greater relevance to Company B. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations thereto.

[0128] In some alternative embodiments, as Figure 3 shown, determining the event entity corresponding to the event in the public opinion text specifically includes:

[0129] Step 301: Perform clause segmentation and word segmentation on the public opinion text to obtain multiple entities;

[0130] Step 302: Determine the frequency of each entity in the multiple entities appearing from the first-person perspective in the public opinion text;

[0131] Step 303: Based on the frequency of each entity appearing from the first-person perspective in the public opinion text, determine the event entity corresponding to the event.

[0132] Specifically, performing clause segmentation and word segmentation on the public opinion text to obtain multiple entities specifically includes:

[0133] In response to the public opinion text being a Chinese text, perform clause segmentation on the public opinion text according to the final identifier to obtain multiple sentences;

[0134] In response to the public opinion text being an English text, perform clause segmentation on the public opinion text according to the combined manner of a preset label and a capital letter to obtain multiple sentences;

[0135] Tokenize each statement in multiple statements according to the sentence composition method to obtain the execution subject of each statement, and use the execution subject as multiple entities that appear in the public opinion text.

[0136] In a specific example, if the public opinion text is in Chinese, the public opinion text is segmented into multiple statements according to the end symbol of the sentence, i.e., the full stop; if the public opinion text is in English, the public opinion text is segmented into multiple statements according to the end symbol of the English sentence, i.e., the position where the dot meets the beginning of the next capital letter.

[0137] In some alternative embodiments, determining the event entity corresponding to the event in the public opinion text includes:

[0138] Perform corpus annotation on the text according to the pre-set event tags and event entity tags to obtain public opinion sample data; perform machine learning on the model according to the binary classification model method based on the public opinion sample data to obtain a second recognition model; determine the event entity corresponding to the event in the public opinion text through the second recognition model.

[0139] For example, in a specific application scenario, the example text is: In May 2019, a certain department received a report on four driving schools in County A. After preliminary verification, the department officially launched an investigation in July 2019. The enterprises involved are Driving School A, Driving School B, Driving School C, and Driving School D, all of which carry out driver training business in County A. Specifically, the marked events include: monopoly information, investigation launched; the entities corresponding to the marked events include: the monopoly information entity is: Driving School A, Driving School B, Driving School C, Driving School D, and the entity for the investigation launched is: Driving School A, Driving School B, Driving School C, Driving School D;

[0140] Specifically, use a pre-trained language model, such as bert, robert, ernie, etc., to perform parameter learning on the multi-event multi-entity extraction model using the marked sample data. The present invention uses a method similar to reading comprehension to split multiple events, such as the above monopoly information and investigation launched, and splice them with the text respectively, i.e., [CLS]monopoly information[SEP]text, and then use the activation function sigmoid to complete the probability prediction of the start position and the probability prediction of the end position for the multi-entities of the event, i.e., binary classification, to determine whether a certain word in the text is the start position or the end position of the event entity. Finally, intercept the entity from the predicted result, and the obtained entity set is the multi-entity of the event. Then, perform the same operation on [CLS]investigation launched[SEP]text to complete the training of the multi-event multi-entity model.

[0141] In some alternative embodiments, due to the problem of uneven numbers and large differences in event entities, for samples with high prediction difficulty, the focal loss function is used to increase the weight coefficient. The design of the loss function is for the reason of unevenness of multiple event samples. Using the loss function can increase the difficulty of sample learning, thereby increasing the weight coefficient of samples with high difficulty. Specifically, the weight coefficient of the event label and the corresponding word in the sample is calculated according to the following formula: Label = a1*trigger + a2*other_word +... + an*low_word. Where a1, a2... an represent weights, trigger represents a type of keyword of the event, other word represents a type of non-event word, and low word represents a non-keyword.

[0142] Step 104, determine the sentiment polarity of the event entity in the public opinion text;

[0143] In some alternative embodiments, determining the sentiment polarity corresponding to the event entity in the public opinion text includes:

[0144] Based on a pre-configured sentiment question template, expand the event entity in the public opinion sample data to obtain the sentiment question text of the event entity in the public opinion sample data;

[0145] Based on the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, perform semantic interaction processing on the sentiment question text and the public opinion text to obtain the semantic interaction data of the sentiment question text and the public opinion text;

[0146] Through the event entity sentiment prediction model, based on the semantic interaction data of the sentiment question text and the public opinion text, determine the sentiment polarity of the event entity in the public opinion text.

[0147] Specifically, based on the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, performing semantic interaction processing on the sentiment question text and the public opinion text to obtain the semantic interaction data of the sentiment question text and the public opinion text includes:

[0148] Determine the absolute value of the difference between the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, the product of the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, and the concatenated data of the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text;

[0149] Determine the semantic interaction data between the characters in the emotional problem text and the characters in the public opinion text based on the absolute value of the difference between the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text, the product of the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text, and the concatenated data of the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text;

[0150] Determine the semantic interaction data between the emotional problem text and the public opinion text based on the semantic interaction data between the characters in the emotional problem text and the characters in the public opinion text.

[0151] In some alternative embodiments, based on a pre-configured emotional problem template, substitute the event entity in the public opinion text into the pre-configured emotional problem template to generate an emotional problem text for the event entity.

[0152] Thereby, an emotional problem text for the event entity can be effectively generated, further enabling more interaction information between the event entity and the public opinion text, and thus improving the accuracy of the emotional polarity of the event entity. It can be understood that the above description is only exemplary, and this embodiment makes no limitations in this regard.

[0153] In this embodiment, the pre-configured emotional problem template can be "Whether one of the positive, neutral, and negative emotional polarities exists for entities such as enterprises and institutions xxx in this text content". It can be understood that the above description is only exemplary, and this embodiment makes no limitations in this regard.

[0154] In a specific example, for instance, when the event entity is "Company B" and the emotional problem template is "Whether one of the positive, neutral, and negative emotional polarities exists for entities such as enterprises and institutions xxx in this text content", "Company B" can be substituted into "Whether one of the positive, neutral, and negative emotional polarities exists for entities such as enterprises and institutions xx in this text content". It can be understood that the above description is only exemplary, and this embodiment makes no limitations in this regard.

[0155] In this embodiment, the event subject sentiment prediction model can be understood as a neural network model for event subject sentiment prediction. The event subject sentiment prediction model can be any suitable neural network model that can implement feature extraction or target object detection, including but not limited to convolutional neural networks, reinforcement learning neural networks, generative networks in adversarial neural networks, etc. The specific structure settings in the neural network can be appropriately set by those skilled in the art according to actual needs, such as the number of convolutional layers, the size of the convolutional kernel, the number of channels, etc. In specific implementation, a reading comprehension model can be used to replace the event subject sentiment prediction model to predict the sentiment polarity of the event subject. The sentiment polarity of the event subject can include non-negative and negative. The label of the sentiment polarity of the event subject can be 0 and 1. Among them, 0 corresponds to negative and 1 corresponds to non-negative. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations thereto.

[0156] In a specific example, a large pre-trained language model, such as the bert model, roberta model, ernie model, etc., can be used to represent the characters in the text, and a deep neural network can be used to learn semantic and syntactic information to construct an opinion text and an event subject sentiment problem text, and then predict the sentiment polarity of the event subject. Finally, for the relationship question and answer after the interaction between the event subject sentiment problem text and the opinion text, that is, to predict the sentiment polarity of the event subject. This task is a text classification task with 3 labels, and the activation function of the last fully connected layer uses softmax.

[0157] For example, the opinion text is "A certain platform: Success and failure are both due to AA. In two years, AA made two strategic misjudgments: heavily betting on business C and plan D. Coincidentally, the two misjudgments had a convergence point on June 30: AA was officially listed, and the trading volume of a certain platform reached the last high point. If the subsequent trend after listing still experienced some games, then the defeat of a certain platform was immediate. On July 1, the trading volume of the entire platform dropped precipitously, and some categories even dropped by 60%", and the event subject sentiment problem text is "Among the three sentiment polarities of positive, neutral, and negative, one of the entities such as enterprises and institutions AA exists in this text content", and the obtained sentiment polarity and its label of the event subject are non-negative and 1 respectively. It can be understood that the above description is only exemplary, and this embodiment does not make any limitations thereto.

[0158] In some alternative embodiments, after determining the sentiment polarity of the event subject, the method further includes: using a regular expression for correcting the sentiment polarity to perform event subject matching on the public opinion text to obtain the matching event subject in the public opinion text; if the matching event subject in the public opinion text is the same as the event subject in the public opinion text, using the sentiment polarity represented by the regular expression to correct the sentiment polarity of the event subject. Thereby, the accuracy of the sentiment polarity of the event subject can be further improved. It can be understood that the above description is only exemplary, and this embodiment does not make any limitation thereto.

[0159] In a specific example, the event subject sentiment prediction model can solve most problems. For the small part of the situations that the model fails to handle and are likely to cause misunderstanding of the model and reduce the model performance, obvious features can be used to handle such situations without modifying the model. For example, when the public opinion text is "According to the data shown by Tianyancha, Tianyancha has a neutral sentiment in the text and is mostly neutral in most cases, but if the data shown by Tianyancha indicates that AA has recently been interviewed and the app has been taken off the shelves", some extremely negative information appears near Tianyancha, which is likely to have a negative impact on the event subject. Using a regular expression such as *entity* data shown / reported, etc., the sentiment polarity of such event subjects can be corrected to improve the accuracy. It can be understood that the above description is only exemplary, and this embodiment does not make any limitation thereto.

[0160] Step 105: In response to determining that the public opinion text meets the early warning condition according to the sentiment polarity corresponding to the event subject, associate the public opinion text with the event subject and the event label.

[0161] In this embodiment, step 105 further includes determining whether the public opinion text meets the early warning condition according to the sentiment polarity corresponding to the event subject; specifically, if there is an event in the public opinion text, there is an event subject corresponding to the event, and the sentiment of the event subject in the public opinion is negative, then it meets the early warning condition.

[0162] In a specific example, the public opinion text is as follows: A certain news website client reported on July 27, 2021 (by Reporter A of a certain evening newspaper) that recently, a certain department launched an operation and conducted an interview with a certain online car-hailing platform. After the interview, the department issued a "Notice" to the platform, and the online car-hailing platform was involved in 10 cases. Specifically, the events in the determined public opinion text include: Event A [0.073, 0.927], Event B [0.273, 0.727]; the event subjects corresponding to the determined events are: a certain online car-hailing platform; the sentiment of the determined event subject: a certain online car-hailing platform, negative, [0.97, 0.02, 0.01]; according to the sentiment polarity, it is determined that this public opinion text meets the early warning conditions: there is an event, the event has an event subject, and the sentiment of the event subject is negative, meeting the above conditions and meeting the early warning conditions.

[0163] By using this method, news with relatively high attention and relatively high risks can be obtained from a large number of public opinion news. Through the screened public opinion news, the subject associated with the public opinion can be further accurately determined, avoiding the occurrence of irrelevant public opinion news being associated with subjects that have little to do with them. At the same time, when there are multiple subjects in the public opinion news, this method can accurately judge the subject corresponding to the sentiment expressed by the public opinion news and associate the public opinion with the subject, so that users can obtain the early warning public opinion corresponding to the subject through the event label and determine the risk information through the public opinion.

[0164] Exemplary apparatus

[0165] Figure 4 It is a schematic structural diagram of an apparatus for public opinion early warning according to an embodiment of the present invention; as Figure 4 shown, the public opinion early warning apparatus 400 includes:

[0166] An acquisition module 401, configured to acquire public opinion text;

[0167] A first determination module 402, configured to determine the events included in the public opinion text according to a preset event label;

[0168] A second determination module 403, configured to determine the event subject corresponding to the event in the public opinion text;

[0169] A third determination module 404, configured to determine the sentiment polarity of the event subject in the public opinion text;

[0170] The association module 405 is configured to, in response to determining that the public opinion text meets the early warning conditions according to the sentiment polarity corresponding to the event subject, associate the public opinion text with the event subject and the event label, so that users can obtain the risk information of the event subject through the event label.

[0171] Optionally, in this embodiment, the device further includes: a judgment module, configured to use the garbage corpus word set to judge the quality of the public opinion text; further configured to determine that the public opinion text is garbage public opinion according to the quality of the public opinion text, and perform filtering processing on the public opinion text; further configured to determine that the public opinion text is effective public opinion according to the quality of the public opinion text, and trigger the first determination module.

[0172] Optionally, in this embodiment, the second determination module is specifically configured to use the event subject trie to mine the event subject of the public opinion text to obtain the first event subject in the public opinion text; use the event subject recognition model to identify the event subject of the public opinion text to obtain the second event subject in the public opinion text; determine the event subject in the public opinion text based on the first event subject in the public opinion text and the second event subject in the public opinion text.

[0173] Optionally, in this embodiment, the device further includes a fourth determination module, configured to determine the frequency of occurrence of the event subject in the public opinion text or the frequency of occurrence of the event subject in the first-person perspective in the public opinion text; determine the relevance between the event subject and the public opinion text based on the frequency of occurrence of the event subject in the public opinion text or the frequency of occurrence of the event subject in the first-person perspective in the public opinion text; determine the final event subject based on the relevance between the event subject and the public opinion text.

[0174] Optionally, in this embodiment, the second determination module specifically includes: a processing unit and a determination unit, where the processing unit is configured to perform clause and word segmentation processing on the public opinion text to obtain multiple subjects; the determination unit is configured to determine the frequency of occurrence of each subject in the multiple subjects in the first-person perspective in the public opinion text; determine the event subject corresponding to the event based on the frequency of occurrence of each subject in the first-person perspective in the public opinion text.

[0175] Optionally, in this embodiment, the processing unit is specifically configured to, in response to the public opinion text being a Chinese text, perform clause segmentation on the public opinion text according to the final identifier to obtain multiple sentences; in response to the public opinion text being an English text, perform clause segmentation on the public opinion text according to the combination of the preset label and capital letters to obtain multiple sentences; perform word segmentation on each sentence in the multiple sentences according to the sentence composition method to obtain the execution subject of each sentence, and use the execution subject as the multiple subjects appearing in the public opinion text.

[0176] Optionally, in this embodiment, the first determination module is specifically configured to perform corpus annotation on the text according to the preset event tags to obtain public opinion sample data; perform machine learning on the model according to the multi-label classification model method based on the public opinion sample data to obtain a first recognition model; determine the events included in the public opinion text through the first recognition model.

[0177] Optionally, in this embodiment, the first determination module is specifically configured to classify the preset event tags into first-level event tags and second-level event tags; perform corpus annotation on the text according to the first-level event tags to obtain first public opinion sample data; perform machine learning on the model based on the first public opinion sample data to obtain first semantic interaction data; determine the events corresponding to the first-level event tags included in the public opinion text through the first semantic interaction data; perform corpus annotation on the text according to the second-level event tags to obtain second public opinion sample data; perform machine learning on the model based on the second public opinion sample data to obtain second semantic interaction data; determine the events corresponding to the second-level event tags included in the public opinion text through the second semantic interaction data.

[0178] Optionally, in this embodiment, the second determination module is specifically configured to perform corpus annotation on the text according to the preset event tags and event subject tags to obtain public opinion sample data; perform machine learning on the model according to the binary classification model method based on the public opinion sample data to obtain a second recognition model; determine the event subject corresponding to the event in the public opinion text through the second recognition model.

[0179] Optionally, in this embodiment, the third determination module is specifically configured to expand the event subject in the public opinion sample data based on a pre-configured emotion question template to obtain the emotion question text of the event subject in the public opinion sample data; perform semantic interaction processing on the emotion question text and the public opinion text based on the semantic feature representation data of the characters in the emotion question text and the semantic feature representation data of the characters in the public opinion text to obtain the semantic interaction data between the emotion question text and the public opinion text; determine the emotion polarity of the event subject in the public opinion text based on the semantic interaction data between the emotion question text and the public opinion text through an event subject emotion prediction model.

[0180] Exemplary electronic device

[0181] Figure 5 This is the structure of an electronic device provided by an exemplary embodiment of the present invention. The electronic device can be any one or both of the first device and the second device, or a stand-alone device independent of them, and the stand-alone device can communicate with the first device and the second device to receive the input signals collected from them. Figure 5The block diagram of an electronic device according to an embodiment of the present invention is illustrated. As Figure 5 shown, the electronic device 500 includes one or more processors 501 and a memory 502.

[0182] The processor 501 may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0183] The memory 502 may include one or more computer programs. The computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage media, and the processor 501 may run the program instructions to implement the public opinion early warning method of various embodiments of the present invention described above and / or other desired functions. In one example, the electronic device may further include: an input device 503 and an output device 504, and these components are interconnected through a bus system and / or other forms of connection mechanisms (not shown).

[0184] In addition, the input device 503 may further include, for example, a keyboard, a mouse, and so on.

[0185] The output device 504 may output various information to the outside. The output device 504 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, and so on.

[0186] Of course, for simplicity, Figure 5 only some of the components related to the present invention in the electronic device are shown in, and components such as buses, input / output interfaces, and so on are omitted. In addition, according to specific application scenarios, the electronic device may further include any other appropriate components.

[0187] Exemplary computer program product and computer-readable storage medium

[0188] In addition to the above methods and devices, an embodiment of the present invention may also be a computer program product, which includes computer program instructions that, when run by a processor, cause the processor to execute the steps in the public opinion early warning method according to various embodiments of the present invention described in the "Exemplary Method" section above of this specification.

[0189] The computer program product can be written in any combination of one or more programming languages for executing the program code of the operations of the embodiments of the present invention. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, executed as an independent software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0190] In addition, an embodiment of the present invention can also be a computer-readable storage medium storing computer program instructions, which, when run by a processor, cause the processor to execute the steps in the public opinion warning method according to various embodiments of the present invention described in the above "Exemplary Method" section of this specification.

[0191] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, 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.

[0192] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, advantages, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, advantages, effects, etc. are essential for each embodiment of the present invention. In addition, the above-disclosed specific details are only for the purpose of illustration and easy understanding, rather than limitations, and the above details do not limit the present invention to necessarily adopt the above specific details for implementation.

[0193] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts between the embodiments, reference can be made to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment.

[0194] The block diagrams of the devices, apparatuses, equipment, and systems involved in the present invention are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open-ended terms, meaning "including but not limited to", and can be used interchangeably with each other. The words "or" and "and" used herein refer to the phrase "and / or" and can be used interchangeably with it, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to" and can be used interchangeably with it.

[0195] The methods and apparatuses of the present invention can be implemented in many ways. For example, the methods and apparatuses of the present invention can be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of the steps for the methods is only for illustration, and the steps of the methods of the present invention are not limited to the specific order described above, unless otherwise specifically stated. In addition, in some embodiments, the present invention can also be implemented as a program recorded in a recording medium, and these programs include machine-readable instructions for implementing the methods according to the present invention. Therefore, the present invention also covers the recording medium storing the programs for executing the methods according to the present invention.

[0196] It should also be noted that in the apparatuses, equipment, and methods of the present invention, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of the present invention. The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the present invention. Therefore, the present invention is not intended to be limited to the aspects shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

[0197] The above description has been given for purposes of illustration and description. In addition, this description is not intended to limit the embodiments of the present invention to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions, and sub-combinations thereof.

Claims

1. A public opinion early warning method, characterized in that, The method includes: Obtain public opinion texts; Determine the events included in the public opinion texts according to preset event tags; Determine the event subjects corresponding to the events in the public opinion texts; Determine the sentiment polarity of the event subjects in the public opinion texts; In response to determining that the public opinion text meets the early warning conditions according to the sentiment polarity corresponding to the event subject, associate the public opinion text with the event subject and the event tag, so that the user can obtain the risk information of the event subject through the event tag; Among them, the determining the event subject corresponding to the event in the public opinion text specifically includes: Use an event subject trie to mine the event subject from the public opinion text to obtain the first event subject in the public opinion text; Identify the event subject from the public opinion text through an event subject recognition model to obtain the second event subject in the public opinion text; Based on the first event subject in the public opinion text and the second event subject in the public opinion text, determine the event subject in the public opinion text; or Perform sentence splitting and word segmentation on the public opinion text to obtain multiple subjects; Determine the frequency of each subject in the multiple subjects appearing from the first-person perspective in the public opinion text; Based on the frequency of each subject appearing from the first-person perspective in the public opinion text, determine the event subject corresponding to the event; or Perform corpus annotation on the text according to the preset event tags and event subject tags to obtain public opinion sample data; Perform machine learning on the model based on the public opinion sample data according to the binary classification model method to obtain a second recognition model; Determine the event subject corresponding to the event in the public opinion text through the second recognition model.

2. The method according to claim 1, wherein The method further includes: Use a set of garbage corpus words to judge the quality of the public opinion text; If it is determined that the public opinion text is garbage public opinion according to the quality of the public opinion text, perform filtering processing on the public opinion text; If it is determined that the public opinion text is valid public opinion according to the quality of the public opinion text, perform the step of determining the events included in the public opinion text according to the preset event tags.

3. The method according to claim 1, wherein After determining the event subject in the public opinion text, it further includes: Determine the frequency of the event subject appearing in the public opinion text or the frequency of the event subject appearing from the first-person perspective in the public opinion text; Based on the frequency of the event subject appearing in the public opinion text or the frequency of the event subject appearing from the first-person perspective in the public opinion text, determine the relevance between the event subject and the public opinion text; Based on the relevance between the event subject and the public opinion text, determine the final event subject.

4. The method according to claim 1, wherein The performing sentence splitting and word segmentation on the public opinion text to obtain multiple subjects specifically includes: In response to the public opinion text being a Chinese text, perform sentence splitting on the public opinion text according to the final identifier to obtain multiple sentences; In response to the public opinion text being an English text, perform sentence splitting on the public opinion text according to the combination of a preset label and capital letters to obtain multiple sentences; Segment each of the multiple statements according to the sentence composition method to obtain the execution subject of each statement, and use the execution subject as the multiple subjects appearing in the public opinion text.

5. The method according to claim 1, characterized in that The determination of the events included in the public opinion text according to the preset event tags includes: Perform corpus annotation on the text according to the preset event tags to obtain public opinion sample data; Based on the public opinion sample data, perform machine learning on the model according to the multi-label classification model method to obtain a first recognition model; Determine the events included in the public opinion text through the first recognition model.

6. The method according to claim 1, characterized in that, The determination of the events included in the public opinion text according to the preset event tags includes: Classify the preset event tags to obtain first-level event tags and second-level event tags; Perform corpus annotation on the text according to the first-level event tags to obtain first public opinion sample data; Perform machine learning on the model based on the first public opinion sample data to obtain first semantic interaction data; Determine the events corresponding to the first-level event tags included in the public opinion text through the first semantic interaction data; Perform corpus annotation on the text according to the second-level event tags to obtain second public opinion sample data; Perform machine learning on the model based on the second public opinion sample data to obtain second semantic interaction data; Determine the events corresponding to the second-level event tags included in the public opinion text through the second semantic interaction data.

7. The method according to claim 1, wherein The determination of the sentiment polarity corresponding to the event subject in the public opinion text includes: Based on a pre-configured sentiment question template, expand the event subject in the public opinion sample data to obtain the sentiment question text of the event subject in the public opinion sample data; Based on the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, perform semantic interaction processing on the sentiment question text and the public opinion text to obtain the semantic interaction data between the sentiment question text and the public opinion text; Through an event subject sentiment prediction model, based on the semantic interaction data between the sentiment question text and the public opinion text, determine the sentiment polarity of the event subject in the public opinion text.

8. The method according to claim 7, wherein The performing of semantic interaction processing on the sentiment question text and the public opinion text based on the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text to obtain the semantic interaction data between the sentiment question text and the public opinion text includes: Determine the absolute value of the difference between the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, the product of the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text, and the concatenated data of the semantic feature representation data of the characters in the sentiment question text and the semantic feature representation data of the characters in the public opinion text; Determine the semantic interaction data of the characters in the emotional problem text and the characters in the public opinion text based on the absolute value of the difference between the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text, the product of the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text, and the concatenated data of the semantic feature representation data of the characters in the emotional problem text and the semantic feature representation data of the characters in the public opinion text; Determine the semantic interaction data of the emotional problem text and the public opinion text based on the semantic interaction data of the characters in the emotional problem text and the characters in the public opinion text.

9. An opinion warning device, characterized in that, The device includes: An acquisition module, configured to acquire a public opinion text; A first determination module, configured to determine an event included in the public opinion text according to a preset event label; A second determination module, configured to determine an event subject corresponding to the event in the public opinion text; A third determination module, configured to determine the emotional polarity of the event subject in the public opinion text; An association module, configured to, in response to determining that the public opinion text meets the early warning condition according to the emotional polarity corresponding to the event subject, associate the public opinion text with the event subject and the event label, so that a user can obtain risk information of the event subject through the event label; Wherein, the second determination module is further configured to: Use an event subject trie to mine event subjects from the public opinion text to obtain a first event subject in the public opinion text; Identify event subjects from the public opinion text through an event subject recognition model to obtain a second event subject in the public opinion text; Determine the event subject in the public opinion text based on the first event subject in the public opinion text and the second event subject in the public opinion text; or Perform sentence splitting and word segmentation processing on the public opinion text to obtain multiple subjects; Determine the frequency of occurrence of each subject in the multiple subjects from the first-person perspective in the public opinion text; Determine the event subject corresponding to the event based on the frequency of occurrence of each subject from the first-person perspective in the public opinion text; or Perform corpus annotation on the text according to the preset event label and event subject label to obtain public opinion sample data; Perform machine learning on the model based on the public opinion sample data according to the binary classification model method to obtain a second recognition model; Determine the event subject corresponding to the event in the public opinion text through the second recognition model.

10. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer-executable program, and the computer-executable program is run to implement the method according to any one of claims 1-8.

11. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory is used to store a computer-executable program, and the processor is used to run the computer-executable program to implement the method according to any one of claims 1-8.

Images