Network public opinion monitoring method and equipment

By generating word sequences and entity category sequences, using symbol representation learning model and Transformer model to extract target trigger words and arguments, identifying public opinion event categories, solving the problem of keyword matching in the existing technology that consumes a lot of human resources, and achieving efficient recognition Monitoring of public opinion events.

CN114282542BActive Publication Date: 2024-11-15AGRICULTURAL BANK OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111631185.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-28
Publication Date
2024-11-15
Estimated Expiration
2041-12-28

AI Technical Summary

Technical Problem

The existing banking public opinion monitoring tools rely on keyword matching, require a large number of manual keyword formulation and extracting information from massive unstructured text consumes a lot of human resources, making it difficult to efficiently identify public opinion events.

Method used

By generating word sequences and entity category sequences, the symbolic representation learning model and the Transformer model are used to extract target trigger words and arguments, and combined with multi-layer perceptrons to classify, identify public opinion event categories, and reduce labor costs.

Benefits of technology

Effectively identifying public opinion-related texts and extracting semi-structured event information, reducing the labor cost of identifying public opinion events and improving monitoring efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114282542B_ABST
    Figure CN114282542B_ABST
Patent Text Reader

Abstract

The embodiment of the present application provides a network public opinion monitoring method and device, including: generating multiple word sequences and multiple entity category sequences according to the collected text data; extracting target trigger words in the text data according to each of the word sequences and each of the entity category sequences, and determining the event category of each target trigger word; extracting arguments in the text data according to each of the word sequences, and determining the entity category of each of the arguments; determining whether there are public opinion events to be processed in the text data according to the event category of each of the target trigger words and the entity category of each of the arguments. The network public opinion monitoring method provided by the embodiment of the present application can effectively reduce the manpower cost of network public opinion monitoring.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of network public opinion monitoring, and in particular to a network public opinion monitoring method and device. Background Art

[0002] With the rapid development of the financial industry and the enhancement of customers' awareness of rights protection, customers have higher and higher requirements for the services provided by banks. Once customers are dissatisfied with the bank's services, they may vent their personal emotions on some mainstream social platforms or news websites, which will be widely disseminated. Such situations will have a certain adverse impact on the bank's reputation.

[0003] The public opinion and reputation monitoring tools currently used by the banking industry mostly use keyword matching to extract information that may be related to specific banks from massive texts. The advantages of this method are simple rules, easy configuration, and high human intervention, but it also has obvious defects: a large number of manually formulated keywords need to be input and need to be supplemented regularly; the matching obtained from massive text data is still a large amount of unstructured text, which requires a lot of human resources for further reading and screening to determine what kind of public opinion events are contained in the extracted text. Summary of the invention

[0004] The embodiments of the present application provide a method and device for monitoring online public opinion, which can effectively reduce the manpower cost of monitoring online public opinion.

[0005] In a first aspect, an embodiment of the present application provides a method for monitoring online public opinion, the method comprising:

[0006] Generate multiple word sequences and multiple entity category sequences according to the collected text data;

[0007] According to each of the word sequences and each of the entity category sequences, target trigger words in the text data are extracted, and event categories of each target trigger word are determined; wherein the event categories include positive categories and negative categories;

[0008] Extracting arguments from the text data according to each of the word sequences, and determining the entity category of each of the arguments;

[0009] According to the event category of each of the target trigger words and the entity category of each of the arguments, it is determined whether there is a public opinion event to be processed in the text data.

[0010] In a feasible implementation manner, the text data is unstructured text, and the generating of multiple word sequences and multiple entity category sequences according to the collected text data includes:

[0011] Determine each sentence in the text data as a word sequence;

[0012] The named entity recognition tool NER is used to mark the nouns in each sentence in the text data to form a plurality of entity category sequences.

[0013] In a feasible implementation, extracting target trigger words in the text data according to each of the word sequences and each of the entity category sequences, and determining the event category of each target trigger word includes:

[0014] Input each of the word sequences and each of the entity category sequences into a symbolic representation learning model to obtain a first vector after encoding each of the word sequences and a second vector after encoding each of the entity category sequences; wherein the symbolic representation learning model includes a word embedding vector layer and a Transformer-based encoder layer;

[0015] Obtaining an interaction relationship vector between the target trigger word and the entity category according to the first vector after encoding each of the word sequences and the second vector after encoding each of the entity category sequences, and a preset trigger word entity interaction relationship learning model;

[0016] Obtaining a target trigger word representation vector according to the interaction relationship vector;

[0017] The target trigger word representation vector is classified based on a multi-layer perceptron (MLP) to obtain the event category of each target trigger word.

[0018] In a feasible implementation manner, extracting arguments from the text data according to each of the word sequences and determining the entity category of each of the arguments includes:

[0019] Using the Transformer-based bidirectional encoder BERT model to learn each of the word sequences to obtain an argument representation vector;

[0020] The argument representation vectors are classified based on MLP to obtain the entity category of each argument.

[0021] In a second aspect, an embodiment of the present application provides a network public opinion monitoring device, the device comprising:

[0022] A processing module, used to generate multiple word sequences and multiple entity category sequences according to the collected text data;

[0023] A trigger word extraction module, used to extract target trigger words in the text data according to each of the word sequences and each of the entity category sequences, and determine the event category of each target trigger word; wherein the event category includes a positive category and a negative category;

[0024] An argument extraction module, used to extract arguments from the text data according to each of the word sequences, and determine the entity category of each of the arguments;

[0025] The monitoring module is used to determine whether there is a public opinion event to be processed in the text data according to the event category of each target trigger word and the entity category of each argument.

[0026] In a feasible implementation manner, the processing module is specifically used for:

[0027] Determine each sentence in the text data as a word sequence;

[0028] The named entity recognition tool NER is used to mark the nouns in each sentence in the text data to form a plurality of entity category sequences.

[0029] In a feasible implementation manner, the trigger word extraction module is specifically used to:

[0030] Input each of the word sequences and each of the entity category sequences into a symbolic representation learning model to obtain a first vector after encoding each of the word sequences and a second vector after encoding each of the entity category sequences; wherein the symbolic representation learning model includes a word embedding vector layer and a Transformer-based encoder layer;

[0031] Obtaining an interaction relationship vector between the target trigger word and the entity category according to the first vector after encoding each of the word sequences and the second vector after encoding each of the entity category sequences, and a preset trigger word entity interaction relationship learning model;

[0032] Obtaining a target trigger word representation vector according to the interaction relationship vector;

[0033] The target trigger word representation vector is classified based on a multi-layer perceptron (MLP) to obtain the event category of each target trigger word.

[0034] In a feasible implementation manner, the argument extraction module is specifically used for:

[0035] Using the Transformer-based bidirectional encoder BERT model to learn each of the word sequences to obtain an argument representation vector;

[0036] The argument representation vectors are classified based on MLP to obtain the entity category of each argument.

[0037] In a third aspect, an embodiment of the present application provides an electronic device, including: at least one processor and a memory;

[0038] The memory stores computer-executable instructions;

[0039] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the network public opinion monitoring method provided in the first aspect.

[0040] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer execution instructions are stored. When a processor executes the computer execution instructions, the network public opinion monitoring method provided in the first aspect is implemented.

[0041] The network public opinion monitoring method and device provided in the embodiment of the present application can obtain deeper semantic information by extracting target trigger words from text data and determining the event category of each target trigger word, as well as extracting arguments from text data and determining the entity category of each argument, and can identify parts of text that are difficult to obtain by keyword matching. Furthermore, in the process of determining whether there are public opinion events to be processed in the text data based on the event category of each target trigger word and the entity category of each argument, the manpower cost of network public opinion monitoring can be effectively reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments of the present application or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0043] Figure 1 A flowchart of a method for monitoring online public opinion provided in an embodiment of the present application;

[0044] Figure 2 A flowchart of another network public opinion monitoring method provided in an embodiment of the present application;

[0045] Figure 3 A schematic diagram of a program module of a network public opinion monitoring device provided in an embodiment of the present application;

[0046] Figure 4 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. In addition, although the disclosure in the present application is introduced according to one or more exemplary examples, it should be understood that each aspect of these disclosures can also constitute a complete implementation method separately.

[0048] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their common and usual meanings.

[0049] The terms "first", "second", etc. in the specification and claims of this application and the above drawings are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise noted. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, for example, they can be implemented in an order other than those given in the diagrams or descriptions of the embodiments of this application.

[0050] In addition, the terms "include" and "have" and any variations thereof are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to those components expressly listed but may include other components not expressly listed or inherent to such products or devices.

[0051] The term "module" as used in this application refers to any known or later developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0052] With the rapid development of the financial industry and the enhancement of customers' awareness of rights protection, customers have higher and higher requirements for the services provided by banks. Once customers are dissatisfied with the bank's services, they may vent their personal emotions on some mainstream social platforms or news websites, which will be widely disseminated. Such situations will have a certain adverse impact on the bank's reputation.

[0053] The public opinion and reputation monitoring tools currently used by the banking industry mostly use keyword matching to extract information that may be related to specific banks from massive texts. The advantages of this method are simple rules, easy configuration, and high human intervention, but it also has obvious defects: a large number of manually formulated keywords need to be input and need to be supplemented regularly; the matching obtained from massive text data is still a large amount of unstructured text, which requires a lot of human resources for further reading and screening to determine what kind of public opinion events are contained in the extracted text.

[0054] In view of the above technical problems, a method for monitoring online public opinion is provided in an embodiment of the present application. By acquiring semantic information, public opinion-related texts can be accurately identified, and semi-structured event information can be extracted from unstructured texts, thereby effectively reducing the manpower cost of identifying public opinion events.

[0055] In some embodiments, the present application is based on deep learning and pre-training models. By integrating deep learning ideas with pre-training models, it uses the two modules of "event detection" and "argument extraction" to extract semi-structured event information from unstructured text.

[0056] The following is an explanation of the relevant terms involved in the embodiments of the present application:

[0057] Event: An event is an objectively observed change in a state that occurs within a certain time and place, involves one or more participants. In layman's terms, the core element of an event is the occurrence of a state change, which in turn involves attributes such as time and place, as well as participants such as people and objects.

[0058] Event extraction: Event extraction is the process of extracting specific events from unstructured text and semi-structuring them. The specific content includes elements such as the type of event, participants of the event, and attributes. Event extraction research includes two subtasks: trigger word extraction and argument extraction.

[0059] Trigger words: words that best represent the occurrence of an event, usually verbs or nouns.

[0060] Arguments: The participants of an event and some event attributes such as time and place are collectively called event arguments.

[0061] Trigger word extraction: Determine whether there is an event in the unstructured text, what the sign of the event (i.e., trigger word) is, extract the sign and mark the category of the event, which is also called event detection.

[0062] Argument extraction: After clarifying the occurrence and category of an event, extract the attributes of the event and the participants of the event.

[0063] Named Entity Recognition (NER): Also known as proper name recognition, it is a basic task in natural language processing. Named entities generally refer to entities with specific meanings or strong referentiality in texts, usually including names of people, places, names of organizations, dates, times, proper nouns, etc.

[0064] Word Embedding: Word embedding refers to the process of vectorizing a word.

[0065] Transformer Encoder: A deep learning model for extracting sequence features.

[0066] BERT: The full name is Bidirectional Encoder Representation from Transformers, which is a bidirectional deep learning pre-trained language model.

[0067] Reference Figure 1 , Figure 1 A flowchart of a method for monitoring online public opinion provided in an embodiment of the present application is provided. In some embodiments, the method for monitoring online public opinion includes:

[0068] S101. Generate multiple word sequences and multiple entity category sequences according to collected text data.

[0069] In some embodiments, each sentence in the collected text data may be determined as a word sequence; and the nouns in each sentence in the text data may be marked using NER to form a plurality of entity category sequences.

[0070] In some embodiments, the above-mentioned collected text data is unstructured text, which serves as input text for the network public opinion monitoring method in this application.

[0071] Among them, NER tools are important basic tools in application fields such as information extraction, question-answering systems, syntactic analysis, and machine translation, and they play an important role in the process of natural language processing technology becoming practical. Generally speaking, the task of named entity recognition is to identify three major categories (entity category, time category, and number category) and seven minor categories (name, organization name, place name, time, date, currency, and percentage) of named entities in the text to be processed. For example, in the sentence "Xiao Ming goes to school at 8 o'clock in the morning.", performing named entity recognition on it should be able to extract the following information: name: Xiao Ming, time: 8 o'clock in the morning, location: school.

[0072] S102: extract target trigger words from the text data according to each word sequence and each entity category sequence, and determine the event category of each target trigger word.

[0073] In some embodiments, each word sequence and each entity category sequence can be input into a symbolic representation learning model to obtain a first vector after encoding each word sequence and a second vector after encoding each entity category sequence; wherein the symbolic representation learning model includes a word embedding vector layer and a Transformer-based encoder layer.

[0074] Furthermore, the interaction relationship vector between the target trigger word and the entity category is obtained based on the first vector after encoding each word sequence and the second vector after encoding each entity category sequence, as well as the preset trigger word entity interaction relationship learning model. Based on the interaction relationship vector, the target trigger word representation vector is obtained; the target trigger word representation vector is classified based on the multi-layer perceptron MLP to obtain the event category of each target trigger word.

[0075] The above event categories include positive categories and negative categories.

[0076] In some embodiments, the above-mentioned positive categories and negative categories may include multiple subcategories. For example, the positive categories may include praise, compliments, meetings, promotions and other categories, and the negative categories may include quarrels, conflicts, confrontations, attacks, and verbal abuse and other categories, which are not limited in the embodiments of the present application.

[0077] S103. Extract arguments from the text data based on each word sequence, and determine the entity category of each argument.

[0078] In some embodiments, the BERT model is used to learn each of the word sequences to obtain an argument representation vector, and the argument representation vector is classified based on a multi-layer perceptron MLP to obtain an entity category of each argument.

[0079] S104. Determine whether there is a public opinion event to be processed in the text data based on the event category of each target trigger word and the entity category of each argument.

[0080] In some embodiments, for example, if there is a negative category in the extracted target trigger words, it can be determined that there is a public opinion event to be processed in the text data.

[0081] The network public opinion monitoring method provided in the embodiment of the present application can obtain deeper semantic information by extracting target trigger words from text data and determining the event category of each target trigger word, as well as extracting arguments from text data and determining the entity category of each argument, and can identify some texts that are difficult to obtain by keyword matching. Furthermore, in the process of determining whether there are public opinion events to be processed in the text data based on the event category of each target trigger word and the entity category of each argument, the manpower cost of network public opinion monitoring can be effectively reduced.

[0082] Reference Figure 2 , Figure 2 A flowchart of another network public opinion monitoring method provided in an embodiment of the present application. In some embodiments, the network public opinion monitoring method includes:

[0083] For the input unstructured text, each sentence is regarded as a word sequence, and the nouns in the text are marked in advance with the help of NER tools to form an entity category sequence.

[0084] In the embodiment of the present application, the above word sequence and entity category sequence together constitute the input information of the trigger word extraction model. For each word sequence of length n, it is formally represented as w = {w1, w2, ..., w n}; For each entity category sequence of length n, it is formally represented as e = {e1, e2, …, e n}. Among them, w i represents the i-th word in the above word sequence, e i Represents the i-th word in the above entity category sequence.

[0085] The above word sequence and the above entity category sequence are firstly subjected to context-integrated sequence encoding by a symbolic representation learning model, wherein the above symbolic representation learning model is composed of a superposition of a word embedding sublayer and a TransformerEncoder sublayer.

[0086] After the above word sequence w and entity category sequence e are learned through the above symbolic representation model, the encoded vector representation x of the above word sequence w and entity category sequence e will be obtained respectively. w ={x w1 ,x w2 ,…,x wn}, and x e ={x e1 ,x e2 ,…,x en},

[0087] In getting x w and x e On the basis of , in order to capture the interactive relationship between the trigger word candidate and the entity category in the text sequence, a trigger word-entity interactive relationship learning model with attention mechanism as the core is adopted. The specific calculation method of this model is shown in the following formula:

[0088] x ′ e ={x ei |1≤i≤n and ei ≠O} (1)

[0090]

[0091] r = α·relu(x ′ e U3+b4) (3)

[0093] in, k is the sequence x e The number of non-O fields in α is the calculated attention weight parameter, x wt Represents the text vector corresponding to the trigger word candidate.

[0094] The above formula (3) is the combination sequence x e ′ and the attention weight parameter α, the formula for calculating the interaction relationship between the trigger word and the entity category. In the process of calculating the interaction relationship, in order to further increase the capacity of the model and enhance the model's expressiveness, it is possible to e Add a layer of fully connected neural network above ′. In formula (3), and Represents the output vector of the interaction relationship encoding module.

[0095] After obtaining the trigger word-entity interaction relationship vector r, it is combined with the text vector x of the trigger word candidate. wt , entity category vector x corresponding to the trigger word candidate et The three are concatenated to form the final trigger word candidate vector representation. Afterwards, the multi-layer perceptron is used to classify the vector representation to obtain the event category of the trigger word candidate. The specific calculation method is shown in the formula:

[0096] H c1 =selu([x wt ,x et ,r]W c1 +b c1 )

[0097] H c2 =selu(H c1 W c2 +b c2 )

[0098] O T =softmax(H c2 W T +b T )

[0099] in, are weight matrices, d ci (i∈{1,2}) is the dimension of the hidden state, d T Represents the number of event types.

[0100] In some embodiments, compared with the trigger word extraction module, the argument extraction module also uses a word sequence as the input of the model, and also defines the module training target as a classification task, but the difference between the two is that a sample only contains one trigger word candidate, but often contains multiple argument candidates. Therefore, the argument extraction module needs to classify multiple arguments in each sample. In view of the above situation, the present application uses the BERT model in the argument extraction module to learn word sequences.

[0101] Z,s=BERT(w) (4)

[0103]

[0104] In formula (4), Z = [z1, z2, …, z n ], Represents the new vector representation of all fields in the original input sample w after being encoded by the BERT model, and the other output result in the formula is It can be regarded as the vector representation of the whole sentence, which contains the information of the whole sentence and is generally used for sentence-level classification tasks. Represents the vector representation corresponding to the candidate word in Z, n A Represents the number of argument candidates. Represents x e The entity category representation corresponding to the argument candidate word in . Represents copying s to n A The final argument candidate word vector representation is denoted by F.

[0105] We also use a multilayer perceptron as an argument classifier to classify n in F. A The specific calculation method is shown in the following formula:

[0106] H c3 =selu(FW c3 +b c3 )

[0107] H c4 =selu(H c3 W c4 +b c3 )

[0108] H c5 =selu(H c4 W c5+b c4 )

[0109] O A =softmax(H c5 W A +b A )

[0110] in are weight matrices, d ci (i∈{3,4,5}) is the dimension of the hidden state, d A Represents the number of argument types.

[0111] The network public opinion monitoring method provided in the embodiment of the present application can obtain deeper semantic information by integrating deep learning and pre-training models, can identify partial text that is difficult to obtain through keyword matching, and can extract semi-structured event information, thereby reducing the manpower cost of further identifying public opinion events.

[0112] Based on the contents described in the above embodiments, a network public opinion monitoring device is also provided in the embodiments of the present application. Figure 3 , Figure 3 This is a schematic diagram of a program module of a network public opinion monitoring device provided in an embodiment of the present application, and the network public opinion monitoring device includes:

[0113] The processing module 301 is used to generate multiple word sequences and multiple entity category sequences according to the collected text data.

[0114] The trigger word extraction module 302 is used to extract the target trigger words in the text data according to each of the word sequences and each of the entity category sequences, and determine the event category of each target trigger word; wherein the event category includes a positive category and a negative category.

[0115] The argument extraction module 303 is used to extract arguments from the text data according to each of the word sequences, and determine the entity category of each of the arguments.

[0116] The monitoring module 304 is used to determine whether there is a public opinion event to be processed in the text data according to the event category of each of the target trigger words and the entity category of each of the arguments.

[0117] In some embodiments, the processing module 301 is specifically used to:

[0118] Each sentence in the text data is determined as a word sequence; and the nouns in each sentence in the text data are marked using NER to form a plurality of entity category sequences.

[0119] In some embodiments, the trigger word extraction module 302 is specifically used to:

[0120] Each word sequence and each entity category sequence is input into a symbolic representation learning model to obtain a first vector after encoding each word sequence and a second vector after encoding each entity category sequence; wherein the symbolic representation learning model includes a word embedding vector layer and a Transformer-based encoder layer.

[0121] According to the first vector after encoding each word sequence and the second vector after encoding each entity category sequence, as well as the preset trigger word entity interaction relationship learning model, the interaction relationship vector between the target trigger word and the entity category is obtained; according to the interaction relationship vector, the target trigger word representation vector is obtained; based on the MLP, the target trigger word representation vector is classified to obtain the event category of each target trigger word.

[0122] In some embodiments, the argument extraction module 303 is specifically used to:

[0123] The BERT model is used to learn each of the word sequences to obtain an argument representation vector; the argument representation vector is classified based on MLP to obtain the entity category of each argument.

[0124] It should be noted that the specific implementation contents of the processing module 301, the trigger word extraction module 302, the argument extraction module 303 and the monitoring module 304 in the embodiment of the present application can be referred to in Figure 1 to Figure 2 The relevant contents in the illustrated embodiment will not be described in detail here.

[0125] Furthermore, based on the contents described in the above embodiments, an electronic device is also provided in an embodiment of the present application, which includes at least one processor and a memory; wherein the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory to implement the various steps of the network public opinion monitoring method described in the above embodiments, and this embodiment will not be repeated here.

[0126] In order to better understand the embodiments of the present application, refer to Figure 4 , Figure 4 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application.

[0127] like Figure 4 As shown, the electronic device 40 of this embodiment includes: a processor 401 and a memory 402; wherein:

[0128] Memory 402, used to store computer-executable instructions;

[0129] Processor 401 is used to execute computer-executable instructions stored in the memory to implement the various steps of the network public opinion monitoring method described in the above embodiment, which will not be repeated here in this embodiment.

[0130] Optionally, the memory 402 may be independent or integrated with the processor 401 .

[0131] When the memory 402 is independently provided, the device further includes a bus 403 for connecting the memory 402 and the processor 401 .

[0132] Furthermore, based on the contents described in the above embodiments, a computer-readable storage medium is also provided in the embodiments of the present application, in which computer execution instructions are stored. When the processor executes the computer execution instructions, the various steps of the network public opinion monitoring method described in the above embodiments are implemented, and this embodiment will not be repeated here.

[0133] Furthermore, based on the contents described in the above embodiments, a computer-readable storage medium is also provided in the embodiments of the present application, in which computer execution instructions are stored. When the processor executes the computer execution instructions, the various steps of the network public opinion monitoring method described in the above embodiments are implemented, and this embodiment will not be repeated here.

[0134] Furthermore, based on the contents described in the above embodiments, a computer program product is also provided in the embodiments of the present application, including a computer program. When the computer program is executed by a processor, the various steps of the network public opinion monitoring method described in the above embodiments are implemented, and this embodiment will not be repeated here.

[0135] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0136] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0137] In addition, each functional module in each embodiment of the present application can be integrated into a processing unit, each module can exist physically separately, or two or more modules can be integrated into one unit. The above module integration unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0138] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to perform some steps of the method described in each embodiment of the present application.

[0139] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the application may be directly implemented as being executed by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.

[0140] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0141] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0142] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0143] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0144] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A network public opinion monitoring method, characterized in that: The method comprises: Generate multiple word sequences and multiple entity category sequences according to the collected text data; According to each of the word sequences and each of the entity category sequences, target trigger words in the text data are extracted, and event categories of each target trigger word are determined; wherein the event categories include positive categories and negative categories; Extracting arguments from the text data according to each of the word sequences, and determining the entity category of each of the arguments; Determining whether there is a public opinion event to be processed in the text data according to the event category of each of the target trigger words and the entity category of each of the arguments; The step of extracting target trigger words from the text data according to each of the word sequences and each of the entity category sequences, and determining the event category of each target trigger word, includes: Input each of the word sequences and each of the entity category sequences into a symbolic representation learning model to obtain a first vector after encoding each of the word sequences and a second vector after encoding each of the entity category sequences; wherein the symbolic representation learning model includes a word embedding vector layer and a Transformer-based encoder layer; Obtaining an interaction relationship vector between the target trigger word and the entity category according to the first vector after encoding each of the word sequences and the second vector after encoding each of the entity category sequences, and a preset trigger word entity interaction relationship learning model; Obtaining a target trigger word representation vector according to the interaction relationship vector; The target trigger word representation vector is classified based on a multi-layer perceptron (MLP) to obtain the event category of each target trigger word.

2. The method according to claim 1, characterized in that The text data is unstructured text, and the generating of multiple word sequences and multiple entity category sequences according to the collected text data includes: Determine each sentence in the text data as a word sequence; The named entity recognition tool NER is used to mark the nouns in each sentence in the text data to form a plurality of entity category sequences.

3. The method according to claim 1, characterized in that: The step of extracting arguments from the text data according to each of the word sequences and determining the entity category of each of the arguments comprises: Using the Transformer-based bidirectional encoder BERT model to learn each of the word sequences to obtain an argument representation vector; The argument representation vectors are classified based on MLP to obtain the entity category of each argument.

4. A network public opinion monitoring device, characterized in that: The device comprises: A processing module, used to generate multiple word sequences and multiple entity category sequences according to the collected text data; A trigger word extraction module, used to extract target trigger words in the text data according to each of the word sequences and each of the entity category sequences, and determine the event category of each target trigger word; wherein the event category includes a positive category and a negative category; An argument extraction module, used to extract arguments from the text data according to each of the word sequences, and determine the entity category of each of the arguments; A monitoring module, used to determine whether there is a public opinion event to be processed in the text data according to the event category of each of the target trigger words and the entity category of each of the arguments; The trigger word extraction module is specifically used for: Input each of the word sequences and each of the entity category sequences into a symbolic representation learning model to obtain a first vector after encoding each of the word sequences and a second vector after encoding each of the entity category sequences; wherein the symbolic representation learning model includes a word embedding vector layer and a Transformer-based encoder layer; Obtaining an interaction relationship vector between the target trigger word and the entity category according to the first vector after encoding each of the word sequences and the second vector after encoding each of the entity category sequences, and a preset trigger word entity interaction relationship learning model; Obtaining a target trigger word representation vector according to the interaction relationship vector; The target trigger word representation vector is classified based on a multi-layer perceptron (MLP) to obtain the event category of each target trigger word.

5. The device according to claim 4, characterized in that The processing module is specifically used for: Determine each sentence in the text data as a word sequence; The named entity recognition tool NER is used to mark the nouns in each sentence in the text data to form a plurality of entity category sequences.

6. The device according to claim 4, characterized in that The argument extraction module is specifically used for: Using the Transformer-based bidirectional encoder BERT model to learn each of the word sequences to obtain an argument representation vector; The argument representation vectors are classified based on MLP to obtain the entity category of each argument.

7. An electronic device, characterized in that: include: at least one processor and memory; The memory stores computer-executable instructions; The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the network public opinion monitoring method as described in any one of claims 1 to 3.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions. When the processor executes the computer-executable instructions, the network public opinion monitoring method according to any one of claims 1 to 3 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the network public opinion monitoring method according to any one of claims 1 to 3 is implemented.