[go: up one dir, main page]

CN117494068B - Network public opinion analysis method and device combining deep learning and causal inference - Google Patents

Network public opinion analysis method and device combining deep learning and causal inference Download PDF

Info

Publication number
CN117494068B
CN117494068B CN202311546925.1A CN202311546925A CN117494068B CN 117494068 B CN117494068 B CN 117494068B CN 202311546925 A CN202311546925 A CN 202311546925A CN 117494068 B CN117494068 B CN 117494068B
Authority
CN
China
Prior art keywords
public opinion
event
network public
analyzed
network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202311546925.1A
Other languages
Chinese (zh)
Other versions
CN117494068A (en
Inventor
王超
王永恒
金星
王靖锐
王震
邵彬
张杨
董世海
恽爽
路游
杨亚飞
鲁艺
周春来
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Lab
Original Assignee
Zhejiang Lab
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang Lab filed Critical Zhejiang Lab
Priority to CN202311546925.1A priority Critical patent/CN117494068B/en
Publication of CN117494068A publication Critical patent/CN117494068A/en
Application granted granted Critical
Publication of CN117494068B publication Critical patent/CN117494068B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/27—Regression, e.g. linear or logistic regression
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G06T11/23—
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Algebra (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

According to the network public opinion analysis method and device combining deep learning and causal inference, a target event is determined, an event to be analyzed related to the target event and a related event affecting the network public opinion of the target event are determined, a time sequence data set of the related event is constructed according to the related event, network public opinion data of the target event is obtained, the network public opinion data is input into a pre-trained analysis model, an emotion tendency characterization value of the network public opinion is obtained, a curve of the emotion tendency characterization value relative to the time sequence data set is fitted according to the time sequence data set and the emotion tendency characterization value, and the influence of the event to be analyzed on the network public opinion is determined according to the occurrence time of the event to be analyzed and the curve. According to the method, whether the event to be analyzed affects the network public opinion is determined on the basis of analyzing the network public opinion.

Description

Network public opinion analysis method and device combining deep learning and causal inference
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to a method and apparatus for online public opinion analysis combining deep learning and causal inference.
Background
With the development of the internet, the public can express his own perspective by speaking through a social network platform for various phenomena and problems in real life. The network public opinion formed by the comments often has strong emotional tendency and influence, reasonably analyzes the network public opinion, monitors the development dynamics of events, is favorable for building a positive public opinion environment for harmonious development of society, and has important realistic demands and significance.
Currently, the mainstream network public opinion analysis method generally only performs visual description on the collected public opinion data of a certain hot event, and cannot determine whether other events occurring simultaneously with the hot event affect the network public opinion.
Therefore, how to determine whether other events affect the internet public opinion of the hot event is a urgent issue to be resolved.
Disclosure of Invention
The embodiment of the specification provides a network public opinion analysis method and device combining deep learning and causal inference, so as to partially solve the problems existing in the prior art.
The embodiment of the specification adopts the following technical scheme:
the present specification provides a network public opinion analysis method combining deep learning and causal inference, the method includes:
determining a target event;
determining an event to be analyzed related to the target event and related events affecting the online public opinion of the target event;
Constructing a time sequence data set of the related event according to the related event;
acquiring network public opinion data of the target event, and inputting the network public opinion data into a pre-trained analysis model to obtain an emotion tendency characterization value of the network public opinion;
Fitting a curve of the emotion tendency characterization value with respect to the time series data set according to the time series data set and the emotion tendency characterization value;
and determining the influence of the event to be analyzed on the network public opinion according to the occurrence time of the event to be analyzed and the curve.
Optionally, constructing a time sequence data set of the related event according to the related event specifically includes:
determining each data acquisition time for acquiring the data of the related event according to the network public opinion and the event to be analyzed;
for each related event, respectively acquiring data of the related event at each data acquisition time;
and constructing a time sequence data set of the related event according to the data acquired at each data acquisition moment.
Optionally, the obtaining the internet public opinion data of the target event specifically includes:
acquiring an acquisition strategy preset for the target event, wherein the acquisition strategy comprises at least one of keywords, target fields and quantity related to the target event;
And acquiring the network public opinion data of the target event at each data acquisition time according to the acquisition strategy.
Optionally, inputting the online public opinion data into a pre-trained analysis model specifically includes:
dividing the network public opinion data into network public opinion data of different time periods according to the occurrence time of the event to be analyzed;
And respectively inputting the network public opinion data of each period into a pre-trained analysis model.
Optionally, the analysis model includes a feature extraction network and an emotion classification network;
the emotional tendency characterization value of the network public opinion is obtained, which concretely comprises the following steps:
extracting the characteristics of the network public opinion data of each period through the characteristic extraction network;
based on the characteristics, analyzing the network public opinion data of each period through the emotion classification network to obtain emotion tendency characterization values of the network public opinion of each period.
Optionally, the analysis model includes a feature extraction network, a topic extraction network;
the method further comprises the steps of:
extracting the characteristics of the network public opinion data of each period through the characteristic extraction network;
performing dimension reduction processing on the characteristics to obtain final characteristics of text information contained in the network public opinion data of each period;
And according to the final characteristics, extracting hot topics contained in the network public opinion data of each period through the topic extraction network.
Optionally, extracting hot topics contained in the online public opinion data of each period specifically includes:
Dividing the text information into a plurality of text information sets through clustering according to the final characteristics, and representing each text information set by using word bags containing a plurality of words;
For each text information set, determining the weight of each word contained in a word bag for representing the text information set;
and determining a subject word in a word bag used for representing the text information set according to the weight of each word, and taking the subject word as a hot topic of the text information set.
Optionally, fitting a curve of the emotion tendency characterization value with respect to the time series data set specifically includes:
and fitting a curve of the emotion tendency characterization value of the network public opinion in each time period with respect to the time sequence data set by a multiple linear regression method according to the emotion tendency characterization value of the network public opinion in each time period and the time sequence data set of the related event in each time period.
Optionally, the curve of the emotion tendency characterization value of the network public opinion fitted in each period with respect to the time series data set specifically includes:
taking the occurrence time of the event to be analyzed as a breakpoint, and fitting a curve of the emotion tendency characterization value of the network public opinion before the occurrence of the related event relative to the time sequence data set before the breakpoint;
after the breakpoint, setting corresponding weights for the events to be analyzed;
fitting a curve of the emotion tendency characterization value of the network public opinion after the related event occurs with respect to the time sequence data set and the event to be analyzed according to the weight of the event to be analyzed;
According to the occurrence time of the event to be analyzed and the curve, determining the influence of the event to be analyzed on the network public opinion specifically comprises the following steps:
And comparing the curves at two sides of the breakpoint to determine the influence of the event to be analyzed on the network public opinion.
The present specification provides an online public opinion analysis device combining deep learning and causal inference, the device includes:
The first determining module is used for determining a target event;
the second determining module is used for determining an event to be analyzed related to the target event and related events affecting the online public opinion of the target event;
The construction module is used for constructing a time sequence data set of the related event according to the related event;
The acquisition module is used for acquiring the network public opinion data of the target event, inputting the network public opinion data into a pre-trained analysis model and obtaining the emotion tendency characterization value of the network public opinion;
The fitting module is used for fitting a curve of the emotion tendency characterization value relative to the time sequence data set according to the time sequence data set and the emotion tendency characterization value;
And the analysis module is used for determining the influence of the event to be analyzed on the network public opinion according to the occurrence time of the event to be analyzed and the curve.
The electronic equipment provided by the specification comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the online public opinion analysis method combining deep learning and causal inference when executing the program.
The above-mentioned at least one technical scheme that this description embodiment adopted can reach following beneficial effect:
According to the online public opinion analysis method combining deep learning and causal inference, a target event is determined, an event to be analyzed related to the target event and related events affecting the online public opinion of the target event are determined, a time sequence data set of the related events is constructed according to the related events, online public opinion data of the target event are obtained, the online public opinion data are input into a pre-trained analysis model to obtain an emotion tendency characterization value of the online public opinion, a curve of the emotion tendency characterization value relative to the time sequence data set is fitted according to the time sequence data set and the emotion tendency characterization value, and the influence of the event to be analyzed on the online public opinion is determined according to the occurrence time of the event to be analyzed and the curve. According to the method, on the basis of analyzing the network public opinion of the target event, a causal inference method is combined, and whether the event to be analyzed affects the network public opinion or not is determined by taking the related event as a reference.
Drawings
The accompanying drawings, which are included to provide a further understanding of the specification, illustrate and explain the exemplary embodiments of the present specification and their description, are not intended to limit the specification unduly. In the drawings:
FIG. 1 is a schematic flow chart of an online public opinion analysis method combining deep learning and causal inference according to an embodiment of the present disclosure;
FIG. 2 is a graph of emotional tendency characterization values provided by embodiments of the present description with respect to related events and events to be analyzed;
FIG. 3 is a schematic diagram of an online public opinion analysis device combining deep learning and causal inference according to an embodiment of the present disclosure;
Fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure, which combines deep learning and causal inference.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the present specification more apparent, the technical solutions of the present specification will be clearly and completely described below with reference to specific embodiments of the present specification and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present specification. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.
The following describes in detail the technical solutions provided by the embodiments of the present specification with reference to the accompanying drawings.
Fig. 1 is a flow chart of a network public opinion analysis method combining deep learning and causal inference in the present specification, which specifically includes the following steps:
S100: a target event is determined.
Currently, a user may acquire and analyze network public opinion for a certain hotspot event by using various applications or services, and after determining the network public opinion of the hotspot event that the user wants to analyze, the applications may provide a corresponding public opinion analysis report for the user, so that the user can know the trend of the network public opinion for the hotspot event or help related personnel to make a corresponding decision according to the network public opinion.
Therefore, the server may be used as a background server of the applications, and after receiving a request of analyzing the internet public opinion of a certain hotspot event, the server may confirm the hotspot event as the target event according to the request.
S102: and determining the event to be analyzed related to the target event and related events affecting the network public opinion of the target event.
After determining the target event, determining an event to be analyzed related to the target event and related events affecting the network public opinion of the target event according to the target event. Specifically, the event to be analyzed refers to an event that may affect the internet public opinion of the target event in the event that occurs simultaneously or recently with the target event, and it is required to further analyze and determine whether the event may affect the internet public opinion of the target event. In contrast, the related events affecting the network public opinion of the target event mentioned herein are some events or factors that have been determined to affect the network public opinion of the target event, and it should be noted that the number of determined events to be analyzed and related events are not limited in the embodiments of the present specification.
For example, for an epidemic situation occurring in a certain place, the event to be analyzed may be a certain sealing policy issued for the epidemic situation, or a certain expert developed a vaccine for the epidemic situation, and the related event may be the number of people dying from the local epidemic situation, the local total production value, and the like.
In addition, the related event may be determined by the user, the server determines the related event through the received request, or after determining the target event, the server determines the related event that may affect the online public opinion of the target event, and performs subsequent analysis of the online public opinion, which is not limited in the embodiment of the present disclosure.
S104: and constructing a time sequence data set of the related events according to the related events.
Specifically, after determining the related event, a corresponding time rule can be formulated for the related event, data of the related event is collected at each moment, and a time sequence data set of the related event is constructed according to the time rule and the collected data of the related event at each moment. If a plurality of related events are determined, different time series data sets can be respectively constructed for each related event according to the same time rule.
Furthermore, when data of related events are collected, in order to improve accuracy of the data, interfaces provided by authoritative software or websites can be accessed, so that needed data can be obtained in real time and efficiently. For example, when it is necessary to acquire domestic total production value (Gross Domestic Product, GDP) data, the GDP data at each time may be acquired through an interface provided by the national statistical office.
S106: and acquiring the network public opinion data of the target event, and inputting the network public opinion data into a pre-trained analysis model to obtain the emotion tendency characterization value of the network public opinion.
Specifically, to analyze the internet public opinion of a target event, the internet public opinion data of the target event needs to be acquired first. The current network public opinion data is comments, articles and the like posted by users aiming at the target event through a social network platform, and correspondingly, the network public opinion data aiming at the target event can be obtained through an interface connected with the network social platform. In addition, network public opinion data input by the user for the target event may also be obtained, and the embodiment of the present description is not limited.
Further, the obtained internet public opinion data is analyzed, the obtained internet public opinion data can be input into a pre-trained analysis model to extract characteristics of the internet public opinion, and emotion tendency characterization values of the internet public opinion are obtained based on the characteristics and are used for representing emotion tendencies of network users publishing the internet public opinion, wherein the emotion tendencies can be positive emotion and negative emotion. Wherein the analysis model may be pre-trained based on text analysis, natural language processing, image recognition, and the like.
S108: fitting a curve of the emotion tendency characterization value with respect to the time series data set according to the time series data set and the emotion tendency characterization value.
Specifically, after the time series data set of the related event is constructed according to the method and the emotion tendency characterization value of the target public opinion is obtained, a suitable curve model, such as a polynomial function, an exponential function or other self-defined functions, can be selected, and a curve of the emotion tendency characterization value with respect to the time series data set is fitted, wherein the curve can reflect the relevance of different related events and emotion tendencies of network users. In the examples of the present specification, the curve is fitted by a multiple linear regression method, and the specific fitting method is not limited in the specification.
S110: and determining the influence of the event to be analyzed on the network public opinion according to the occurrence time of the event to be analyzed and the curve.
Specifically, when the curve is fitted by a causal inference method, according to the occurrence time of the event to be analyzed, the curve of the emotion tendency characterization value about the event to be analyzed and the related event is fitted after the occurrence time, so as to determine whether the event to be analyzed affects the network public opinion of the target event.
Based on the network public opinion analysis method combining deep learning and causal inference provided by fig. 1, after determining a target event, an event to be analyzed and related events and obtaining the network public opinion of the target event, determining whether the event to be analyzed affects the network public opinion by taking the related events as a reference on the basis of analyzing the network public opinion of the target event.
Further, the time sequence data set of the related event is constructed according to the related event, specifically, each data acquisition time for acquiring the data of the related event is determined according to the network public opinion and the event to be analyzed, the data of the related event is acquired at each data acquisition time for each related event, and the time sequence data set of the related event is constructed according to the data acquired at each data acquisition time.
In other words, each data acquisition time for acquiring the data of the related event may be determined according to the characteristics of the target event and the event to be analyzed. For example, the target event duration is 3 months, the occurrence time interval of each event to be analyzed is usually several weeks, and then each data acquisition time for acquiring the related event data can be determined at a day-by-day interval. After each data acquisition time is determined, the data of the related event are acquired at each data acquisition time. And then, constructing a time sequence data set of related events comprising each data acquisition time and the data acquired at each data acquisition time according to the data acquired at each data acquisition time. Specifically, the representation of the time series data set of the related event is as follows:
Ei＝[(D1,T1),(D2,T2),...,(Dt-1,Tt-1),(Dt,Tt)]
Wherein E i represents the ith relevant event such as GDP, rate of loss of business, population growth rate, etc., which affects the network public opinion of the target event, T t represents the T data acquisition time, and D t represents the data of the relevant event at the time T t, such as a specific value of rate of loss of business.
In addition, the number of the determined events to be analyzed may be multiple, and for the multiple events to be analyzed, a time sequence data set of the events to be analyzed may be constructed according to the occurrence time of each event to be analyzed, where a specific representation form is as follows:
TD＝[(A1,T1),(A1,T1),(A2,T2),...,(At-1,Tt-1),(At,Tt)]
Wherein a t represents a T-th event to be analyzed which may affect the online public opinion of the target event, and T t represents an occurrence time of the T-th event to be analyzed.
Further, the acquiring the network public opinion data of the target event may specifically be acquiring an acquisition policy preset for the target event, where the acquisition policy includes at least one of a keyword, a target field and a number related to the target event, and the network public opinion data of the target event is acquired at each data acquisition time according to the acquisition policy.
Specifically, to obtain the online public opinion data of the target event, after determining the target event, the data acquisition may be queried and performed according to an acquisition policy set in advance for the target event. The collection strategy mentioned herein contains at least one of keywords, target fields and quantity related to the target event, wherein the keywords are words which must be contained in text information of collected network public opinion data, and an appropriate amount of words related to the keywords can be determined at the same time of determining the keywords. The target field is a field to be collected, for example, a nickname of the sender, a geographical location of the sender, a time of the sender, a content of the sender, a forwarding number, a comment number, a praise number, and the like, which are determined according to a data format of the social network platform. In addition, the determined collection strategy may not be queried for the collection strategy preset for the target event, and the collection strategy is manually input by the user, that is, the server collects the network public opinion according to the received request after the user determines the information such as the keyword, the target field, the number and the like, and the embodiment of the specification is not limited in specific manner.
It should be noted that, each time of collecting the network public opinion data should correspond to each data collection time in the time sequence data set for constructing the related event, that is, the network public opinion data of the target event is obtained according to the collection policy at each data collection time of collecting the data of the related event, so as to facilitate subsequent analysis of the network public opinion.
Further, the network public opinion data is input to a pre-trained analysis model, specifically, the network public opinion data is divided into network public opinion data of different time periods according to the occurrence time of the event to be analyzed, and the network public opinion data of each time period is respectively input to the pre-trained analysis model.
Specifically, in order to facilitate the subsequent analysis of the network public opinion data before and after the occurrence of the event to be analyzed and determine the influence of the event to be analyzed on the network public opinion, the obtained network public opinion data can be divided into network public opinion data of different time periods according to the occurrence time of the event to be analyzed, and the network public opinion data of different time periods are respectively input into a pre-trained analysis model for analysis.
Further, the analysis model includes a feature extraction network and an emotion classification network, so as to obtain an emotion tendency characterization value of the network public opinion, which may specifically be that features of the network public opinion data of each period are extracted through the feature extraction network, and based on the features, the network public opinion data of each period is analyzed through the emotion classification network, so as to obtain an emotion tendency characterization value of the network public opinion of each period.
In other words, the analysis model comprises two sub-networks of a feature extraction network and an emotion classification network, wherein the analysis model is used for extracting features of network public opinion data, and the emotion analysis network is used for performing emotion analysis on the network public opinion data. Specifically, for the text information in the online public opinion, such as comments and articles for the target event, features of the text information are extracted through a feature extraction network for subsequent processing. The feature extraction network mentioned here may be a pre-trained model based on BERT improvement, which converts the text information of the network public opinion to obtain word embedding vectors, i.e. text information that cannot be directly calculated is converted into mathematical vectors that can be calculated. For example, word embedding vectors may be calculated using the paraphrase-multilingual-MiniLM-L12-v2 model integrated in the sentence-transducer dependency library of Python. In addition, to avoid word embedding vectors from being too redundant to affect subsequent computations, the resulting word embedding vectors may be pruned, leaving only the first 768-dimensional features.
And carrying out emotion analysis through the emotion analysis network based on the characteristics of the extracted text information of the network public opinion. In the embodiment of the present specification, the emotion analysis task is constructed as a classification problem, that is, emotion tendencies of text information only have two categories of negative emotion and positive emotion, and emotion tendencies can be represented by emotion tendencies characterization values. Specifically, the emotion tendency characterization value can be defined as v, the value range of v is between 0 and 1, and if v is less than or equal to 0.5, the corresponding public opinion information is negative emotion; if v >0.5, the corresponding public opinion information is positive emotion. The emotion analysis network is a deep learning-based pre-training classification model, such as a BERT downstream Chinese emotion classification model which is already trained in an open source code warehouse, and takes the characteristics of the word embedded vector, namely the network public opinion data of each period as input to calculate emotion tendency characterization values of the corresponding network public opinion data without additional training. It should be noted that, for different target events, the model may also be trained by using a data set formed by corresponding network public opinion data based on the existing classification model, so as to further improve accuracy of the model.
Further, the curve of the emotion tendency characterization value with respect to the time series data set may be specifically a curve of the emotion tendency characterization value of the network public opinion in each period with respect to the time series data set according to the emotion tendency characterization value of the network public opinion in each period and the time series data set of the related event in each period by a multiple linear regression method.
Specifically, the method may use the relevant emotion tendency characterization value of each data collection time in each period as a dependent variable, use the sequence in the time sequence data set of the related event as an independent variable, fit a function of the emotion tendency characterization value of the network public opinion in each period with respect to the time sequence data set, and obtain the curve according to the function. The function obtained by fitting contains variables associated with each related event, and in the fitting process, the weight of each related event needs to be determined, namely the weight of the variable corresponding to each related event is determined. For example, the fitted function is shown in the following graph:
z＝ax1+bx2+cx3
where x i represents the ith correlation event and the coefficients preceding x i represent the weights of the correlation events, e.g., a represents the weight of the correlation event x 1.
Further, the curve of the emotion tendency characterization value of the network public opinion in each period with respect to the time series data set may specifically be that the occurrence time of the event to be analyzed is taken as a breakpoint, the curve of the emotion tendency characterization value of the network public opinion before the occurrence of the related event with respect to the time series data set is fitted before the breakpoint, after the breakpoint, a corresponding weight is set for the event to be analyzed, and the curve of the emotion tendency characterization value of the network public opinion after the occurrence of the related event with respect to the time series data set and the event to be analyzed is fitted according to the weight of the event to be analyzed.
Specifically, the method is a breakpoint design method, the occurrence time of the event to be analyzed is taken as a breakpoint, the weight of the event to be analyzed is set to zero before the breakpoint, namely, a relation curve of the emotion tendency characterization value and the related event is fitted as a comparison without considering the event to be analyzed, and after the breakpoint, the curve of the emotion tendency characterization value relative to the time sequence data set and the event to be analyzed is obtained by setting the corresponding weight of the event to be analyzed and fitting the curve.
For example, for a fitted function z=ax 1+bx2+cx3+dy1, before the breakpoint, setting a coefficient of an event to be analyzed corresponding item dy 1 to 0, obtaining a curve of an emotion tendency characterization value with respect to a related event before the occurrence time of the event to be analyzed y 1, after the breakpoint, determining a value of d in dy 1, that is, a weight of the event to be analyzed, and obtaining a curve of an emotion tendency characterization value with respect to the time series dataset and the event to be analyzed after the occurrence time of the event to be analyzed y 1.
If a plurality of events to be analyzed exist, a plurality of breakpoints can be set for analysis. As shown in fig. 2, the graph is a graph obtained after the emotion tendency characterization value is fitted with respect to the related event and the event to be analyzed, the time of the x axis in the graph is the time of each data acquisition, the number of the y axis is the emotion tendency characterization value, each point in the graph is a point according to which the graph is fitted, the emotion tendency characterization value corresponding to each sequence in the time sequence data set of each related event and the data acquisition time in each sequence is determined, and the graph in the graph is obtained by fitting according to each point. In addition, the dot-dashed line perpendicular to the x-axis in fig. 2 is generated according to the occurrence time of each event to be analyzed, so as to observe the change trend of the curves before and after the occurrence of the event to be analyzed.
Further, after the curves on the two sides of the breakpoint are obtained, according to the occurrence time of the event to be analyzed and the curves, the influence of the event to be analyzed on the network public opinion is determined, specifically, the curves on the two sides of the breakpoint are compared, and the influence of the event to be analyzed on the network public opinion is determined.
Specifically, after the curves at the two sides of the breakpoint are obtained, the curves at the two sides of the breakpoint are compared, and whether the event to be analyzed affects the network public opinion of the target event can be determined. As shown in fig. 2, for the dotted line in the graph, curves on two sides of the dotted line are compared, and if the trend changes greatly, it is indicated that the event to be analyzed affects the network public opinion of the target event.
In addition, the analysis model further comprises a topic extraction network on the basis of comprising a feature extraction model network and an emotion analysis network, and topic extraction can be carried out on text information in the acquired network public opinion data. The method specifically comprises the steps of extracting the characteristics of the network public opinion data of each period through the characteristic extraction network, performing dimension reduction processing on the characteristics to obtain final characteristics of text information contained in the network public opinion data of each period, and extracting hot topics contained in the network public opinion data of each period through the topic extraction network according to the final characteristics.
Specifically, after the characteristics of the network public opinion data are obtained through the characteristic extraction network according to the method, hot topics contained in the network public opinion data in each period can be extracted through the topic extraction network based on the characteristics. The present embodiment uses the BERT model-based topic model algorithm BerTopic algorithm for topic extraction, berTopic is an unsupervised learning algorithm for automatically extracting topics from a large amount of text. Because the feature obtained by the method, namely the dimension of the word embedding vector is higher, the calculation of the clustering module in the follow-up BeTopic is difficult to realize, the word embedding vector can be subjected to dimension reduction processing based on a BerTopic default nonlinear dimension reduction algorithm (Uniform Manifold Approximation and Projection, UMAP), and the final feature of the network public opinion data is obtained and is used for the follow-up topic extraction. The UMAP algorithm maps high-dimensional data to a low-dimensional space through learning the manifold structure of the data, retains structural information of the high-dimensional data, is commonly used for visualization and embedding of the high-dimensional data, and can effectively extract main characteristics of network public opinion data through the built-in UMAP dimension reduction of BerTopic, so that more concise input characteristics are provided for subsequent work.
Further, the hot topics contained in the network public opinion data of each period may be extracted, specifically, the text information is divided into a plurality of text information sets by clustering according to the final feature, each text information set is represented by using a word bag containing a plurality of words, the weight of each word contained in the word bag for representing the text information set is determined for each text information set, and the subject word is determined in the word bag for representing the text information set according to the weight of each word as the hot topic of the text information set.
Specifically, after obtaining the final feature after dimension reduction, the data may be clustered based on a density-based clustering technique HDBSCAN, divided into a plurality of text information sets, and the result obtained by the clustering is represented by using a word bag containing a plurality of words. Wherein HDBSCAN is a density-based clustering algorithm, when performing cluster analysis on network public opinion feature data, HDBSCAN can automatically determine the actual cluster number of text information, and divide the feature data into text sets of different topic information, so as to provide support for subsequent analysis. After the clustering is completed, a class-based TF-IDF algorithm can be utilized to generate a theme representation in the word bag, namely, the weight of each word contained in the word bag for representing the text information set is determined, and the theme word is determined in the word bag for representing the text information set according to the weight of each word and is used as a hot topic of the text information set. Specifically, determining the weight of each vocabulary can be achieved by calculating the influence of each vocabulary, and the specific formula is as follows:
Wherein tf x,c represents the frequency of the vocabulary x in the clustered text information set c; f x denotes the frequency of the vocabulary x in all text information sets; a represents the average vocabulary that each text information set contains.
Furthermore, after emotion analysis and topic extraction are performed on the network public opinion data, multidimensional analysis can be performed on the network public opinion according to information such as the number of endorsements, the author of a lesson, the geographic position and the like, and the specific analysis mode is not limited in the embodiment of the specification. For example, according to the geographic position in the network public opinion data, the emotion tendencies and hot topics of each region can be visually displayed.
And finally, completing the analysis of the network public opinion data, and after determining the influence of the event to be analyzed on the network public opinion, presenting the analysis result to the user in the form of an analysis report. The analysis report can comprise fitting curves and result text descriptions of emotion tendency characterization values about related events, breakpoint regression analysis results of emotion tendency characterization values of events to be analyzed, which occur at different moments, information such as emotion evolution processes of network users, topic keywords of network discussion and the like, wherein the emotion tendency characterization values are obtained based on multiple linear regression.
The above provides a network public opinion analysis method combining deep learning and causal inference for the embodiments of the present specification, and based on the same thought, the present specification further provides a corresponding device, a storage medium, and an electronic apparatus.
Fig. 3 is a schematic structural diagram of an online public opinion analysis device combining deep learning and causal inference according to an embodiment of the present disclosure, where the device includes:
a first determining module 300, configured to determine a target event;
a second determining module 302, configured to determine an event to be analyzed related to the target event and a related event affecting network public opinion of the target event;
a construction module 304, configured to construct a time sequence data set of the related event according to the related event;
the obtaining module 306 is configured to obtain online public opinion data of the target event, input the online public opinion data into a pre-trained analysis model, and obtain an emotion tendency characterization value of the online public opinion;
a fitting module 308, configured to fit a curve of the emotion tendency characterization value with respect to the time series data set according to the time series data set and the emotion tendency characterization value;
the analysis module 310 is configured to determine an influence of the event to be analyzed on the online public opinion according to the occurrence time of the event to be analyzed and the curve.
Optionally, the construction module 304 is specifically configured to determine each data collection time for collecting the data of the related event according to the online public opinion and the event to be analyzed, collect, for each related event, the data of the related event at each data collection time, and construct a time sequence dataset of the related event according to the data collected at each data collection time.
Optionally, the acquiring module 306 is specifically configured to acquire an acquisition policy set in advance for the target event, where the acquisition policy includes at least one of a keyword, a target field, and a number related to the target event, and acquire, according to the acquisition policy, the online public opinion data of the target event at each data acquisition time.
Optionally, the obtaining module 306 is specifically configured to divide the online public opinion data into online public opinion data of different time periods according to the occurrence time of the event to be analyzed, and input the online public opinion data of each time period into a pre-trained analysis model respectively.
Optionally, the analysis model includes a feature extraction network and an emotion classification network, and the obtaining module 306 is specifically configured to extract, through the feature extraction network, features of the network public opinion data of each period, and analyze, based on the features, the network public opinion data of each period through the emotion classification network, to obtain emotion tendency characterization values of the network public opinion of each period.
Optionally, the analysis model includes a feature extraction network and a topic extraction network, and the device is further configured to extract, through the feature extraction network, features of the network public opinion data of each period, perform dimension reduction processing on the features to obtain final features of text information included in the network public opinion data of each period, and extract, according to the final features, hot topics included in the network public opinion data of each period through the topic extraction network.
Optionally, the device is further configured to divide the text information into a plurality of text information sets by clustering according to the final feature, and, for each text information set, determine weights of words included in a word bag for representing the text information set by using a word bag including a plurality of words, and determine, according to the weights of the words, subject words in the word bag for representing the text information set as hot topics of the text information set.
Optionally, the fitting module 308 is specifically configured to fit, according to the emotion tendency characterization value of the online public opinion in each period and the time series data set of the related event in each period, a curve of the emotion tendency characterization value of the online public opinion in each period with respect to the time series data set by using a multiple linear regression method.
Optionally, the fitting module 308 is specifically configured to use an occurrence time of the event to be analyzed as a breakpoint, and before the breakpoint, fit a curve of an emotion tendency characterization value of the network public opinion before occurrence of the related event with respect to the time series data set, after the breakpoint, set a corresponding weight for the event to be analyzed, and fit a curve of an emotion tendency characterization value of the network public opinion after occurrence of the related event with respect to the time series data set and the event to be analyzed according to the weight of the event to be analyzed; the analysis module 310 is specifically configured to compare the curves at two sides of the breakpoint, and determine an influence of the event to be analyzed on the network public opinion.
The present disclosure also provides a computer readable storage medium storing a computer program which when executed by a processor is operable to perform the above-described network public opinion analysis method of fig. 1 combining deep learning and causal inference.
Based on the online public opinion analysis method combining deep learning and causal inference shown in fig. 1, the embodiment of the present disclosure further provides a structural schematic diagram of the electronic device shown in fig. 4. At the hardware level, as in fig. 4, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile storage, although it may include hardware required for other services. The processor reads the corresponding computer program from the nonvolatile memory into the memory and then runs the computer program to realize the network public opinion analysis method combining deep learning and causal inference as described in the above figure 1.
Of course, other implementations, such as logic devices or combinations of hardware and software, are not excluded from the present description, that is, the execution subject of the following processing flows is not limited to each logic unit, but may be hardware or logic devices.
In the 90 s of the 20 th century, improvements to one technology could clearly be distinguished as improvements in hardware (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software (improvements to the process flow). However, with the development of technology, many improvements of the current method flows can be regarded as direct improvements of hardware circuit structures. Designers almost always obtain corresponding hardware circuit structures by programming improved method flows into hardware circuits. Therefore, an improvement of a method flow cannot be said to be realized by a hardware entity module. For example, a programmable logic device (Programmable Logic Device, PLD) (e.g., field programmable gate array (Field Programmable GATE ARRAY, FPGA)) is an integrated circuit whose logic functions are determined by user programming of the device. A designer programs to "integrate" a digital system onto a PLD without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, such programming is mostly implemented with "logic compiler (logic compiler)" software, which is similar to the software compiler used in program development and writing, and the original code before being compiled is also written in a specific programming language, which is called hardware description language (Hardware Description Language, HDL), but HDL is not just one, but a plurality of kinds, such as ABEL(Advanced Boolean Expression Language)、AHDL(Altera Hardware Description Language)、Confluence、CUPL(Cornell University Programming Language)、HDCal、JHDL(Java Hardware Description Language)、Lava、Lola、MyHDL、PALASM、RHDL(Ruby Hardware Description Language), and VHDL (very-high-SPEED INTEGRATED Circuit Hardware Description Language) and verilog are currently most commonly used. It will also be apparent to those skilled in the art that a hardware circuit implementing the logic method flow can be readily obtained by merely slightly programming the method flow into an integrated circuit using several of the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer readable medium storing computer readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, application SPECIFIC INTEGRATED Circuits (ASICs), programmable logic controllers, and embedded microcontrollers, examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic of the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller in a pure computer readable program code, it is well possible to implement the same functionality by logically programming the method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Such a controller may thus be regarded as a kind of hardware component, and means for performing various functions included therein may also be regarded as structures within the hardware component. Or even means for achieving the various functions may be regarded as either software modules implementing the methods or structures within hardware components.
The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. One typical implementation is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being functionally divided into various units, respectively. Of course, the functions of each element may be implemented in one or more software and/or hardware elements when implemented in the present specification.
It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
The present description is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
The description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, as relevant to see a section of the description of method embodiments.
The foregoing is merely exemplary of the present disclosure and is not intended to limit the disclosure. Various modifications and alterations to this specification will become apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, or the like, which are within the spirit and principles of the present description, are intended to be included within the scope of the claims of the present description.

Claims (11)

1.一种结合深度学习与因果推断的网络舆情分析方法，其特征在于，所述方法包括：1. A network public opinion analysis method combining deep learning and causal inference, characterized in that the method comprises: 确定目标事件；Identify target events; 确定与所述目标事件相关的待分析事件和影响所述目标事件的网络舆情的相关事件；Determine the events to be analyzed that are related to the target event and the related events that affect the network public opinion of the target event; 根据所述相关事件，构建所述相关事件的时间序列数据集；According to the related events, construct a time series data set of the related events; 获取所述目标事件的网络舆情数据，将所述网络舆情数据输入到预先训练的分析模型，得到所述网络舆情的情感倾向表征值；Acquire the network public opinion data of the target event, input the network public opinion data into a pre-trained analysis model, and obtain a sentiment tendency representation value of the network public opinion; 根据所述时间序列数据集和所述情感倾向表征值，拟合所述情感倾向表征值关于所述时间序列数据集的曲线；According to the time series data set and the sentiment tendency representation value, fitting a curve of the sentiment tendency representation value with respect to the time series data set; 根据所述待分析事件的发生时刻以及所述曲线，确定所述待分析事件对网络舆情的影响；Determine the impact of the event to be analyzed on network public opinion according to the occurrence time of the event to be analyzed and the curve; 其中，获取所述网络舆情数据的各时刻与所述相关事件的时间序列数据集中的各数据的采集时刻相对应；The time at which the network public opinion data is obtained corresponds to the time at which each data in the time series data set of the relevant event is collected; 所述拟合所述情感倾向表征值关于所述时间序列数据集的曲线，具体包括：The step of fitting the curve of the sentiment tendency representation value with respect to the time series data set specifically includes: 将所述待分析事件的发生时刻作为断点，在所述断点之前，拟合所述相关事件发生前的网络舆情的情感倾向表征值关于所述时间序列数据集的曲线；The occurrence time of the event to be analyzed is taken as a breakpoint, and before the breakpoint, a curve of the sentiment tendency representation value of the network public opinion before the occurrence of the relevant event with respect to the time series data set is fitted; 在所述断点之后，为所述待分析事件设置相应的权重；After the breakpoint, setting a corresponding weight for the event to be analyzed; 根据所述待分析事件的权重，拟合所述相关事件发生后的网络舆情的情感倾向表征值关于所述时间序列数据集和所述待分析事件的曲线；According to the weight of the event to be analyzed, fitting a curve of the sentiment tendency representation value of the network public opinion after the relevant event occurs with respect to the time series data set and the event to be analyzed; 根据所述待分析事件的发生时刻以及所述曲线，确定所述待分析事件对网络舆情的影响，具体包括：Determining the impact of the event to be analyzed on network public opinion according to the occurrence time of the event to be analyzed and the curve, specifically including: 对所述断点两侧曲线进行比较，确定所述待分析事件对所述网络舆情的影响。Compare the curves on both sides of the breakpoint to determine the impact of the event to be analyzed on the network public opinion. 2.如权利要求1所述的方法，其特征在于，根据所述相关事件，构建所述相关事件的时间序列数据集，具体包括：2. The method according to claim 1, characterized in that, according to the related events, constructing a time series data set of the related events specifically comprises: 根据所述网络舆情的和所述待分析事件，确定采集所述相关事件的数据的各数据采集时刻；Determine the data collection time for collecting data of the related events according to the network public opinion and the events to be analyzed; 针对每个相关事件，分别在所述各数据采集时刻采集该相关事件的数据；For each relevant event, respectively collect data of the relevant event at each data collection moment; 根据所述在所述各数据采集时刻采集的数据，构建该相关事件的时间序列数据集。A time series data set of the related events is constructed based on the data collected at each data collection moment. 3.如权利要求2所述的方法，其特征在于，获取所述目标事件的网络舆情数据，具体包括：3. The method according to claim 2, wherein obtaining the network public opinion data of the target event specifically comprises: 获取预先针对所述目标事件设置的采集策略，所述采集策略包括与所述目标事件相关的关键词、目标字段和数量中的至少一种；Acquire a collection strategy pre-set for the target event, the collection strategy including at least one of a keyword, a target field, and a quantity related to the target event; 根据所述采集策略，分别在所述各数据采集时刻，获取所述目标事件的所述网络舆情数据。According to the collection strategy, the network public opinion data of the target event is obtained at each data collection moment. 4.如权利要求1所述的方法，其特征在于，将所述网络舆情数据输入到预先训练的分析模型，具体包括：4. The method according to claim 1, wherein inputting the network public opinion data into a pre-trained analysis model specifically comprises: 根据所述待分析事件的发生时刻，将所述网络舆情数据分割为不同时段的网络舆情数据；According to the occurrence time of the event to be analyzed, the network public opinion data is divided into network public opinion data of different time periods; 分别将所述各时段的网络舆情数据输入到预先训练的分析模型。The network public opinion data of each time period is respectively input into the pre-trained analysis model. 5.如权利要求4所述的方法，其特征在于，所述分析模型包括特征提取网络、情感分类网络；5. The method according to claim 4, wherein the analysis model comprises a feature extraction network and a sentiment classification network; 得到所述网络舆情的情感倾向表征值，具体包括：Obtaining the sentiment tendency representation value of the network public opinion specifically includes: 通过所述特征提取网络，提取所述各时段的网络舆情数据的特征；Extracting the features of the network public opinion data of each time period through the feature extraction network; 基于所述特征，通过所述情感分类网络，对所述各时段的网络舆情数据进行分析，得到所述各时段的网络舆情的情感倾向表征值。Based on the features, the network public opinion data of each time period is analyzed through the sentiment classification network to obtain the sentiment tendency representation value of the network public opinion of each time period. 6.如权利要求1所述的方法，其特征在于，所述分析模型包括特征提取网络、话题提取网络；6. The method according to claim 1, wherein the analysis model comprises a feature extraction network and a topic extraction network; 所述方法还包括：The method further comprises: 通过所述特征提取网络，提取各时段的网络舆情数据的特征；Extracting the features of the network public opinion data of each time period through the feature extraction network; 将所述特征进行降维处理，得到所述各时段的网络舆情数据中包含的文本信息的最终特征；Performing dimensionality reduction processing on the features to obtain final features of the text information contained in the network public opinion data of each time period; 根据所述最终特征，通过所述话题提取网络，提取所述各时段的网络舆情数据中包含的热点话题。According to the final feature, the hot topics contained in the network public opinion data of each time period are extracted through the topic extraction network. 7.如权利要求6所述的方法，其特征在于，提取所述各时段的网络舆情数据中包含的热点话题，具体包括：7. The method according to claim 6, characterized in that extracting the hot topics contained in the network public opinion data of each time period specifically includes: 根据所述最终特征，将所述文本信息通过聚类分为多个文本信息集，并，通过使用包含若干个词汇的词袋表示各所述文本信息集；According to the final features, the text information is divided into a plurality of text information sets by clustering, and each of the text information sets is represented by using a bag of words containing a plurality of words; 针对所述各文本信息集，确定用于表示该文本信息集的词袋中包含的各词汇的权重；For each of the text information sets, determining the weight of each word contained in a word bag used to represent the text information set; 根据所述各词汇的权重，在用于表示该文本信息集的词袋中确定主题词，作为该文本信息集的热点话题。According to the weights of the words, a subject word is determined in a word bag used to represent the text information set as a hot topic of the text information set. 8.如权利要求5所述的方法，其特征在于，拟合所述情感倾向表征值关于所述时间序列数据集的曲线，具体包括：8. The method according to claim 5, wherein fitting the curve of the sentiment tendency representation value with respect to the time series data set specifically comprises: 根据所述各时段的网络舆情的情感倾向表征值，以及各时段的所述相关事件的时间序列数据集，通过多元线性回归方法，拟合在所述各时段的网络舆情的情感倾向表征值关于所述时间序列数据集的曲线。According to the sentiment tendency representation value of the network public opinion in each time period and the time series data set of the related events in each time period, a curve of the sentiment tendency representation value of the network public opinion in each time period with respect to the time series data set is fitted by a multivariate linear regression method. 9.一种装置，其特征在于，所述装置包括：9. A device, characterized in that the device comprises: 第一确定模块，用于确定目标事件；A first determination module, used to determine a target event; 第二确定模块，用于确定与所述目标事件相关的待分析事件和影响所述目标事件的网络舆情的相关事件；The second determination module is used to determine the events to be analyzed related to the target event and the related events of the network public opinion that affect the target event; 构建模块，用于根据所述相关事件，构建所述相关事件的时间序列数据集；A construction module, used to construct a time series data set of the related events according to the related events; 获取模块，用于获取所述目标事件的网络舆情数据，将所述网络舆情数据输入到预先训练的分析模型，得到所述网络舆情的情感倾向表征值；An acquisition module is used to acquire the network public opinion data of the target event, input the network public opinion data into a pre-trained analysis model, and obtain a sentiment tendency representation value of the network public opinion; 拟合模块，用于根据所述时间序列数据集和所述情感倾向表征值，拟合所述情感倾向表征值关于所述时间序列数据集的曲线；A fitting module, used for fitting a curve of the emotional tendency representation value with respect to the time series data set according to the time series data set and the emotional tendency representation value; 分析模块，用于根据所述待分析事件的发生时刻以及所述曲线，确定所述待分析事件对网络舆情的影响；An analysis module, used to determine the impact of the event to be analyzed on network public opinion according to the occurrence time of the event to be analyzed and the curve; 其中，所述获取所述网络舆情数据的各时刻与所述相关事件的时间序列数据集中的各数据的采集时刻相对应；The time at which the network public opinion data is obtained corresponds to the time at which each data in the time series data set of the relevant event is collected; 所述拟合模块，具体用于将所述待分析事件的发生时刻作为断点，在所述断点之前，拟合所述相关事件发生前的网络舆情的情感倾向表征值关于所述时间序列数据集的曲线；在所述断点之后，为所述待分析事件设置相应的权重；根据所述待分析事件的权重，拟合所述相关事件发生后的网络舆情的情感倾向表征值关于所述时间序列数据集和所述待分析事件的曲线；The fitting module is specifically used to take the occurrence time of the event to be analyzed as a breakpoint, and before the breakpoint, fit the curve of the emotional tendency representation value of the network public opinion before the relevant event occurs with respect to the time series data set; after the breakpoint, set a corresponding weight for the event to be analyzed; according to the weight of the event to be analyzed, fit the curve of the emotional tendency representation value of the network public opinion after the relevant event occurs with respect to the time series data set and the event to be analyzed; 所述分析模块，具体用于对所述断点两侧曲线进行比较，确定所述待分析事件对所述网络舆情的影响。The analysis module is specifically used to compare the curves on both sides of the breakpoint to determine the impact of the event to be analyzed on the network public opinion. 10.一种计算机可读存储介质，其特征在于，所述存储介质存储有计算机程序，所述计算机程序被处理器执行时实现上述权利要求1-8任一项所述的方法。10. A computer-readable storage medium, characterized in that the storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented. 11.一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述程序时实现上述权利要求1-8任一项所述的方法。11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 8 when executing the program.
CN202311546925.1A 2023-11-17 2023-11-17 Network public opinion analysis method and device combining deep learning and causal inference Active CN117494068B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311546925.1A CN117494068B (en) 2023-11-17 2023-11-17 Network public opinion analysis method and device combining deep learning and causal inference

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311546925.1A CN117494068B (en) 2023-11-17 2023-11-17 Network public opinion analysis method and device combining deep learning and causal inference

Publications (2)

Publication Number Publication Date
CN117494068A CN117494068A (en) 2024-02-02
CN117494068B true CN117494068B (en) 2024-04-19

Family

ID=89676332

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311546925.1A Active CN117494068B (en) 2023-11-17 2023-11-17 Network public opinion analysis method and device combining deep learning and causal inference

Country Status (1)

Country Link
CN (1) CN117494068B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119151333A (en) * 2024-09-13 2024-12-17 怀化学院 Digital economic digital governance analysis method and system
CN119622457B (en) * 2024-11-21 2025-12-05 甘肃政法大学 AI-based methods and systems for monitoring public opinion on social networks
CN119671285A (en) * 2024-12-11 2025-03-21 南京航空航天大学 A dual-channel low-altitude wind shear aviation safety risk causal extraction method based on knowledge integration and competition mechanism

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106339463A (en) * 2016-08-26 2017-01-18 中国传媒大学 Network public opinion early-warning system based on logistic model and early-warning method thereof
CN109145216A (en) * 2018-08-29 2019-01-04 中国平安保险(集团)股份有限公司 Network public-opinion monitoring method, device and storage medium
CN109325524A (en) * 2018-08-31 2019-02-12 中国科学院自动化研究所 Event tracking and change stage division method, system and related equipment
CN110232109A (en) * 2019-05-17 2019-09-13 深圳市兴海物联科技有限公司 A kind of Internet public opinion analysis method and system
CN114528416A (en) * 2022-02-25 2022-05-24 上海数仑数字信息技术有限公司 Enterprise public opinion environment monitoring method and system based on big data
WO2022116771A1 (en) * 2020-12-02 2022-06-09 Zhejiang Dahua Technology Co., Ltd. Method for analyzing emotion shown in image and related devices
CN115270807A (en) * 2022-06-29 2022-11-01 国家计算机网络与信息安全管理中心 Method, device and equipment for judging emotional tendency of network user and storage medium
CN116881395A (en) * 2023-06-20 2023-10-13 中国工商银行股份有限公司 Public opinion information detection method and device

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8341110B2 (en) * 2005-04-04 2012-12-25 Geoeye Analytics Inc. Temporal-influenced geospatial modeling system and method
US20130290232A1 (en) * 2012-04-30 2013-10-31 Mikalai Tsytsarau Identifying news events that cause a shift in sentiment

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106339463A (en) * 2016-08-26 2017-01-18 中国传媒大学 Network public opinion early-warning system based on logistic model and early-warning method thereof
CN109145216A (en) * 2018-08-29 2019-01-04 中国平安保险(集团)股份有限公司 Network public-opinion monitoring method, device and storage medium
CN109325524A (en) * 2018-08-31 2019-02-12 中国科学院自动化研究所 Event tracking and change stage division method, system and related equipment
CN110232109A (en) * 2019-05-17 2019-09-13 深圳市兴海物联科技有限公司 A kind of Internet public opinion analysis method and system
WO2022116771A1 (en) * 2020-12-02 2022-06-09 Zhejiang Dahua Technology Co., Ltd. Method for analyzing emotion shown in image and related devices
CN114528416A (en) * 2022-02-25 2022-05-24 上海数仑数字信息技术有限公司 Enterprise public opinion environment monitoring method and system based on big data
CN115270807A (en) * 2022-06-29 2022-11-01 国家计算机网络与信息安全管理中心 Method, device and equipment for judging emotional tendency of network user and storage medium
CN116881395A (en) * 2023-06-20 2023-10-13 中国工商银行股份有限公司 Public opinion information detection method and device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于舆情评论数据挖掘的政府回应策略优化方法研究——以新冠疫情援助物资使用舆情事件为例;马宁等;《电子政务》;20210910;全文 *

Also Published As

Publication number Publication date
CN117494068A (en) 2024-02-02

Similar Documents

Publication Publication Date Title
CN117494068B (en) Network public opinion analysis method and device combining deep learning and causal inference
CN116227474B (en) A method, device, storage medium and electronic device for generating confrontational text
CN108596410B (en) A kind of wind control event automatic processing method and device
US20190114711A1 (en) Financial analysis system and method for unstructured text data
CN115840802A (en) Service processing method and device
CN117033668B (en) Knowledge graph quality assessment method and device, storage medium and electronic equipment
CN105989066A (en) Information processing method and device
CN117787418A (en) A risk identification method, device, storage medium and electronic equipment
CN118691072A (en) A training method, device and readable storage medium for risk control model
CN119988973A (en) Model training method, business execution method, device, medium and electronic equipment
CN110008394B (en) A method, device and equipment for identifying public opinion information
CN115862675B (en) Emotion recognition method, device, equipment and storage medium
CN118193757B (en) Task execution method and device, storage medium and electronic equipment
CN114676257A (en) A method and device for determining a dialogue topic
CN118152591B (en) A method, device, equipment and readable storage medium for constructing a knowledge graph
CN118036668B (en) GPT model-oriented comprehensive evaluation method
CN117332282B (en) Knowledge graph-based event matching method and device
CN114519631B (en) Financing data processing method and device based on product matching
CN115130000B (en) Information recommendation method and device, storage medium and electronic equipment
CN118964167A (en) A large model risk assessment method, device and equipment
CN117593003A (en) A method, device, storage medium and electronic equipment for model training
CN117217858A (en) Artificial intelligence-based item recommendation methods, devices, equipment and storage media
CN106776529B (en) Business emotion analysis method and device
CN117113174A (en) A method, device, storage medium and electronic equipment for model training
CN116629248A (en) Keyword extraction method, device and equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant