Network public opinion key influence factor identification method based on influence degree optimization

Abstract

The invention relates to a network public opinion key influence factor identification method based on influence degree optimization, and the method comprises the steps: carrying out the value adjustment according to a single dimension, so as to find the single dimension which can remarkably improve the influence degree; and combining the dimensions with small weights in pairs so as to find a dimension combination capable of remarkably improving the influence degree. After the single dimension capable of remarkably improving the influence degree is recognized, the value of the influence factor is adjusted in the dimension so as to find the key factor capable of remarkably improving the influence degree. And after the combination dimension capable of remarkably improving the influence degree is identified, selecting the influence factors with small weights in the combination dimension, and carrying out inter-dimension pairwise combination so as to find key factors capable of remarkably improving the influence degree. After key influence factors are identified, important reference data can be provided as health guidance public opinions.

Description

TECHNICAL FIELD

[0001] The application relates to a network public opinion key influence factor identification method based on influence degree tuning, and belongs to the technical field of network public opinion data analysis and mining. BACKGROUND

[0002] At present, network public opinion prevention and control mostly adopts a mode of "discovery first and intervention then": first, relying on big data and artificial intelligence to realize real-time monitoring and trend early warning of public opinion, and then combining administrative management, legal means, authoritative voice and positive guidance and other measures to deal with it. However, such methods often belong to post-position passive prevention and control, and it is difficult to timely curb the spread and diffusion in the initial stage of public opinion outbreak. SUMMARY

[0003] In order to solve the above problems, the application discloses a network public opinion key influence factor identification method based on influence degree tuning, and the specific technical scheme is as follows:

[0004] A network public opinion key influence factor identification method based on influence degree tuning comprises the following steps:

[0005] Step 1: influence degree tuning: the influence dimension of network public opinion evolution law is V i , the influence factor under each influence dimension is x ij , , i is the number of dimensions, a total of 5 dimensions, i takes the value of 1, 2, 3, 4 and 5, j is the number of influence factors under each influence dimension, and takes the value of 2, 3 and 4, the value of the influence factor under each influence dimension is integrated in a weighted average manner by applying a natural language processing technology, that is, the value of each influence dimension can be obtained, as shown in formula 1, (Formula 1) The values of the influence dimensions are integrated to obtain the influence degree D of the network public opinion evolution law, as shown in formula 2, (Formula 2) After calculation, each piece of public opinion information obtains one level of low, medium and high, and the number of times of occurrence of public opinions of different levels in different time periods is counted, that is, an evolution law graph is formed; If the value of the influence factor x ij is changed, the value of the influence degree D will be changed, so that the development direction of the evolution law is finally changed; Step 2: key influence factor identification: first, adjust the influence dimension V i , and then adjust the influence factor x ij under the corresponding influence dimension V i ; 2.1 First, adjust a single influence dimension V ithe value of the degree of influence D; 2.2 If the value of the degree of influence D is found to have a significant change, the weight of the influence factor under the single influence dimension is not greater than 50%, and the influence factor with a weight greater than 50% is not adjusted; 2.3 The influence factor adjustment adopts a two-by-two combination mode, that is, any two influence factors are selected, the first influence factor is fixed at a value, and the second influence factor is adjusted from 0 to 1 as a round of adjustment. After the end of this round of adjustment, the first influence factor is adjusted by a certain amount, and the second influence factor is adjusted from 0 to 1. The adjustment of the two influence factors selected is completed in turn, and if the value of the degree of influence D has a significant change, the two influence factors are simultaneously identified as key influence factors; 2.4 Replace the influence factor and repeat step 2.3 until all two-by-two combinations of the influence factors that can be adjusted under the single influence dimension are completed; 2.5 If the value of the degree of influence D does not have a significant change after the single influence dimension adjustment, a two-influence dimension combination adjustment mode is used; 2.6 Two influence dimension combination adjustment, if the value of the degree of influence D has a significant change, the influence factor with a weight not greater than 50% under the two influence dimensions is selected for adjustment, and the influence factor with a weight greater than 50% is not adjusted; 2.7 The influence factor adjustment adopts a two-by-two combination adjustment mode of any selected influence factor under the two influence dimensions, and the adjustment and identification of key influence factors are performed according to steps 2.3 and 2.4; 2.8 Repeat step 2.7 until all two-by-two combinations of the influence factors under the two influence dimensions are completed; The standard for judging whether the value of the degree of influence D has a significant change is whether the degree of influence grade changes; Step 3: After identifying the key influence factors, corresponding measures are taken according to the characteristics of different key influence factors, which can play a role in public opinion emergency prevention and control.

[0006] Further, the dimensions include five, namely, objectivity, timeliness, spread, emotional tendency, and standardization; The influence factors corresponding to each dimension are as follows: the influence factors of objectivity include the proportion of fuzzy words and the proportion of objective sentences; The influence factors of timeliness include the publishing frequency and the freshness; The influence factors of spread include the number of likes, the number of follow-up posts, and the spread rate; The influence factors of emotional tendency include the emotional intensity and the emotional consistency; The influence factors of standardization include the organization, the readability, the similarity, and the proportion of sensitive words.

[0007] Further, the influence factor value improvement method is as follows: Fuzzy word proportion: reduce the use of ambiguous words, Objective sentence proportion: use declarative sentences, Post frequency: reduce the interval between posts, Freshness: shorten the time interval between the occurrence and the release of public opinion information, Like amount: increase the number of likes, Follow-up amount: increase the number of follow-up replies, Transmission rate: increase the number of forwarding platforms, Emotional intensity: clearly express emotional stance, Emotional consistency: the emotional stance contained in the public opinion information remains consistent, Orderliness: use logical conjunctions to make the structure of the released public opinion information complete, hierarchical, and the theme clear, and reduce similar content, Readability: smooth text organization, Similarity: reduce the repetition of the content of the published public opinion information, Sensitive word proportion: reduce sensitive words.

[0008] Further, the process of obtaining the quantitative value of the dimension is: 1.1: Determine the weight of each dimension: give the weight value to the weight of the influencing factor, and the sum of the weight values of all influencing factors under each dimension is 1; 1.2: Quantitative calculation of the quantitative value of each dimension: according to the weight value of the influencing factor of each dimension, the weighted average is carried out according to the weight value of the influencing factor, and the quantitative value of each dimension is obtained.

[0009] Further, the process of obtaining the quantitative value of the influence degree is: S1: For the quantitative value of each influencing dimension, use the analytic hierarchy process to determine the weight of different influencing dimensions: S1.1: Establish a hierarchical structure model: network public opinion evolution rule influence degree is the target layer, and 5 influencing dimensions are the criterion layer, S1.2: Construct a judgment matrix: use the 1-9 scale method to compare the influencing dimensions of the criterion layer two by two to form a judgment matrix, S1.3: Consistency check: calculate the eigenvector and the maximum eigenvalue of the judgment matrix, and perform consistency check, and after passing the test, the weight vector w is finally obtained as the weight of each influencing dimension , Step S2: Integrate the quantitative values of each influencing dimension, use the membership function to construct a fuzzy relationship matrix, each influencing dimension of the public opinion information will get three membership degrees, and finally form a 5-row 3-column fuzzy relationship matrix U, as shown below, .

[0010] Further, the process of obtaining the influence degree level is: With the five influence dimensions of the influence degree as the evaluation influence factors, denoted as Wherein, is the quantitative value of each influence dimension, The level set of the influence degree is [low, medium, high], and the influence degree value of each public opinion information finally corresponds to a level in the level set; The fuzzy relationship matrix U is multiplied by the eigenvector w to obtain the fuzzy comprehensive evaluation set S: Wherein, is a value, , , Corresponding to low, medium and high three influence degree levels, according to the rules of fuzzy comprehensive evaluation, the maximum value in the three values is the influence degree, and the level corresponding to the maximum value is the influence degree level.

[0011] The beneficial effects of the present application are:

[0012] The influence degree of public opinion information is modeled as a comprehensive quantity determined by multiple influence factors in the present application, and the influence degree is dynamically optimized by adjusting the values of the influence factors, which directly acts on the public opinion propagation dynamics. Through calculation and verification, the adjustment of different influence factors will lead to significant changes in the evolution law of public opinion; when such changes significantly reduce the number of potential public opinion releases or the propagation heat, the optimization strategy can be used as the main basis for public opinion prevention and control. Unlike the traditional passive prevention mode, the present application can actively identify and optimize key influence factors, realize the transformation from "passive discovery" to "active intervention", and achieve precise containment and effective prevention and control before or at the early stage of public opinion outbreak. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 is a network public opinion evolution law diagram based on influence degree distribution statistics of the present application, Figure 2 is an evolution law diagram before and after the influence degree optimization of the present application, Figure 3 is a key factor identification scheme diagram of the present application, Figure 4 is a comparison diagram before and after the propagation degree dimension adjustment in the embodiment of the present application, Figure 5 is a comparison diagram before and after the propagation degree influence factor adjustment in the embodiment of the present application. DETAILED DESCRIPTION

[0014] The present application will be further clarified by the following description and examples with reference to the accompanying drawings. It should be understood that the following description and examples are merely illustrative of the present application and are not intended to limit the scope of the present application.

[0015] The data preparation underlying the present application is the quantitative calculation of the influence degree. The process is as follows: natural language processing and mathematical statistics methods are used to calculate the quantitative values of each influencing factor, in order to eliminate the differences in the values of the influencing factors between different methods. The values of the influencing factors are standardized to make the value range of each influencing factor fall within the interval [0-1], and the value is directly proportional to the influence degree, so as to unify the quantitative influencing factors and the influence degree of public opinion information. Secondly, the weights of different influencing factors under the same influencing dimension are determined by fuzzy comprehensive evaluation method, and the quantitative values of each influencing dimension are calculated. For each influencing dimension value, the weights of different influencing dimensions are determined by analytic hierarchy process, and the specific numerical value of the network public opinion evolution rule influence degree is calculated by combining the fuzzy comprehensive evaluation method, the influence degree is divided into grades, and the grade evaluation of the network public opinion evolution rule influence degree is obtained according to the corresponding relationship between the specific numerical value of the influence degree and the grades of the influence degree. The specific execution process of each step is described below:

[0016] Step 1: Construct the network public opinion evolution rule influence degree evaluation framework: including 5 influencing dimensions and 13 influencing factors, in turn: Influencing dimension one: objectivity: including influencing factors fuzzy word proportion and objective sentence proportion, Influencing dimension two: timeliness: including influencing factors release frequency and freshness, Influencing dimension three: propagation degree: including influencing factors like like amount, follow-up amount and propagation rate, Influencing dimension four: sentiment tendency: including influencing factors sentiment intensity and sentiment consistency, Influencing dimension five: standardization: including influencing factors orderliness, readability, similarity and sensitive word proportion.

[0017] Step 2: Use natural language processing and mathematical statistics methods to calculate the quantitative values of each influencing factor, and standardize the values of each influencing factor to make the quantitative values of each influencing factor fall within the interval [0-1], and the quantitative values of the influencing factors are directly proportional to the influence degree.

[0018] (1) Objectivity

[0019] Objectivity refers to the measurement of the objectivity of the content of network public opinion information in word usage. The higher the objectivity of a piece of public opinion information, the higher the credibility, and thus the greater the influence on the evolution rule of public opinion. The influencing factors are fuzzy word proportion and objective word proportion.

[0020] ① The proportion of vague terms refers to the percentage of vague pronouns such as "it is said," "I heard," and "insider information" in public opinion information. To standardize vague pronouns, the research project constructed a vague pronoun database. Using this database, the proportion of vague pronouns in the total length of current public opinion was statistically analyzed. Since the more such words appear, the higher the proportion, and the lower the veracity of the events involved, the reciprocal of this ratio was taken as the final quantitative representation of the proportion of vague terms.

[0021] ② The proportion of objective sentences refers to the percentage of objective sentences in public opinion information. Subjective sentences are those with obvious emotions and subjective judgments. Objective sentences, on the other hand, are generally declarative sentences that objectively describe events without obvious emotional coloring. Because objective sentences are more credible than subjective sentences, the more objective sentences there are, the clearer the event description, and the greater the impact on the evolution of public opinion. Therefore, this study uses the proportion of objective sentences in public opinion information as the quantitative value of this influencing factor, that is, the percentage of objective sentences in public opinion information out of the total number of sentences in the public opinion information. Objective sentences generally use declarative statements, and whether a sentence is a declarative statement can be determined using the Natural Language Processing (NLTK) package.

[0022] (2) Timeliness

[0023] Timeliness refers to the measurement of the time interval between the release of public opinion information. Generally speaking, the shorter the interval, the greater the impact on the evolutionary pattern. Sub-factors include: release frequency and freshness.

[0024] ① Release frequency refers to the relative rate at which public opinion information is released. It is calculated by comparing the interval between two public opinion messages that are adjacent to the current public opinion message in terms of release time. The calculation rule is as follows: Select two public opinion messages that are adjacent to the current public opinion message in terms of release time. Calculate the average of the time interval between the current public opinion message and these two public opinion messages in minutes, where the absolute value of the interval is taken. After obtaining the average value, take its reciprocal as the release frequency value, so that the smaller the interval, the greater the impact.

[0025] ② Freshness refers to the absolute rate at which public opinion information is released, calculated and compared with the release interval of the first piece of public opinion information. Also measured in minutes, the reciprocal of the time interval between the current public opinion and the first piece of public opinion is taken as the freshness value. The freshness of the first piece of public opinion is considered to be 1.

[0026] If the intervals mentioned above are all within the same minute, the interval is considered to be 1.

[0027] (3) Spread

[0028] Dissemination rate is a measure of the speed at which public opinion spreads; the higher the dissemination rate, the greater its impact on the evolutionary pattern. Sub-factors include: number of likes, number of comments, and dissemination rate.

[0029] ① The number of likes refers to the number of times users like public opinion information, and its value is a natural number.

[0030] ② The number of comments refers to the number of replies to the current public opinion information, and its value is a natural number.

[0031] ③ The propagation rate refers to the time interval between the spread of information from one online media to another. The calculation rule is as follows: calculate the interval between the time when the current public opinion information is published on the current platform and the time when that related information first appears on another platform. Also measured in minutes, the reciprocal of its absolute value is taken as the propagation rate.

[0032] The number of likes and comments may be relatively large, so for the same set of public opinion, they are normalized to convert their values ​​to [0-1], so as to keep them consistent with the value range of other influencing factors.

[0033] (4) Emotional Tendency

[0034] Sentiment bias is a measure of user emotions contained in public opinion information. The more pronounced the user emotions, the greater their influence on the evolution of public opinion. Sub-factors include: sentiment intensity and sentiment consistency.

[0035] ① Sentiment intensity refers to the degree of positive, negative, or neutral emotion expressed by users towards public opinion. The sentiment analysis tool TextBlob can be used to calculate sentiment intensity, which has a sentiment polarity value ranging from -1 to 1, where -1 represents extremely negative, 0 represents neutral, and 1 represents extremely positive. Here, we focus on intensity, so the absolute value is taken as the sentiment intensity value.

[0036] ② Emotional consistency refers to whether there is inconsistency in the emotions expressed within the content of public opinion. The calculation rule is as follows: Using TextBlob, the emotional polarity of each sentence is calculated, and the extreme values ​​are summed and the absolute value is taken. Since the summed value may exceed 1, the result is normalized to a range of [0-1]. 0 indicates completely opposite sentiments, i.e., complete inconsistency. 1 indicates that one emotion completely dominates, i.e., complete consistency.

[0037] (5) Normative

[0038] Standardization refers to the degree of standardization in the organization and expression of public opinion information. The higher the standardization, the greater the impact on the evolution of public opinion. Sub-factors include: coherence, readability, similarity, and the proportion of sensitive words.

[0039] ① Organizationality is a measure of whether public opinion information is logically organized in its text. Organizationality can be comprehensively evaluated from four aspects: logical connector density, semantic coherence analysis, topic concentration, and structural integrity. Logical connector density refers to the proportion of logical connectors such as "firstly," "secondly," "therefore," "however," and "for example" in the public opinion information. Semantic coherence analysis uses BERT to calculate the average similarity between adjacent sentences in the public opinion information. Topic concentration uses TextRank to extract keywords and calculates the topic distribution entropy as the concentration. Structural integrity checks for the existence of a general-specific-general structure; a complete structure scores 1 point, a partial structure scores 0.5 points, and no structure scores 0 points. The final organizationality score = 0.3 * logical connector density + 0.4 * semantic coherence + 0.2 * topic concentration + 0.1 * structural score.

[0040] ②Readability refers to the fluency of public opinion content. This study uses the readability method of the natural language processing tool cntext to calculate it directly, with a value ranging from 0 to 1. A higher value indicates greater readability.

[0041] ③ Similarity refers to the degree of similarity between a single piece of public opinion content and other public opinion information in the public opinion set. This project uses cosine similarity based on TF / IDF as the basic algorithm. The current public opinion information is compared pairwise with other public opinion information in the set, and the median is taken as the initial similarity. Since a higher similarity has a smaller impact on the pattern, the initial similarity is subtracted from 1 to obtain the final similarity.

[0042] ④ The percentage of sensitive words refers to the proportion of politically sensitive, stability-related, abusive, or personally attacking words in a given post. The calculation method for the percentage of sensitive words is similar to that for the percentage of vague words; it also requires constructing a sensitive word database and calculating the initial percentage of words in the database within the post. Since the more words in the database, the less compliant it is with current Chinese online information dissemination standards, the final percentage of sensitive words is obtained by subtracting the initial percentage from 1. A higher value indicates greater compliance and a stronger influence on the evolution of public opinion.

[0043] Applying the above calculation rules, the value of each influencing factor is a real number between 0 and 1, and the larger the value, the greater the impact on the evolution of public opinion.

[0044] The execution code for the above-mentioned influencing factors can be found in the code given in the specific implementation of the patent application filed by the inventor's team earlier (patent number 202411689716.7).

[0045] Step 3: Determine the weight values ​​of different influencing factors under the same influence dimension, and calculate the quantitative value of each influence dimension accordingly.

[0046] (1) Determine the weights of the influencing factors for each influencing dimension

[0047] This invention determines the weights of each influencing factor under different influence dimensions, as shown in Table 1. The weights shown in Table 1 are only examples.

[0048] Table 1 Weights of each influencing factor

[0049] Influence dimension Influence factor weight Objectivity Fuzzy word proportion (59.5%), objective sentence proportion (40.5%). Timeliness Release frequency (22.7%), freshness (77.3%). Propagation degree Like amount (26.24%), follow-up amount (46.16%), propagation rate (27.6%). Sentiment tendency Emotional intensity (74.95%), emotional consistency (25.05%). Normativity Organization (37.45%), readability (16.95%), similarity (16.52%), sensitive word proportion (29.08%).

[0050] (2) Calculation of quantified values ​​of each influencing dimension

[0051] Based on the weights shown in Table 1, the influencing factor values ​​are weighted and averaged according to the influencing dimension to obtain the quantitative value of each influencing dimension.

[0052] Step 4: For the quantitative values ​​of each influence dimension, determine the weight of different influence dimensions, and then use the fuzzy comprehensive evaluation method to calculate the specific numerical value of the influence of the evolution law of online public opinion.

[0053] 4.1 Calculate the weights of each influencing dimension using the AHP method. The steps are as follows:

[0054] ① Establish a hierarchical structure model: The influence of the evolution of online public opinion is the target layer, and the five influence dimensions are the criteria layer.

[0055] ② Constructing the judgment matrix: The influence dimensions of the criterion layer are compared pairwise using the 1-9 scaling method to form the judgment matrix, as shown in Table 2.

[0056] Table 2 Importance Judgment Matrix

[0057] Influence dimension Objectivity Timeliness Propagation degree Sentiment tendency Normativity Objectivity 1 1 / 3 1 / 5 3 5 Timeliness 3 1 1 / 3 5 7 Propagation degree 5 3 1 7 9 Sentiment tendency 1 / 3 1 / 5 1 / 7 1 3 Normativity 1 / 5 1 / 7 1 / 9 1 / 3 1

[0058] ③ Consistency check: Calculate the eigenvectors and the largest eigenvalue of the matrix, perform a consistency check, and after passing the check, obtain the final weight vector w, which is used as the weight of each influence dimension.

[0059]

[0060] 4.2 Constructing the Fuzzy Relation Matrix

[0061] The fuzzy comprehensive evaluation method uses membership functions to construct a fuzzy relation matrix, representing the degree of conformity of each influence dimension to the influence level. This invention selects the trapezoidal function as the membership function, with the specific formula as follows:

[0062] When the level of influence is [low], ,

[0063] When the level of influence is [Medium], ,

[0064] When the level of influence is [high], ,

[0065] Where x represents each influencing dimension V i The value of the parameter is determined by the function, while the parameters a, b, c, and d are used to divide the optimal interval. That is, by adjusting the parameter values, the influence dimension values ​​are made to fall within the range of [0-1]. This is also the important reason why this paper chooses [0-1] as the range of influence dimension values. The adjustment of this parameter also needs to be determined based on the actual influence value. According to this function, each influence dimension of public opinion information will obtain three membership degrees, ultimately forming a 5-row, 3-column fuzzy relation matrix U, as shown in the following formula.

[0066]

[0067] Step 5: Classify the level of influence. Based on the correspondence between the specific numerical value of the influence and the level of influence, obtain the level of influence.

[0068] 5.1 Determine the influencing factors and their levels

[0069] The five dimensions of influence of regularity, namely the influencing factors, are denoted as follows: .in This refers to the quantitative value of the influence dimension. The set of evaluation levels for the influence degree of the pattern is then denoted as T = [low, medium, high], representing the magnitude of the influence on the pattern of public opinion evolution. Through calculation, the influence value of each piece of public opinion information will ultimately correspond to a certain level in the evaluation set. The influence value of public opinion information is calculated from the fuzzy relation matrix and weights of the influencing factors.

[0070] 5.2 Calculate the fuzzy comprehensive evaluation set to evaluate the impact.

[0071] Multiplying the fuzzy relation matrix U by the eigenvector w yields the fuzzy comprehensive evaluation set S:

[0072]

[0073] in, For a value, , , These correspond to three levels of influence: low, medium, and high. According to the rules of fuzzy comprehensive evaluation, the maximum value among these three values ​​is the influence level, and the level corresponding to the maximum value is the influence level.

[0074] This invention further proposes to identify key influencing factors affecting the evolution of public opinion by optimizing the impact level.

[0075] In practical applications, this invention can formulate relevant strategies for key influencing factors, guide the healthy development of public opinion, and realize the transformation from "passive discovery" to "proactive intervention," enabling public opinion to be accurately located and effectively controlled before or in its early stages.

[0076] The main steps in the practical application of this invention are as follows:

[0077] (1) Obtain public opinion information data.

[0078] (2) First adjust the values ​​according to a single dimension to find the single dimension that can significantly improve the influence.

[0079] (3) Combine dimensions with small weights in pairs to find dimension combinations that can significantly improve the influence.

[0080] (4) After identifying a single dimension that can significantly improve the impact, adjust the value of the influencing factors within that dimension to discover the key factors that can significantly improve the impact.

[0081] (5) After identifying the combined dimensions that can significantly improve the influence, select the influencing factors with small weights in the combined dimensions and combine them in pairs across dimensions to find the key factors that can significantly improve the influence.

[0082] (6) After identifying key influencing factors, apply the key influencing factors to guide the development of public opinion in a healthy direction.

[0083] The specific execution process of this invention is as follows:

[0084] 1. Optimization of Online Public Opinion Influence Based on Influencing Factors

[0085] Let V be the dimension of influence of the evolution of online public opinion. i , Each influencing dimension has x as its influencing factor. ij , , In this equation, i represents the number of dimensions (5 in total), with values ​​of 1, 2, 3, 4, and 5. j represents the number of influencing factors under each dimension, with values ​​of 2, 3, and 4. By applying natural language processing techniques and integrating the values ​​of influencing factors under each dimension using a weighted average, the values ​​of each dimension can be obtained, as shown in Equation 1.

[0086] (Equation 1)

[0087] By integrating the values ​​of various influencing dimensions, the influence degree D of the evolution pattern of online public opinion is obtained, as shown in Equation 2.

[0088] (Equation 2)

[0089] After calculation, each piece of public opinion information was assigned a low, medium, or high level. The frequency of occurrence of each level of public opinion information in different time periods was then counted, forming a [database structure]. Figure 1 The evolutionary pattern shown.

[0090] If the influencing factor x is changed i The value of will inevitably change the value of influence D, thus ultimately altering the direction of evolutionary development, such as... Figure 2 As shown.

[0091] As shown in the figure, the purpose of optimization is to expedite the dissipation phase of public opinion. This invention, through extensive data analysis, reveals that high-impact public opinion significantly drives its evolution; therefore, optimizing impact primarily aims to enhance the influence of public opinion information. The higher the proportion of high-impact public opinion, the smaller the impact of medium- and low-impact public opinion. Improving the influence of public opinion information mainly involves identifying key influencing factors that significantly enhance its impact.

[0092] 2. Identification of key influencing factors

[0093] Influence optimization aims to identify key influencing factors that can significantly improve influence. Since influence is determined by the values ​​of each dimension and its subordinate factors, adjusting factor values ​​can change dimension values, thereby affecting the overall influence. However, adjusting factors individually may have limited effect; adjusting dimension values ​​first, observing changes in influence, and then adjusting influencing factors is more efficient. Based on this, this invention proposes a key factor identification method of "adjusting dimensions first, then factors," the identification scheme of which is described below. Figure 3 .

[0094] As shown in the figure, the values ​​of individual dimensions are first adjusted sequentially from 0 to 1. If a significant improvement is observed, the influencing factors under that dimension are then adjusted from 0 to 1, but the influencing factor with the highest weight is not adjusted, unless its weight is still less than 40%. Factor adjustments are performed in pairs. If a significant improvement is observed, both factors are identified as key factors. If a single-dimensional adjustment does not reveal a significant improvement, a two-dimensional combination adjustment is used. If a significant improvement is observed, influencing factors under different dimensions are selected and combined in pairs according to the above rules to identify the key factors. The criterion for judging a significant change is whether a change in the degree of influence occurs.

[0095] 3. Emergency Prevention and Control Methods for Online Public Opinion Based on Key Influencing Factors

[0096] After identifying key influencing factors, writing and disseminating public opinion information based on the characteristics of different key factors can, in theory, serve as a means of emergency public opinion control. This invention identified 5 dimensions and 13 influencing factors through research, and the general methods for improving the values ​​of each influencing factor are shown in Table 1.

[0097] Table 1. Methods for improving the impact factor values ​​of each dimension.

[0098]

[0099] When time permits, increasing the values ​​of all influencing factors can naturally produce high-impact public opinion information. Otherwise, the decision must be made based on the type, urgency, and characteristics of the public opinion. Different types of public opinion may only require adjusting a few influencing factors to achieve good results in a short period of time. Therefore, historical public opinion data should be regularly collected and analyzed to identify key influencing factors, so that when new public opinion emerges, key influencing factors can be quickly matched and located, and targeted guidance can be provided for the release of public opinion to achieve rapid resolution of the public opinion crisis.

[0100] The following is a specific application example of the present invention:

[0101] This invention extracted public opinion information on the "American Chinese restaurant conflict incident" from online media such as Sina Weibo, Toutiao, and NetEase. This information was publicly available and used as research material, without infringing on any legal rights of others. After sorting, a total of 5,000 pieces of public opinion information were obtained. The structure of each piece of information is {content, source website, date, media type, author}, mainly text. After statistical analysis, the impact of all 5,000 pieces of information was calculated, and different levels of public opinion information were statistically analyzed. The final results are as follows: low-impact public opinion information accounts for approximately 26.7%, medium-impact public opinion information accounts for approximately 64.2%, and low-impact public opinion information accounts for approximately 9.2%, showing a distribution pattern of small at both ends and large in the middle, which is close to a normal distribution and conforms to the general law of the development of things.

[0102] When applying the method of this invention to optimize influence, there is no room for further improvement in public opinion information that already has high influence. Therefore, whether adjusting dimensions or factors, it is only necessary to target low and medium influence. Taking dissemination as the adjustment target, three public opinion messages were randomly selected for adjustment, and the results are as follows. Figure 4 As shown.

[0103] As shown in the figure, the adjustment is effective for low impact. After the adjustment, the impact of two out of the three factors increased from low to high. The impact of medium factors, however, did not change significantly. Although one factor changed, it started low, moved to medium, and then back to low; since it was already a medium impact factor, this cannot be considered a significant change. This adjustment demonstrates that adjusting the dissemination rate will cause a significant change in the impact of public opinion information. The dissemination rate value is derived from the weighted average of its subordinate influencing factors, which include likes (26.24%), comments (46.16%), and dissemination rate (27.6%). Comments have a high weight of 46.16%, significantly influencing this dimension and can be directly considered a key factor. The adjustment here targets the combination of likes and dissemination rate, and the result is shown below.Figure 5 As shown.

[0104] like Figure 5 When both the number of likes and the speed of dissemination increase simultaneously, two pieces of public opinion information with low impact show significant changes: one's impact increases from low to high, and the other from medium to high. Therefore, the number of likes and the speed of dissemination can be considered key influencing factors; their combination can significantly change the impact of public opinion information. However, this requires the number of likes to increase to at least approximately 0.6 for this combination to effectively enhance the impact of the public opinion information.

[0105] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0106] Based on the above-described preferred embodiments of the present invention, and through the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention.

Claims

1. A method for identifying key influencing factors of online public opinion based on influence optimization, characterized in that, Includes the following steps: Step 1: Influence Optimization: Let V be the influence dimension of the evolution pattern of online public opinion. i Each influencing dimension has x as its influencing factor. ij , , In this equation, i represents the number of dimensions (5 in total), with values ​​of 1, 2, 3, 4, and 5. j represents the number of influencing factors under each dimension, with values ​​of 2, 3, and 4. By applying natural language processing techniques and integrating the values ​​of influencing factors under each dimension using a weighted average, the values ​​of each dimension can be obtained, as shown in Equation 1. (Equation 1) By integrating the values ​​of various influencing dimensions, the influence degree D of the evolution pattern of online public opinion is obtained, as shown in Equation 2. (Equation 2) After calculation, each piece of public opinion information was assigned a level of low, medium, or high. The number of times different levels of public opinion appeared in different time periods was counted to form an evolution pattern diagram. If the influencing factor x is changed ij The value of will inevitably change the value of influence D, thereby ultimately changing the direction of evolutionary development. Step 2: Identification of key influencing factors: First adjust the influencing dimension V i Then adjust the corresponding influence dimension V. i The following influencing factors x ij ; Step 2.1 First, adjust the individual influence dimensions V sequentially from 0 to 1. i The value; Step 2.2 If a significant change is found in the value of influence D, then the influence factors with a weight of no more than 50% under that single influence dimension will be adjusted, while the influence factors with a weight of more than 50% will not be adjusted. Step 2.3 The adjustment of influencing factors adopts a pairwise combination method, that is, select any two influencing factors, fix the first influencing factor at a fixed value, and adjust the second influencing factor from 0 to 1 as one round of adjustment. After this round of adjustment, the first influencing factor is adjusted upward by a certain amount, and the second influencing factor is adjusted from 0 to 1 again. The adjustment of the two currently selected influencing factors is completed in turn. If the value of the influence degree D changes significantly, the two influencing factors are identified as key influencing factors. Step 2.4 Change the influencing factor and repeat step 2.3 until all pairwise combinations of the influencing factors that can be adjusted under this single influencing dimension are completed; Step 2.5 If adjusting a single influence dimension does not significantly change the value of influence D, then use a combination of two influence dimensions for adjustment. Step 2.6 Adjust the combination of the two influence dimensions. If the value of influence D changes significantly, select the influence factors with a weight of no more than 50% under the two influence dimensions for adjustment. Influence factors with a weight of more than 50% will not be included in the adjustment. Step 2.7 Adjusting influencing factors involves selecting one influencing factor from each of the two influencing dimensions and combining them for adjustment. After combination, the key influencing factors are adjusted and identified according to the methods in Steps 2.3 and 2.

4. Step 2.8 Repeat step 2.7 until all pairwise combinations of the influencing factors under the two influence dimensions of the current combination are completed; The criterion for judging whether the value of influence degree D has changed significantly is whether the influence degree level has changed. Step 3: After identifying the key influencing factors, take corresponding measures based on the characteristics of different key influencing factors to play a role in public opinion emergency prevention and control.

2. The method for identifying key influencing factors of online public opinion based on influence degree optimization according to claim 1, characterized in that, The dimensions are five in total: objectivity, timeliness, dissemination, emotional bias, and standardization. The influencing factors for each dimension are as follows: the influencing factors for objectivity include the proportion of vague words and the proportion of objective sentences; Factors affecting timeliness include publication frequency and freshness; Factors influencing the spread of information include the number of likes, the number of comments, and the rate of dissemination. Factors influencing affective tendency include affective intensity and affective consistency; Factors influencing standardization include logical structure, readability, similarity, and the proportion of sensitive words.

3. The method for identifying key influencing factors of online public opinion based on influence optimization according to claim 2, characterized in that, The methods for increasing the values ​​of the influencing factors are as follows: Percentage of vague words: Reduce the use of vague words. Objective sentence percentage: Declarative sentences were used. Release frequency: Reduce release intervals. Freshness: Shorten the time interval between the occurrence of an event and the release of public opinion information. Likes: Increase the number of likes. Number of comments: Increase the number of comments and replies. Propagation rate: Increase the number of forwarding platforms. Emotional intensity: Clearly express emotional stance, Emotional consistency: The emotional stance contained in public opinion information remains consistent. Organization: Use logical connectors to ensure the released public opinion information has a complete structure, clear hierarchy, and a clear theme, while reducing similar content. Readability: The text is well-organized and flows smoothly. Similarity: Reduce duplication with already published public opinion information. Sensitive word percentage: Reduce sensitive words.

4. The method for identifying key influencing factors of online public opinion based on influence optimization according to claim 1, characterized in that, The process of obtaining the quantitative value of the dimension is as follows: Determine the weights of influencing factors for each dimension: Assign weight values ​​to the influencing factors, and the sum of the weight values ​​of all influencing factors under each dimension is 1; Quantitative calculation of each dimension: Based on the weight values ​​of the influencing factors of each dimension, a weighted average is calculated to obtain the quantitative value of each dimension.

5. The method for identifying key influencing factors of online public opinion based on influence degree optimization according to claim 1, characterized in that, The process for obtaining the quantitative value of the influence is as follows: Step S1: For the quantitative values ​​of each influencing dimension, use the analytic hierarchy process (AHP) to determine the weights of different influencing dimensions: Step S1.1: Establish a hierarchical model: the influence of the evolution of online public opinion is the target layer, and the five influence dimensions are the criteria layer. Step S1.2: Construct the judgment matrix: Use the 1-9 scaling method to perform pairwise comparisons of the influence dimensions of the criterion layer to form a judgment matrix. Step S1.3: Consistency Check: Calculate the eigenvectors and the largest eigenvalue of the judgment matrix, perform a consistency check, and if the check passes, obtain the final weight vector w, which will be used as the weights for each influence dimension. , Step S2: Integrate the quantitative values ​​of each influence dimension, and use the membership function to construct a fuzzy relation matrix. Each influence dimension of public opinion information will obtain three membership degrees, ultimately forming a 5x3 fuzzy relation matrix U, as shown below. 。 6. The method for identifying key influencing factors of online public opinion based on influence optimization according to claim 1, characterized in that, The process of obtaining the influence level is as follows: The five dimensions of influence are used to evaluate the influencing factors, denoted as: ,in, These are quantitative values ​​for each influencing dimension. Then, the influence level set is recorded as [low, medium, high], and the influence value of each piece of public opinion information will eventually correspond to a certain level in the level set; Multiplying the fuzzy relation matrix U by the eigenvector w yields the fuzzy comprehensive evaluation set S: in, For a value, , , These correspond to three levels of influence: low, medium, and high. According to the rules of fuzzy comprehensive evaluation, the maximum value among these three values ​​is the influence level, and the level corresponding to the maximum value is the level of influence.

Images