Social risk identification and prediction method and system based on artificial intelligence

Abstract

The invention discloses a social risk identification and prediction method and system based on artificial intelligence, and belongs to the technical field of artificial intelligence. The method comprises the following steps: preprocessing discrete information of network public opinions to obtain a plurality of words, calculating the similarity between each word and a risk emotion intention label in a risk word bank, and classifying the discrete information of the internet according to a comparison result of the plurality of words and a threshold value; calculating the amount of information of a plurality of words of discrete information of each type of network public opinions in each time period, adding the amount of information in each time period to obtain the amount of information in each time period, adding all the amount of information in each time period before the last moment of each time period to obtain a time sequence of an amount of information accumulated value, and calculating the amount of information in each time period; the Logistic regression model is trained into a risk prediction model, and the risk prediction model is utilized to generate the social risk trend in a period of time in the future. According to the method, wide social public opinion information can be covered, appeals and emotional changes of the public can be accurately captured, and the complexity and diversity of social risks can be dealt with.

Description

Technical Field

[0001] The present invention relates to a method and system for identifying and predicting social risks based on artificial intelligence, and belongs to the technical field of artificial intelligence. Background Art

[0002] Existing technologies primarily rely on traditional statistical methods, expert assessments, and specific risk indicator systems to identify and predict social risks. These methods are typically based on limited data samples and pre-set risk indicators, making it difficult to fully and dynamically reflect the complexity and diversity of social risks. Specific issues include:

[0003] Single data source: Traditional methods mainly rely on government statistics, expert opinions, etc. The data sources are limited and it is difficult to cover a wide range of social public opinion information.

[0004] Limited recognition accuracy: Due to the lack of in-depth mining and analysis of public opinion big data, existing methods have low accuracy in identifying social risks and find it difficult to accurately capture the demands and emotional changes of the public.

[0005] Poor prediction results: Existing methods mostly conduct short-term static predictions, which are difficult to cope with the complexity and diversity of social risks and have poor prediction results. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the invention of the present invention provides a method and system for social risk identification and prediction based on artificial intelligence, which can cover a wide range of social public opinion information, accurately capture the demands and emotional changes of the people, cope with the complexity and diversity of social risks, and have good prediction effects.

[0007] To achieve the above-mentioned purpose, the present invention provides a method for identifying and predicting social risks based on artificial intelligence, which comprises the following steps: S1: Preprocess the discrete information of online public opinion to obtain multiple words, calculate the similarity between each word and the risk sentiment intention label in the risk vocabulary according to the recognition rules, and compare the similarity with the threshold. Classify the discrete information on the Internet based on the comparison results of multiple words and the threshold; S2: Divide the previous set time into multiple time periods of equal length, calculate the information volume of multiple words of discrete information of each type of network public opinion in each time period, add up the information volume of each time period to obtain the information volume of each time period, and add up all the information volumes of each time period before the last moment of each time period to obtain the time series of the accumulated information volume. ,use The Logistic regression model is trained into a risk prediction model, and the risk prediction model is used to generate social risk trends in the future.

[0008] To achieve the aforementioned purpose of the invention, the present invention also provides a system comprising a storage medium and one or more processors, wherein the storage medium is used to store code for compiling the aforementioned method of social risk identification and prediction based on artificial intelligence into a computer program using a computer language, and the computer program can be called and executed by one or more processors.

[0009] Compared with the existing technology, the method and system for social risk identification and prediction based on artificial intelligence provided by the present invention can cover a wide range of social public opinion information, accurately capture the demands and emotional changes of the people, cope with the complexity and diversity of social risks, and have good prediction effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flow chart of a method for social risk identification and prediction based on artificial intelligence provided by the present invention.

[0011] Figure 2 It is a schematic diagram of the time series of social risk information provided by the present invention.

[0012] Figure 3 This is a flowchart of constructing the emotional intention dictionary provided by the present invention.

[0013] Figure 4 A schematic diagram of emotional risk classification provided by the present invention.

[0014] Figure 5 This is a flow chart of the claim risk classification and identification based on the claim risk vocabulary provided by the present invention.

[0015] Figure 6 This is a flow chart of claim risk identification provided by the present invention. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0017] Figure 1 The present invention provides a flowchart of a method for identifying and predicting social risks based on artificial intelligence, such as Figure 1 As shown, the method for social risk identification and prediction based on artificial intelligence provided by the present invention includes the following steps: S1: Preprocess the discrete information of online public opinion to obtain multiple words, calculate the similarity between each word and the risk sentiment intention label in the risk vocabulary according to the recognition rules, and compare the similarity with the threshold. Classify the discrete information on the Internet according to the comparison results of the multiple words and the threshold. The recognition rules include word frequency restriction, weight setting, multi-word co-occurrence rules, and syntactic rules; S2: Divide the previous set time into multiple time periods of equal length, calculate the information volume of multiple words of discrete information of each type of network public opinion in each time period, add up the information volume of each time period to obtain the information volume of each time period, and add up all the information volumes of each time period before the last moment of each time period to obtain the time series of the accumulated information volume. ,use The Logistic regression model is trained into a risk prediction model, and the risk prediction model is used to generate social risk trends in the future.

[0018] Figure 2 This is a schematic diagram of the time series of social risk information provided by the present invention, such as Figure 2 As shown, the social risk information time series , is the sum of the information of multiple words in the i-th time period, , The pth word The amount of information, P is the number of words that the discrete information of a type of public opinion big data is divided into. For example ; .

[0019] In the present invention, the risk vocabulary includes multiple categories of emotional intent dictionaries, each category of emotional intent dictionaries includes multiple risk emotional intent labels, and the multiple categories of risk emotional intent labels are structured texts converted from unstructured text of risk emotional intent information in a set network medium. Preferably, emotional intent includes: emotion and appeal. The multiple categories of emotional intent dictionaries include at least an emotional risk vocabulary and an appeal risk vocabulary.

[0020] Figure 3 This is a flowchart of constructing the emotional intention dictionary provided by the present invention. Figure 3 As shown in the figure, the risk word library that reflects the public's emotional intentions includes: S1-1: De-duplicate and clean the risk emotion intention information in the set network media to obtain multiple non-repeating risk emotion intention information, and perform word segmentation on each risk emotion intention information; S1-2: Tag each word with part of speech, filter the words according to the tagged part of speech, and build a custom stop dictionary; S1-3: synthesizing the customized stop dictionary and the general stop dictionary to obtain a synthesized stop dictionary; S1-4: Remove the words in the synthetic stop dictionary from the words obtained in step 1-1 to obtain the emotional intent information words.

[0021] In the present invention, constructing a risk vocabulary that reflects public sentiment and intentions also includes: S1-5: Calculate the word frequency of risk emotional intention information in the set network media, and calculate its inverse document word frequency as its weight to obtain the domain basic dictionary; S1-6: Use the word vector model to expand the emotional intention information words to obtain the domain expansion dictionary; S1-7: Merge the domain basic dictionary and the domain extended dictionary to obtain the sentiment intention risk vocabulary.

[0022] In the present invention, constructing a risk vocabulary that reflects public sentiment and intentions also includes: S1-8: Based on the emotional intensity of the words and the emotional intensity of the text in the assumed online media, the words in the emotional risk vocabulary are divided into multiple categories. Each category is further divided into multiple subcategories, and each subcategory can be further divided. This provides a basis for predicting the social risk trends of the subdivided emotions.

[0023] Figure 4 The emotional risk classification diagram provided by the present invention is as follows: Figure 4 As shown, the multiple categories include satisfaction, fairness, confusion, fear and willingness to fight. The satisfaction category includes subcategories: happiness, respect, praise, love, and wish; the fairness category includes subcategories: peace of mind and trust; the confusion category includes subcategories: thinking, surprise, doubt, and boredom; the fear category includes subcategories: panic, shame, and fear; the willingness to fight category includes subcategories: sadness, disappointment, blame, anger, and hatred.

[0024] In the present invention, the sentiment intensity of a word is calculated according to the following formula: , in, For words emotional intensity; is the polarity of sentiment words. When the sentiment words are neutral, ,here It is not 0, because as long as there is emotion, it may affect risk perception. When the emotion word is positive, , when the sentiment word is negative, ; Adverbs of degree The weight of The emotional intensity annotated for the emotional vocabulary can be divided into nine levels; The calculation of sentiment intensity takes into account the polarity of sentiment words, the influence of degree adverbs, and the weight of sentence patterns, and can accurately reflect the sentiment tendency of the text.

[0025] The emotional intensity of text in network media is calculated according to the following formula: , Where, For words The sentiment intensity; m is the number of negative words, Negate the result for the negation word; Whether the negative word appears, 0.8 if it appears, and 1 if it does not appear; is the sentence weight, see Table 1 below:

[0026] Table 1 .

[0027] By constructing a sentiment-intention dictionary and appeal lexicon, the present invention extracts public sentiment and intention information from public opinion big data and identifies keywords related to social risks. By constructing a sentiment-intention lexicon, the unstructured text in public opinion big data can be converted into structured appeal labels. The present invention can train the Word2vec model into a word vector model, which can capture the semantic relationships between words, thereby expanding the vocabulary with similar semantics to those in the basic dictionary, forming an extended dictionary.

[0028] In the present invention, the Logistic regression model is: , Where, is the intrinsic growth rate, K is the adjustable coefficient; is the amount of emotional intention information about a social event at the initial moment, is the amount of social risk information predicted by the Logistic regression model.

[0029] In the present invention, the Logistic regression model training process includes: S2-1: Establish the loss function value according to the following formula: , Update K according to the following formula: , Where, is the adjustment coefficient, is the gradient of K; S2-2 updates r according to the following formula: , Where, is the adjustment coefficient, is the gradient of r; S2-3: Determine whether the loss function value L is the smallest. If so, output the optimal K and r respectively. and , and then go to step S2-4; if the loss function value, L does not reach the minimum, so that the changed K and r are replaced by the values ​​before the change, and then repeat steps S2-1 to S2-3; S2-4: Predict the social risk trend in the future based on the following formula: .

[0030] The present invention can predict risk trends in future time periods by training the Logistic regression model into a risk prediction model.

[0031] Figure 5 This is a flow chart of the claim risk classification and identification based on the claim risk vocabulary provided by the present invention, such as Figure 5 As shown, the classification and identification of claim risks based on the claim risk vocabulary includes: S3-1: De-duplicate and clean the discrete information on the Internet to obtain multiple non-repeating public opinion information, and perform word segmentation on each piece of public opinion information; S3-2: Remove the words obtained in step 3-1 from the words in the synthetic stop dictionary to obtain P discrete information emotion intention information words of the public opinion big data; S3-3: Match P discrete information emotion intention information words with words in the appeal vocabulary to obtain Q matching appeal information words, where Q is less than or equal to P, and the ratio of Q to P is the information proportion. Weighting the information proportion can obtain the appeal similarity; S3-4: Compare the similarity with the threshold. If the similarity is higher than the threshold, it is the appeal information; if the similarity is lower than the threshold, it is not the appeal information; S3-5: Classify the claim information using the claim risk domain dictionary. The classification result is the risk domain label corresponding to the claim information. Specifically, the word frequency of Q matching claim information words in the claim risk domain dictionary is calculated. The largest category among multiple categories in the claim risk vocabulary corresponding to the word frequency is used as the claim risk label. S3-6: Calculate the time series of the demand risk information based on the risk area label. Use the time series of the demand risk information to obtain the demand risk prediction model through steps S2-1 to S2-4, and predict the demand trend in the future time period based on the demand risk prediction model.

[0032] Figure 6 This is a flow chart of the emotional risk identification based on the emotional risk vocabulary provided by the present invention, such as Figure 6 As shown, the emotional risk identification based on the emotional risk vocabulary provided by the present invention includes: S4-1: De-duplicate and clean the discrete information on the Internet to obtain multiple non-repeating public opinion information, and perform word segmentation on each piece of public opinion information; S4-2: remove the words obtained in step 4-1 from the words in the synthetic stop dictionary to obtain R discrete information sentiment intention information words of the public opinion big data; S4-3: Match the R discrete information emotion intention information with the words in the emotion risk vocabulary to obtain S matching emotion information words, where S is less than or equal to R, calculate the emotion intensity of each word in the S matching emotion information words, and calculate the emotion intensity of the discrete information on the Internet based on the emotion intensity of each word; S4-4: Classify the emotional intensity of discrete information on the Internet using the emotional risk domain dictionary. The classification result is the emotional risk domain label corresponding to the emotional risk information in the discrete information on the Internet. S4-5: Calculate the time series of emotional risk information based on the emotional risk domain label. Use the time series of emotional risk information to obtain an emotional risk prediction model through steps S2-1 to S2-4, and predict the emotional trend of the future time period based on the emotional risk prediction model.

[0033] The present invention also provides a system comprising a storage medium and one or more processors, wherein the storage medium is used to store code for compiling the above-mentioned method of social risk identification and prediction based on artificial intelligence into a computer program using a computer language, and the computer program can be called and executed by one or more processors.

[0034] Although the present invention is described using emotions and appeals as examples, the present invention is not limited to these two types of intentions. It can identify various social public opinion information and make predictions based on the identification results.

[0035] The invention provides an artificial intelligence-based social risk identification and prediction method and system, which can cover a wide range of social public opinion information, accurately capture the demands and emotional changes of the people, cope with the complexity and diversity of social risks, and have good prediction effects.

[0036] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the contents of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can better understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A method for identifying and predicting social risks based on artificial intelligence, characterized in that: The steps include: S1: Preprocess the discrete information of online public opinion to obtain multiple words, calculate the similarity between each word and the risk sentiment intention label in the risk vocabulary according to the recognition rules, and compare the similarity with the threshold. Classify the discrete information on the Internet based on the comparison results of multiple words and the threshold; S2: Divide the previous set time into multiple time periods of equal length, calculate the information volume of multiple words of discrete information of each type of network public opinion in each time period, add up the information volume of each time period to obtain the information volume of each time period, and add up all the information volumes of each time period before the last moment of each time period to obtain the time series of the accumulated information volume. ,use The Logistic regression model is trained into a risk prediction model, and the risk prediction model is used to generate social risk trends in the future.

2. The method for identifying and predicting social risks based on artificial intelligence according to claim 1, characterized in that: The risk prediction model is obtained by training the Logistic regression model, which is: ， Where, is the intrinsic growth rate, K is the information limit; is the amount of emotional intention information about a social event at the initial moment, is the amount of social risk information predicted by the Logistic regression model.

3. The method for identifying and predicting social risks based on artificial intelligence according to claim 2, characterized in that: The Logistic regression model training process includes: S2-1: Establish the loss function value according to the following formula: ， Update K according to the following formula: ， Where, is the adjustment coefficient, is the gradient of K; S2-2: Update r according to the following formula: ， Where, is the adjustment coefficient, is the gradient of r; S2-3: Determine whether the loss function value L is the smallest. If so, output the optimal K and r respectively. and , and then execute step S2-4; if the loss function value L does not reach the minimum, the changed K and r are replaced by the values ​​before the change, and then repeat steps S2-1 to S2-3; S2-4: Predict the social risk trend in the future based on the following formula: 。 4. The method for identifying and predicting social risks based on artificial intelligence according to any one of claims 1 to 3, characterized in that: The construction of the risk vocabulary includes: S1-1: De-duplicate and clean the risk emotion intention information in the set network media to obtain multiple non-repeating risk emotion intention information, and perform word segmentation on each risk emotion intention information; S1-2: Tag each word with part of speech, filter the words according to the tagged part of speech, and build a custom stop dictionary; S1-3: synthesizing the customized stop dictionary and the general stop dictionary to obtain a synthesized stop dictionary; S1-4: Remove the words obtained in step 1-1 and synthesize the words contained in the stop dictionary to obtain the emotional intent information words.

5. The method for identifying and predicting social risks based on artificial intelligence according to claim 4, characterized in that: The construction of the risk vocabulary also includes: S1-5: Calculate the word frequency of risk emotional intention information in the set network media, and calculate its inverse document word frequency as its weight to obtain the domain basic word vector dictionary; S1-6: Use the word vector model to expand the emotional intention information words to obtain the domain expansion word vector dictionary; S1-7: Merge the domain basic word vector dictionary and the domain extended word vector dictionary to obtain the sentiment intent word vector library.

6. The method for identifying and predicting social risks based on artificial intelligence according to claim 5, characterized in that: Preprocessing the discrete information of the public opinion big data to obtain multiple words includes: deduplicating and cleaning the discrete information of the public opinion big data to obtain multiple non-repeating public opinion information, and segmenting each public opinion information.

7. The method for identifying and predicting social risks based on artificial intelligence according to claim 6, characterized in that: The recognition rules include word frequency restriction, weight setting, multi-word co-occurrence rules, and syntactic rules.

8. The method for identifying and predicting social risks based on artificial intelligence according to claim 7, characterized in that: Emotional intent includes: emotion and appeal; the multi-category emotional intent dictionary includes at least an emotional risk vocabulary and an appeal risk vocabulary.

9. The method for identifying and predicting social risks based on artificial intelligence according to claim 8, characterized in that: Also includes: S1-8: Divide the words in the emotional risk vocabulary into multiple categories according to the emotional intensity of the words and the emotional intensity of the text in the set network media, and calculate the emotional intensity of the words according to the following formula: ， in, For words The emotional intensity, For words Sentiment word polarity; Adverbs of degree The weight of The sentiment intensity of the sentiment word list, Indicates taking the maximum value; The emotional intensity of text in network media is calculated according to the following formula: ，， Where, For words Where m is the number of negative words, The result of negation is negation. is the sentence weight; Indicates whether a negative word appears.

10. A system comprising a storage medium and one or more processors, wherein the storage medium is used to store code that uses a computer language to compile the method for social risk identification and prediction based on artificial intelligence as described in any one of claims 1 to 9 into a computer program, and the computer program can be called and executed by one or more processors.

Images