CN106598944A - Civil aviation security public opinion emotion analysis method - Google Patents

Abstract

A sentiment analysis method for civil aviation public opinion. It includes retrieval, preprocessing and word segmentation of microblog texts on the Internet containing civil aviation security public opinion keywords; building a dictionary; scoring microblogs to obtain the sentiment score of the microblog; Obtaining the threat degree score of the microblog to civil aviation safety through objective discrimination; judging the threat degree level of the comments in the microblog text to civil aviation safety according to the threat degree score. The invention uses the combination of text semantics and microblog emoticons to determine the emotion score of the microblog text, overcomes the limitations of dictionaries and semantic rules, and improves the accuracy of emotion score judgment. Make full use of the characteristics of microblog text to make the judgment of threat level more reasonable. The present invention is different from the machine learning method, does not need large-scale labeled data for training, and therefore is more suitable for real-time data flow processing.

Description

A sentiment analysis method for civil aviation security public opinion

technical field

The invention belongs to the technical field of text sentiment analysis in natural language processing, and in particular relates to a method for sentiment analysis of civil aviation security public opinion.

Background technique

In the Internet age with rapid information expansion, more and more users tend to share their opinions or experiences through the Internet, so there are a large number of short texts with subjective emotions in social networks. Sina Weibo is an information sharing and communication platform that provides entertainment, leisure and life services for the public. At present, the number of active users of Sina Weibo remains at about 200 million. It inherits the advantages of traditional forums and blogs, combined with mobile terminals such as mobile phones, Enable information to be released and acquired quickly in real time. Weibo integrates entertainment, social networking, and marketing. It has gradually evolved from meeting people's social needs for "weak ties" to a popular public opinion platform, and has become the most important source of real-time information and an increasingly influential network of public opinion dissemination. Center, more and more institutions and public figures are publishing or disseminating information through Weibo.

Sentiment analysis is the process of processing, analyzing and applying emotional text, and it is a cutting-edge research field in natural language processing. It is a specific application combining many existing research results, and it has important practical value when combined with Weibo as a new network social media. The main purpose of microblog sentiment analysis is to identify subjective information from microblog information, and to mine users' views and attitudes on comment information such as products, news, and hot events.

In the field of civil aviation, the high degree of liberalization of online public opinion has brought some negative effects, such as publishing false threats, rumors, extreme language, etc. By analyzing the sentiment tendency of microblog texts related to civil aviation, microblogs that threaten the safety of civil aviation can be filtered out, so as to lock key users with criminal tendencies and push them to relevant public security departments for processing in a timely manner. In addition, text sentiment analysis has the following applications: predicting movie box office, stock trends, market dynamics, etc. Therefore, it is of great significance to analyze the sentiment tendency of microblog text.

At present, there are mainly two types of methods based on semantic understanding and machine learning based on Chinese text sentiment analysis methods. However, these two methods mainly have the following problems in the application of microblog sentiment analysis: ① The method based on semantic understanding uses the method of constructing a benchmark lexicon and defining expression rules to perform pattern matching on the corpus. There are great limitations on blog text processing. ②The method based on machine learning is limited by the selection of features and the size of the corpus, and is prone to over-fitting effects, so it is not suitable for real-time bulk text processing.

Contents of the invention

In order to solve the above problems, the object of the present invention is to provide a method for analyzing civil aviation security public opinion sentiment.

In order to achieve the above object, the civil aviation security public opinion emotion analysis method provided by the invention comprises the following steps carried out in order:

(1) Retrieve, preprocess, and word-segment operations on microblog texts on the Internet that contain civil aviation security public opinion keywords;

(2) Construct various dictionaries required for semantic analysis of microblog texts, and the construction methods are divided into selecting existing dictionaries and self-constructing methods;

(3) according to the dictionary that above-mentioned step (2) constructs, above-mentioned microblog after step (1) participle is scored, obtains the emotion score of this microblog;

(4) According to the emotional score obtained in step (3), the subjective and objective discrimination of microblogs is carried out, which is used to filter objective microblogs including news reports, keep subjective microblogs, and finally obtain the microblog’s impact on civil aviation Security threat score;

(5) According to the threat score obtained in step (4), determine the threat level of the remarks in the microblog text to civil aviation safety, and then select the key personnel with high threat level, and report it to the relevant department as early warning information.

In step (1), the method for retrieving, preprocessing and word segmentation operations on the microblog text containing civil aviation security public opinion keywords on the Internet is: grabbing the microblog texts containing civil aviation security public opinion keywords on the Internet, Retrieve keywords related to civil aviation security public opinion from these microblog texts. The keywords are divided into two types: location words and action words, and the retrieval strategies are divided into single word retrieval and combination retrieval; and then perform preprocessing operations on the above retrieval results , to remove noise information including user nicknames, hashtags, and special characters when web links, reposts, and replies to Weibo, and extract emoticons; then use word segmentation tools to segment the above preprocessed results. The word segmentation tool uses Java Open source word segmentation tool Ansj.

In step (2), the dictionaries include sentiment dictionaries, negative dictionaries, modification dictionaries, conjunction dictionaries, emoticon dictionaries, Internet hot word dictionaries and civil aviation security public opinion dictionaries.

In step (3), the described dictionary constructed according to the above-mentioned steps (2) scores the above-mentioned microblog after the word segmentation through step (1), and the method for obtaining the emotional score of the microblog includes the following steps:

1) Extract or determine the emotional words from the above-mentioned microblog text after step (1) word segmentation:

The method for extracting emotional words is to match the words obtained after word segmentation in the above-mentioned microblog text with the above-mentioned emotional dictionary and the network hot word dictionary, and if a certain word exists in the above-mentioned two dictionaries, it is selected as an emotional word;

The method of determining the emotional word is to use the semantic similarity method for words that do not appear in the emotional dictionary and the online hot word dictionary; the specific method is for two words w ₁ and w ₂ , if the word w ₁ has n meanings or concepts : x ₁ , x ₂ …, x _n , the word w ₂ has m meanings or concepts: y ₁ , y ₂ …, y _m , it is stipulated that the similarity between words w ₁ and w ₂ is the maximum of the similarities between the meanings or concepts value, namely:

The formula for calculating the similarity between two sememes is:

Among them, λ is a positive variable parameter; d(x ₁ , y ₂ ) represents the distance between sememe x ₁ and sememe y ₂ in the hierarchical tree;

Calculate the similarity between the word w and each seed word in the positive sentiment dictionary according to formula (1) and formula (2) to obtain the similarity between the word and the positive seed word, and then compare the word w with each seed word in the negative sentiment dictionary Carry out the similarity calculation to obtain the similarity between the word and the negative seed word. By comparing the mean difference between them, the emotional tendency value of the word w is finally obtained. The calculation formula is as follows:

Among them, p _i represents a positive emotional seed word, n _j represents a negative emotional seed word; the value range of the emotional tendency value S _w is (-1, 1); setting the threshold T, the calculated emotional tendency The value S _w is compared with the threshold T to determine whether the word w belongs to an emotional word; when |S _w |>T, it is determined that the word w is an emotional word, and the strength of the emotional word is set at 10·S _w ;

2) Determine the text emotion score of each microblog clause containing the above-mentioned emotional words in the microblog;

2.1) If the microblog clause contains emotional words, and there are negative words in the negative dictionary or modifier words in the modification dictionary before it, the text sentiment score Sa of the microblog clause is calculated according to the following situations:

e) Adverbs of degree + emotional words, the intensity of emotional words changes with the intensity of adverbs, and the emotional score of the text is:

Sa＝M _a ps pa (4)

f) Negative words + emotional words, the polarity of the emotional words changes according to the number of negative words, and the sentiment score of the text is:

Sa=(-1) ⁿ ps pa (5)

g) degree adverb + negative word + emotional word, the polarity of the emotional word is reversed, and the intensity changes with the intensity of the adverb, the text emotion score is:

Sa=(-1) M _a ps pa (6)

h) Negative words + degree adverbs + emotional words, since negation appears before degree adverbs, after the polarity of emotional words is reversed, the strength of emotional words is weakened compared with direct negation, the first weight factor z ₁ = 0.5 is introduced, and the text emotion score for:

Sa=(-1) M _a ps pa z ₁ (7)

Among them, ps represents the intensity of emotional words, pa represents the polarity of emotional words, and M _a represents the intensity of degree adverbs:

2.2) If the microblog clause contains transitional conjunctions in the conjunction dictionary, the microblog clause is a compound sentence. Considering the emotional polarity transfer between sentences, the text sentiment score of the microblog clause is calculated according to the following situations:

d) Turning relationship: when semantic reversal words such as "but" and "however" appear in a microblog clause, the polarity of the previous microblog clause will change, and the overall polarity of the two microblog clauses will be the same as The latter microblog clause is the same, and the second weight factor z ₂ =-1 is introduced, and the sentiment score of the text is:

Sen＝z ₂ Sen ₁ +Sen ₂ (8)

e) Progressive relationship: the two microblog clauses before and after have the same polarity, and the strength is enhanced. The third weight factor z ₃ =1.5 is introduced, and the text sentiment score is:

Sen＝z ₃ (Sen ₁ +Sen ₂ ) (9)

f) Concession relationship: the polarity of the latter microblog clause will be reversed, the polarity of the whole sentence is the same as that of the previous microblog clause, and the fourth weight factor z ₄ =-1 is introduced, and the text sentiment score is:

Sen＝Sen ₁ +z ₄ Sen ₂ (10)

Among them, Sen ₁ represents the text sentiment score of the previous microblog clause, and Sen ₂ represents the text sentiment score of the latter microblog clause;

3) Determining the score of emoticons in Weibo;

According to the emoticon dictionary, find out the polarity and intensity of all emoticons in the microblog, and record the number of each emoticon; let N _i be the number of the i-th emoticon, e _i be the number of the emoticon Intensity, p _i is the polarity of the emoticon, then the calculation formula of the emoticon score in Weibo is:

4) The above-mentioned microblog text sentiment score and emoticon score are weighted and summed to obtain the sentiment score of each microblog. The formula is as follows:

S ₁ ＝α·score _emo +β·score _text (12)

Among them, α and β are adjustable weights, the value range is (0,1), α+β=1, the value of α and β when the probability of correct classification is the highest can be selected through cross-test set verification; score _text is the micro The text sentiment score of the blog is the average of the text sentiment scores of each microblog clause.

In step (4), the subjective and objective discrimination is carried out to microblogs according to the emotional score obtained in step (3), for filtering objective microblogs including news reports, and retaining microblogs with subjectivity, The method to finally obtain the threat degree score of the Weibo to civil aviation safety is:

First, the following methods are used to judge the subjective and objective microblog text:

1) For a microblog with an emotional score S ₁ =0, if it contains first-person nouns or pronouns, it is considered to be a subjective microblog text, otherwise it is an objective microblog text;

2) For a microblog with an emotional score S ₁ ≠ 0, if it contains special predicate words for news reports, or the number of reposts in the microblog text is at least 2 times, it is considered as an objective microblog text, otherwise it is a subjective microblog text;

Set the threat score of the objective microblog text to 0, and do not calculate the threat score, but only calculate the threat score of the subjective microblog. The calculation formula is shown in formula (13):

Among them, D represents the threat score, and the range is between [-10,10]; S ₁ represents the sentiment score of the microblog text; S ₂ <w ₁ , w ₂ > is the civil aviation security public opinion threat score, and w ₁ represents place words, w ₂ represents behavior words;

The calculation process of civil aviation security public opinion threat score S ₂ ＜w ₁ ,w ₂ ＞is as follows: find the behavioral word w ₂ in the microblog text, and then judge the type of the behavioral word; when the behavioral word is direct, the civil aviation security public opinion The value of the threat score S ₂ <w ₁ ,w ₂ > is the strength of the behavioral words; when the behavioral words are indirect, judge whether there are location words in the microblog text at the same time, if they exist at the same time, civil aviation security public opinion threat The value of the score S ₂ <w ₁ ,w ₂ > is the strength of the behavior word, and if it does not exist at the same time, the threat score S ₂ <w ₁ ,w ₂ >is 0.

In step (5), the method for determining the degree of threat level of remarks in the microblog text to civil aviation safety according to the degree of threat score that step (4) obtains is:

When the threat degree score D > 0, the microblog text expresses positive emotions and belongs to safety speech, so the threat level judgment is not performed; when the threat degree score D ≤ 0, it is determined that the microblog text contains civil aviation security public opinion keywords, and express negative emotions, which need to be focused on, and then judge the threat level of the microblog text according to the following threat level standard; the threat level standard is obtained by testing the existing microblog text ,details as follows:

1) When -4.5≤D≤0, it is a low threat degree;

2) When -7≤D<-4.5, it is a medium threat;

3) When -10≤D<7, it is a high threat level.

The civil aviation security public opinion sentiment analysis method provided by the present invention has the following advantages: (1) the present invention utilizes the mode that text semantics and microblog emoticons combine to determine the sentiment score of microblog text, overcomes the limitation of dictionary and semantic rule, Improved the accuracy of sentiment score judgments. (2) Calculate the threat score of the microblog text on the basis of the emotional score of the microblog text, and obtain the threat level, which improves the early warning ability of the civil aviation public security department, which is of great significance. (3) Make full use of the characteristics of microblog text to make the threat level judgment more reasonable. (4) The present invention is different from machine learning methods and does not need large-scale labeled data for training, so it is more suitable for real-time data stream processing.

Description of drawings

Fig. 1 is a flow chart of the method for analyzing civil aviation security public opinion sentiment provided by the present invention.

Fig. 2 is a flow chart of the calculation method of emotion score in the present invention.

detailed description

The civil aviation security public opinion sentiment analysis method provided by the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

As shown in Figure 1, the civil aviation security public opinion emotion analysis method provided by the present invention comprises the following steps carried out in order:

(1) Retrieve, preprocess, and word-segment operations on microblog texts on the Internet that contain civil aviation security public opinion keywords;

Grab the microblog texts that contain civil aviation security public opinion keywords on the Internet, as the analysis object of the present invention, retrieve keywords related to civil aviation security public opinion from these microblog texts, keywords are divided into two types of location words and behavior words, for To ensure the comprehensiveness of data acquisition, the retrieval strategy is divided into two methods: single word retrieval and combination retrieval. Single word retrieval refers to the retrieval of two types of words separately. Among them, location words such as airport, runway, terminal building, flight, etc., behavior Words such as plane bombing, hijacking, fuss, brawl, protest, etc.; combined retrieval is the retrieval mode of location + behavior, such as "airport + bomb", "flight + explosion", "airport + hijacking", etc., and store the retrieval results in the database;

At the same time, in order to improve the efficiency of the system, a "white list" of Weibo users is set up. The Weibo users in this list are the official Weibo of the airport police and news portal Weibo. Since these microblog users often post microblogs containing keywords of civil aviation security public opinion, but they are not within the scope of early warning and monitoring, they are removed from the keyword search.

Then preprocess the above retrieval results to remove noise information irrelevant to emotional expression in the microblog text, for example: (1) Web page links, such as "http://t.cn/Rtj0WWN", etc., because they do not contain Useful information, so it is removed during preprocessing. (2) User nicknames, hashtags, special characters, etc. when reposting and replying to Weibo, in the form of "@李琼子: Reply to @船图托老王: This matter belongs to the police uncle in the mental hospital", where the Weibo after the @ symbol The blog username needs to be removed. And extract the emoticons, because the emoticons in the captured Weibo text are represented by square brackets, such as "This is the testimony of word-of-mouth and effect [strong]", the emoticons are The text in the square brackets is extracted for the calculation of the emotional value of the emoticon.

Then use the word segmentation tool to segment the above preprocessed results. The word segmentation tool uses the Java open source word segmentation tool Ansj.

(2) Construct all kinds of dictionaries required for semantic analysis of microblog texts. The construction methods are divided into selecting existing dictionaries and self-construction. The composition of various dictionaries is as follows:

1) Emotional dictionaries: The current mainstream emotional dictionaries include NTUSD Dictionary of National Taiwan University, HowNet Chinese-English Emotional Dictionary, Dalian University of Technology Emotional Lexicon Ontology Database, etc. Among these mainstream dictionaries, the Dalian University of Technology Emotional Vocabulary Ontology Database divides words into three categories: positive, negative, and neutral, and marks the emotional polarity according to five intensities of 1, 3, 5, 7, and 9, which is beneficial to micro The calculation of blog text emotion value, so select this dictionary as the emotion dictionary in the present invention.

2) Negative dictionary: when negative words appear in front of emotional words, the inversion of emotional polarity can take place, the present invention has counted 19 commonly used negative words, is constructed into negative dictionary thus. The negative words described are: no, no, nothing, no, Mo, Eph, no, no, no, other, nothing, not enough, not, never, not necessarily, no, not, difficult, never.

3) Modification dictionary: According to the semantic rules of Chinese grammar, the strength of emotional value is directly related to the modification function of degree adverbs, so it is necessary to select modifiers as one of the judgment rules. The modified dictionary in the present invention selects the Chinese level words in HowNet (HowNet), and is divided into 6 levels, as shown in Table 1.

Table 1 Modification dictionary example

4) Conjunction dictionary: In compound sentences, conjunctions can lead to changes in emotional polarity, thereby affecting the judgment of emotional polarity. The present invention selects the conjunctions with turning relationship, progressive relationship and concession relationship as the conjunction dictionary. The above-mentioned conjunctions expressing turning relationship are: but, but, however, but, but, however, that is, only, can; the conjunctions expressing progressive relationship are: and, even, even, and, especially; expressing concessional relationship The conjunctions of are: even though, despite, even though, even if, even though.

5) Emoji dictionary: Sina Weibo provides a large number of emoticons, and the present invention selects the built-in emoticons of Sina Weibo, and the emotional polarity is divided into two categories, positive and negative, and manually marked according to intensity. As shown in table 2.

Table 2 Example of emoji dictionary

6) Dictionary of online hot words: Weibo, as a social media, has the characteristics of informality and colloquialism, so the frequency of use of online words is very high, but these popular words are not included in the traditional emotional dictionary. It is very necessary to add new words on the Internet to the emotional dictionary. In Wangci.net (http://wangci.net), there are thesaurus and definitions of Internet popular terms. The present invention captures 494 network hot words of the website, marks the emotional polarity and intensity according to the same criterion as the sentiment dictionary, and constructs a network hot word dictionary as a supplement to the sentiment dictionary.

7) Civil aviation security public opinion dictionary: select the location words and behavior words in the civil aviation security public opinion keywords to construct the civil aviation security public opinion dictionary. Except the basic words, the present invention collects the homophonic typos of common words to increase the completeness of the dictionary. For example, the homophonic typos of "boom" and "jue" and the homophonic typos of "jie" are "cut", etc., and the words added in the dictionary are "fraud machine", "fraud bomb", "cut machine" and so on.

(3) according to the dictionary that above-mentioned step (2) constructs, above-mentioned microblog after step (1) participle is scored, obtains the emotion score of this microblog;

The specific steps are shown in Figure 2.

1) Extract or determine the emotional words from the above-mentioned microblog text after step (1) word segmentation:

The words obtained after word segmentation in the above-mentioned microblog text are matched with the above-mentioned emotional dictionary and the network hot word dictionary, and if a certain word exists in the above-mentioned two dictionaries, then it is selected as an emotional word;

If the word does not appear in the sentiment dictionary and the network hot word dictionary, the sentiment word is determined by the method of semantic similarity. In order to reduce the amount of computation, only nouns, verbs and adjectives are reserved as candidate emotional words. The invention uses the HowNet semantic similarity algorithm as a benchmark algorithm, which has a good effect on measuring the similarity of two words. The specific method is for two words w ₁ and w ₂ , if word w ₁ has n meanings or concepts: x ₁ , x ₂ ..., x _n , word w ₂ has m meanings or concepts: y ₁ , y ₂ ... ,y _m , it is stipulated that the similarity of words w ₁ and w ₂ is the maximum value of the similarity of each meaning item or concept, that is:

The formula for calculating the similarity between two sememes is:

Wherein, λ is a positive variable parameter; d(x ₁ , y ₂ ) represents the distance between sememe x ₁ and sememe y ₂ in the hierarchical tree.

For any word, its emotional tendency value can be obtained by calculating the similarity between the word and the seed word in the sentiment dictionary. The calculation method is: the word w and each seed word in the positive sentiment dictionary according to formula (1) and formula (2) Carry out the similarity calculation to obtain the similarity between the word and the positive seed word, and then calculate the similarity between the word w and each seed word in the negative sentiment dictionary to obtain the similarity between the word and the negative seed word, by comparing them The average difference between the words, and finally get the emotional tendency value of the word w, the calculation formula is as follows:

Among them, p _i represents a positive emotional seed word, n _j represents a negative emotional seed word; the value range of the emotional tendency value S _w is (-1,1). Set the threshold T, and compare the calculated emotional tendency value S _w with the threshold T to determine whether the word w belongs to the emotional word. When |S _w |>T, the word w is judged to be an emotional word. The strength of the sentiment word is set as 10·S _w , which is consistent with the magnitude of strength in the sentiment dictionary.

2) Determine the text emotion score of each microblog clause containing the above-mentioned emotional words in the microblog;

2.1) If the microblog clause contains emotional words, and there are negative words in the negative dictionary or modifier words in the modification dictionary before it, the text sentiment score Sa of the microblog clause is calculated according to the following situations:

i) degree adverb+emotional word, the intensity of the emotional word changes with the intensity of the adverb, and the emotional score of the text is:

Sa＝M _a ps pa (4)

j) Negative words + emotional words, the polarity of emotional words changes according to the number of negative words, and the sentiment score of the text is:

Sa=(-1) ⁿ ps pa (5)

k) Degree adverb + negative word + emotional word, the polarity of the emotional word is reversed, and the intensity changes with the intensity of the adverb, the text emotion score is:

Sa=(-1) M _a ps pa (6)

l) Negative words + degree adverbs + emotional words, since negation appears before degree adverbs, after the polarity of emotional words is reversed, the strength of emotional words is weakened compared with direct negation, the first weight factor z ₁ = 0.5 is introduced, and the text emotion score for:

Sa=(-1) M _a ps pa z ₁ (7)

Among them, ps represents the intensity of emotional words, pa represents the polarity of emotional words, and M _a represents the intensity of degree adverbs:

2.2) If the microblog clause contains transitional conjunctions in the conjunction dictionary, the microblog clause is a compound sentence. Considering the emotional polarity transfer between sentences, the text sentiment score of the microblog clause is calculated according to the following situations:

g) Turning relationship: when semantic reversal words such as "but" and "however" appear in a microblog clause, the polarity of the previous microblog clause will change, and the overall polarity of the two microblog clauses will be the same as The latter microblog clause is the same, and the second weight factor z ₂ =-1 is introduced, and the sentiment score of the text is:

Sen＝z ₂ Sen ₁ +Sen ₂ (8)

h) Progressive relationship: the two microblog clauses before and after have the same polarity, and the strength is enhanced. The third weight factor z ₃ =1.5 is introduced, and the sentiment score of the text is:

Sen＝z ₃ (Sen ₁ +Sen ₂ ) (9)

i) Concession relationship: the polarity of the latter microblog clause will be reversed, the polarity of the whole sentence is the same as that of the previous microblog clause, and the fourth weight factor z ₄ =-1 is introduced, and the text sentiment score is:

Sen＝Sen ₁ +z ₄ Sen ₂ (10)

Among them, Sen ₁ represents the text sentiment score of the previous microblog clause, and Sen ₂ represents the text sentiment score of the latter microblog clause;

3) Determining the score of emoticons in Weibo;

Sina Weibo provides a large number of emoticons, and the use of emoticons in Weibo can clearly express the emotional tendency of the Weibo. Taking emoticons as a weighted item of emotional score has a certain corrective effect on the judgment of the emotional tendency of the entire microblog text. According to the emoticon dictionary, find out the polarity and intensity of all emoticons in the microblog, and record the number of each emoticon; let N _i be the number of the i-th emoticon, e _i be the number of the emoticon Intensity, p _i is the polarity of the emoticon, then the calculation formula of the emoticon score in Weibo is:

4) The text sentiment score and emoticon score of the above microblogs are weighted and summed to obtain the sentiment score of each microblog. The formula is as follows:

S ₁ ＝α·score _emo +β·score _text (12)

Among them, α and β are adjustable weights, the value range is (0,1), α+β=1, the value of α and β when the probability of correct classification is the highest can be selected through cross-test set verification; score _text is the micro The text sentiment score of the blog is the average of the text sentiment scores of each microblog clause. When the sentiment score S ₁ is positive, it is determined that the microblog expresses positive emotion; when the sentiment score S ₁ is negative, it is determined that the microblog expresses negative emotion.

(4) According to the emotional score obtained in step (3), the subjective and objective discrimination of microblogs is carried out, which is used to filter objective microblogs including news reports, keep subjective microblogs, and finally obtain the microblog’s impact on civil aviation Security threat score;

First, the following methods are used to judge the subjective and objective microblog text:

1) For a microblog with an emotion score S ₁ =0, if it contains first-person nouns or pronouns, it is considered as a subjective microblog text, otherwise it is an objective microblog text.

2) For a microblog with an emotional score S ₁ ≠ 0, if it contains special predicate words for news reports, or the number of reposts in the microblog text is at least 2 times, it is considered as an objective microblog text, otherwise it is a subjective microblog text.

Set the threat score of the objective microblog text to 0, and do not perform the following calculation of the threat score, and only calculate the threat score of the subjective microblog. The calculation formula is shown in formula (13):

Among them, D represents the threat score, and the range is between [-10,10]; S ₁ represents the sentiment score of the microblog text; S ₂ <w ₁ , w ₂ > is the civil aviation security public opinion threat score, and w ₁ represents place words, w ₂ represents behavior words;

In the Civil Aviation Security Public Opinion Dictionary, location words include airport, runway, terminal building, flight, etc., and behavior words include bombing, hijacking, air trouble, fight, protest, etc. Among them, behavior words have two attributes, the first attribute is Intensity, which measures the threat level of the word to civil aviation security. The measurement standard is divided into five intensities of 1, 3, 5, 7, and 9, which are consistent with the intensity measurement of emotional words. The second attribute is the word type. The word type is divided into two types. One is the direct type, that is, only the appearance of this word can be judged as a threat to civil aviation, such as bombing, hijacking, and overlord, etc.; the other is Indirect type, that is, it must appear together with the location words to determine whether there is a threat to civil aviation security, such as fights, protests, smoking, etc. When there are only indirect behavior words, it is not enough to judge that they pose a threat to civil aviation security.

The calculation process of civil aviation security public opinion threat score S ₂ ＜w ₁ ,w ₂ ＞is as follows: find the behavioral word w ₂ in the microblog text, and then judge the type of the behavioral word; when the behavioral word is direct, the civil aviation security public opinion The value of the threat score S ₂ <w ₁ ,w ₂ > is the strength of the behavioral words; when the behavioral words are indirect, judge whether there are location words in the microblog text at the same time, if they exist at the same time, civil aviation security public opinion threat The value of the score S ₂ <w ₁ ,w ₂ > is the strength of the behavior word, and if it does not exist at the same time, the threat score S ₂ <w ₁ ,w ₂ >is 0.

(5) According to the threat score obtained in step (4), determine the threat level of the remarks in the microblog text to civil aviation safety, and then select the key personnel with high threat level, and report it to the relevant department as early warning information.

From the threat degree score obtained in step (4), it can be seen that when the threat degree score D>0, the microblog text expresses positive emotions and belongs to safe speech, so the threat level judgment is not performed; when the threat degree When the degree score D≤0, it is determined that the microblog text contains civil aviation security public opinion keywords and expresses negative emotions, which needs to be focused on, and then the threat level of the microblog text is judged according to the following threat level standards. The threat level standard is obtained by testing existing Weibo texts, as follows:

1) When -4.5≤D≤0, it is a low threat degree.

2) When -7≤D<-4.5, it is a medium threat.

3) When -10≤D<7, it is a high threat degree.

Table 3 lists the threat degree scores and threat degree levels obtained after processing some microblog texts according to the method of the present invention. It can be seen from the table that the method of the present invention can more accurately determine whether the microblog text is a threat to civil aviation safety.

Table 3 Judgment results of threat degree of microblog text

Claims (6)

1. a kind of civil aviaton's security public sentiment sentiment analysis method, it is characterised in that：Described civil aviaton's security public sentiment sentiment analysis method Including the following steps for carrying out in order：

(1) operation of line retrieval, pretreatment and participle is entered to the microblogging text comprising civil aviaton's security public sentiment key word on the Internet；

(2) build for all kinds of dictionaries needed for the analysis of microblogging text semantic, construction method is divided into the existing dictionary of selection and autonomous The mode of construction；

(3) dictionary built according to above-mentioned steps (2), gives a mark to the above-mentioned microblogging Jing after step (1) participle, obtains this micro- Rich emotion score value；

(4) subjective and objective differentiation is carried out to microblogging according to the emotion score value obtained in step (3), for news report being filtered interior Objective microblogging, retains the microblogging with subjectivity, finally gives Threat score value of the microblogging to safety of civil aviation；

(5) the Threat score value obtained according to step (4) judges the Threat grade of speech in microblogging text to safety of civil aviation, Then the high emphasis personnel of Threat grade are filtered out, and relevant departments are reported and submitted as early warning information.

2. civil aviaton's security public sentiment sentiment analysis method according to claim 1, it is characterised in that：It is in step (1), described To on the Internet comprising civil aviaton's security public sentiment key word microblogging text enter line retrieval, pretreatment and participle operation method It is：Microblogging text comprising civil aviaton's security public sentiment key word on crawl the Internet, the retrieval from these microblogging texts are related to civil aviaton The key word of security public sentiment, key word are divided into two class of place word and behavior word, search strategy be divided into single word retrieval and Combined retrieval two ways；Then pretreatment operation is carried out to above-mentioned retrieval result, to remove web page interlinkage, forward, reply micro- User's pet name, topic label when rich, spcial character is in interior noise information, and extracts emoticon；Participle work is utilized afterwards Tool carries out participle through pretreated result to above-mentioned, and participle instrument is increased income participle instrument Ansj using Java.

3. civil aviaton's security public sentiment sentiment analysis method according to claim 1, it is characterised in that：It is in step (2), described Dictionary include sentiment dictionary, negative dictionary, modification dictionary, conjunction dictionary, emoticon dictionary, network hot word dictionary and civil aviaton Security public sentiment dictionary.

4. civil aviaton's security public sentiment sentiment analysis method according to claim 1, it is characterised in that：It is in step (3), described The dictionary built according to above-mentioned steps (2), the above-mentioned microblogging Jing after step (1) participle is given a mark, the microblogging is obtained The method of emotion score value comprises the following steps：

1) extract or determination emotion word from the above-mentioned microblogging text Jing after step (1) participle：

The method for extracting emotion word is the word and above-mentioned sentiment dictionary and net that will be obtained after participle in above-mentioned microblogging text Network hot word dictionary is matched, if a certain word is present in above-mentioned two dictionary, is chosen for emotion word；

The method for determining emotion word is to adopt semantic similitude to not appearing in the word in sentiment dictionary and network hot word dictionary Degree method is carried out；Concrete grammar is for two word w₁And w₂If, word w₁There are the n senses of a dictionary entry or concept：x₁,x₂…,x_n, word Language w₂There are the m senses of a dictionary entry or concept：y₁,y₂…,y_m, it is stipulated that word w₁And w₂Similarity be each senses of a dictionary entry or concept similarity most It is big to be worth, i.e.,：

$Sim(w_1,w_2)=\underset{i\&Element;\&lsqb;1,n\&rsqb;,j\&Element;\&lsqb;1,m\&rsqb;}{max}\left(Sim\right(x_i,y_j\left)\right)---\left(1\right)$

Two former calculating formula of similarity of justice are：

$Sim(x_1,y_2)=\frac{\mathrm{\&lambda;}}{\mathrm{\&lambda;}+d(x_1,y_2)}---\left(2\right)$

Wherein, λ is positive variable element；d(x₁,y₂) represent adopted original x₁With adopted original y₂Distance in hierarchical tree；

By each seed words in word w and positive emotion dictionary by formula (1) and formula (2) carry out Similarity Measure obtain the word with just The similarity of face seed words, then by each seed words in word w and negative emotion dictionary carry out Similarity Measure obtain the word with The similarity of negative seed words, by comparing the equal difference between them, finally gives the Sentiment orientation value of word w, calculates public Formula is as follows：

$S_w=\frac{1}{n}\underset{i=1}{\overset{n}{\&Sigma;}}Sim(w,p_i)-\frac{1}{m}\underset{j=1}{\overset{m}{\&Sigma;}}Sim(w,n_j)---\left(3\right)$

Wherein, p_iRepresent a certain positive emotion seed words, n_jRepresent a certain negative emotion seed words；Sentiment orientation value S_wValue Scope is (- 1,1)；Given threshold T, by Sentiment orientation value S for calculating_wIt is compared with threshold value T, whether to judge word w Belong to emotion word；When | S_w| during ＞ T, word w is judged as emotion word, the intensity of the emotion word is set to 10S_w；

2) determine the text emotion score of each microblogging clause comprising above-mentioned emotion word in microblogging；

2.1) if emotion word is included in microblogging clause, and occur before it belonging to the negative word or modification dictionary in negative dictionary In qualifier when, calculate text emotion score Sa of microblogging clause by following several situations：

A) degree adverb+emotion word, emotion word intensity change with adverbial word intensity, and text emotion must be divided into：

Sa=M_a·ps·pa (4)

B) negative word+emotion word, the polarity of emotion word change according to the number of negative word, and text emotion must be divided into：

Sa=(- 1)ⁿ·ps·pa (5)

C) degree adverb+negative word+emotion word, emotion word polarity inversion, and intensity change with adverbial word intensity, and text emotion is obtained It is divided into：

Sa=(- 1) M_a·ps·pa (6)

D) negative word+degree adverb+emotion word, before occurring in degree adverb due to negative, after emotion word polarity inversion, emotion Word intensity negates more directly to have weakened, and introduces the first weight factor z₁=0.5, text emotion must be divided into：

Sa=(- 1) M_a·ps·pa·z₁ (7)

Wherein, ps represents the intensity of emotion word, and pa represents emotion word polarity, M_aRepresent the intensity of degree adverb：

If 2.2) comprising the adversative conjunction in conjunction dictionary in microblogging clause, microblogging clause belongs to compound sentence, it is contemplated that between sentence Feeling polarities transfer, calculate the text emotion score of microblogging clause by following several situations：

A) turning relation：When occurring the semantic reversion vocabulary such as " but ", " but " in microblogging clause, the pole of previous microblogging clause Property will change, the integral polarity of the two microbloggings clause will be identical with latter microblogging clause, introduce the second weight because Sub- z₂=-1, text emotion must be divided into：

Sen=z₂Sen₁+Sen₂ (8)

B) progressive relationship：Former and later two microbloggings clause's polarity is identical, intensity enhancing, introduces the 3rd weight factor z₃=1.5, text Emotion must be divided into：

Sen=z₃(Sen₁+Sen₂) (9)

C) concession relation：The polarity of latter microblogging clause can occur reversion, and the polarity of whole sentence is identical with previous microblogging clause, draws Enter the 4th weight factor z₄=-1, text emotion must be divided into：

Sen=Sen₁+z₄Sen₂ (10)

Wherein, Sen₁Represent the text emotion score of previous microblogging clause, Sen₂Represent the text emotion of latter microblogging clause Score；

3) determine emoticon score in microblogging；

According to emoticon dictionary, the polarity and intensity of all emoticons in the microblogging are found, and records each emoticon Number；Make N_iFor the number of i-th emoticon, e_iFor the intensity of the emoticon, p_iFor the polarity of the emoticon, then Emoticon score computing formula in microblogging is：

${\mathrm{score}}_{emo}=\underset{i=1}{\overset{N}{\&Sigma;}}{N}_i\&CenterDot;e_i\&CenterDot;p_i---\left(11\right)$

4) above-mentioned microblog text affective score and emoticon score are weighted into summation, obtain the emotion of each microblogging Score value, formula are as follows：

S₁=α score_emo+β·score_text (12)

Wherein, α, β are adjustable weights, and span is that (0,1), alpha+beta=1 can select correct by the checking of cross-beta collection α, β value when class probability is maximum；score_textFor the text emotion score of the microblogging, it is each microblogging clause text emotion score Meansigma methodss.

5. civil aviaton's security public sentiment sentiment analysis method according to claim 1, it is characterised in that：It is in step (4), described Subjective and objective differentiation is carried out to microblogging according to the emotion score value obtained in step (3), for filtering news report interior objective Microblogging, retains the microblogging with subjectivity, and finally giving the microblogging to the method for the Threat score value of safety of civil aviation is：

Subjective and objective differentiation is carried out to microblogging text initially with following methods：

1) for emotion score value S₁=0 microblogging, if wherein including first person noun or pronoun, then it is assumed that be subjective microblogging text This, is otherwise objective microblogging text；

2) for emotion score value S₁≠ 0 microblogging, if the wherein special predicate word comprising news report, or in microblogging text Hop count at least 2 times, then it is assumed that be objective microblogging text, is otherwise subjective microblogging text；

The Threat score value of objective microblogging text is set as into 0, and degree of impending score value is not calculated, only calculate subjective microblogging Threat score value, shown in computing formula such as formula (13)：

$D=\frac{1}{2}(S_1-S_2<w_1,w_2>)---\left(13\right)$

Wherein, D represents Threat score value, and scope is between [- 10,10]；S₁Represent the emotion score value of microblogging text；S₂＜ w₁,w₂ ＞ is that civil aviaton's security public sentiment threatens fraction, w₁Represent place word, w₂Expression behavior word；

Civil aviaton's security public sentiment threatens fraction S₂＜ w₁,w₂The calculating process of ＞ is as follows：Search the behavior word w in microblogging text₂, so The type of behavior word is judged afterwards；When behavior word is Direct-type, civil aviaton's security public sentiment threatens fraction S₂＜ w₁,w₂＞ Value take the intensity of behavior word；When behavior word is indirect-type, judge whether there is ground simultaneously in the microblogging text Point word, if while existed, civil aviaton's security public sentiment threatens fraction S₂＜ w₁,w₂The value of ＞ takes the intensity of behavior word, such as Exist when fruit is different, threaten fraction S₂＜ w₁,w₂＞ is 0.

6. civil aviaton's security public sentiment sentiment analysis method according to claim 1, it is characterised in that：It is in step (5), described The Threat score value obtained according to step (4) judge the method for speech in microblogging text to the Threat grade of safety of civil aviation It is：

As Threat score value D ＞ 0, the microblogging text representation is positive emotion, belongs to safe speech, therefore does not impend Degree grade judges；When Threat score value D≤0, judge that the microblogging text contains civil aviaton's security public sentiment key word, and express be Negative Affect, needs to pay close attention to, and then microblogging text degree of impending grade is sentenced according to following Threat classification standard It is fixed；Threat classification standard be existing microblogging text is tested obtained from, it is specific as follows：

1) during -4.5≤D≤0 it is low Threat；

2) during -7≤D ＜ -4.5 it is medium Threat；

3) during -10≤D ＜ 7 it is high Threat.