CN108897792B - Disaster monitoring and analysis method for extracting multi-dimensional disaster-related information from the Internet - Google Patents

Abstract

The present disclosure provides a disaster monitoring and analysis method for extracting multi-dimensional disaster-related information from the Internet, including: S1: real-time acquisition and preprocessing of multi-source data; S2: assisting in building a disaster information extraction knowledge base through feature word multi-dimensional expansion algorithm FWME to extract The disaster loss information and the emotional feedback information of the disaster audience are used to obtain multi-dimensional disaster-related information; S3: Jointly monitor and analyze the disaster area according to the multi-dimensional disaster-related information. Through the construction of the disaster information extraction knowledge base, the disaster loss information contained in the Internet platform and the emotional feedback information of the audience during the disaster process are accurately extracted, and combined with the time and space dimensions for joint analysis, the disaster progress process is described in detail, and the auxiliary It is used for real-time disaster monitoring, disaster damage assessment and analysis, disaster rescue feedback, and follow-up impact assessment.

Description

Disaster monitoring and analysis method for extracting multi-dimensional disaster-related information from the Internet

technical field

The invention relates to the technical field of disaster monitoring and analysis, in particular to the extraction of multi-dimensional disaster-related information based on Internet multi-source text data for disaster monitoring and analysis.

Background technique

my country is a disaster-prone country. Various disasters cause a lot of human and property losses every year. Traditional disaster monitoring methods, such as satellite remote sensing and manual surveys, are often difficult to play in time due to harsh implementation conditions and high costs. And some disaster information, such as the emotional feedback of the disaster audience, is difficult to achieve by traditional disaster monitoring methods, and this information is also extremely important in disaster monitoring and analysis.

At present, many scholars monitor and analyze disasters based on social media, such as exploring typhoon movement trajectories and human behavior patterns in time and space, or mining public attention, emotional attitudes and other information based on text content when typhoons, earthquakes and other disasters occur. However, these methods are relatively coarse-grained and less targeted, and cannot specifically describe the details of the losses caused by disasters, nor can they reflect the public's emotional feedback on government rescue activities.

Extract fine-grained disaster-related information from Internet texts. Usually, the contextual features of such information are very sparse, and the same text often includes multiple fine-grained disaster themes. Traditional supervised learning methods not only require manual annotation of large-scale training corpora, but are mostly aimed at text classification of a single topic. The traditional rule-based method requires a large amount of expert knowledge to summarize the extraction rules, and the portability is poor.

SUMMARY OF THE INVENTION

(1) Technical problems to be solved

The present disclosure provides a disaster monitoring and analysis method for extracting multi-dimensional disaster-related information from the Internet, so as to at least partially solve the above-mentioned technical problems.

(2) Technical solutions

According to one aspect of the present disclosure, a disaster monitoring and analysis method for extracting multi-dimensional disaster-related information from the Internet is provided, including:

S1: Real-time acquisition and preprocessing of multi-source data;

S2: The disaster information extraction knowledge base is assisted by the feature word multi-dimensional expansion algorithm FWME to extract the disaster loss information and the emotional feedback information of the disaster audience, which is used to obtain multi-dimensional disaster-related information;

S3: Joint monitoring and analysis of the disaster area based on multi-dimensional disaster-related information.

In some embodiments of the present disclosure, the multi-source data in the step S1 comes from multiple Internet platforms, and is obtained through a search engine using keywords related to the specified disaster situation; the preprocessing includes text deduplication and simple-to-traditional conversion. And full half-width conversion, as well as the time stamp information and location information of text data are extracted and stored separately.

In some embodiments of the present disclosure, in step S2, the feature word multidimensional expansion algorithm FWME is used to assist in constructing a disaster information extraction knowledge base to extract disaster loss information and emotional feedback information of disaster audience groups, including:

S21: Build a knowledge base for disaster information extraction;

S22: Extract various feature words in the knowledge base based on the feature word multi-dimensional expansion algorithm FWME vertically and horizontally expand disaster information;

S23: Extract the disaster loss information and the emotional feedback information of the disaster audience.

In some embodiments of the present disclosure, in the step S21, the disaster information extraction knowledge base includes a disaster damage information identification knowledge base and a public sentiment feedback knowledge base, using the disaster damage and public sentiment feature words contained in the text for identification and classification, taking into account both negative words that counteract the semantics of the text and degree words that promote it.

In some embodiments of the present disclosure, the structure of the disaster information extraction knowledge base is represented as a tree with a height of 4 layers, wherein the top layer of the tree is the disaster information extraction knowledge base, and the second layer includes two subtrees, in:

The left subtree of the second layer represents the knowledge base of disaster damage information, and the corresponding node of the third layer represents each disaster damage category and negative word. Each disaster damage category node contains multiple leaf nodes that are not fixed, and the entity represented by each leaf node The information is independent of each other; each leaf node of the fourth layer stores a plurality of feature word pairs representing disaster damage events corresponding to each node of the third layer, and the feature words include the feature words representing disaster damage objects and the feature words of disaster damage actions , used to identify disaster damage events; the negative word node contains a series of negative word information, which is stored in the leaf nodes of the fourth layer below it; the feature words and negative words in each leaf node dynamically reserve a storage space for Save the index position information of the feature word or negative word in the text to be extracted, namely the disaster damage feature word index Index(W _i ) and the negative feature word index Index(N);

The second layer of the right subtree represents the public sentiment feedback knowledge base, the second layer of nodes represents sentiment feature words, negative feature words and degree feature words, and each related node has a leaf node, which stores the corresponding feature thesaurus, in which each feature word Dynamically reserve two storage spaces to save the score information and index information of the corresponding feature words in the text to be extracted, including the emotional feature word score Score(E) and its index Index(E); Negative feature word score Score (N) and its index Index(N); the degree feature word score Score(D) and its index Index(D).

In some embodiments of the present disclosure, in the step S22, the FWME algorithm is used to perform feature expansion for various types of feature words in the disaster information extraction knowledge base respectively from vertical and horizontal dimensions, wherein,

Vertical expansion includes using the word vector model integrated in the FWME algorithm, using the Internet disaster-related text as the corpus, calculating the similarity of various feature words, and taking similar words that meet the threshold as vertical supplementary words for the feature words to be expanded in the knowledge base. ;

Horizontal expansion includes, on the basis of vertical expansion, for existing feature words, using the synonymous relationship between words, the words that meet the synonymy conditions are used as synonymous supplementary words for each feature word in the knowledge base.

In some embodiments of the present disclosure, in the step S23, the extraction of disaster loss information based on the wind disaster knowledge base includes the following steps:

_S231 : The text to be processed is segmented according to _punctuation to form a set of short sentences {S ₁ , S ₂ , . Participle words, remove stop words, and record the index position information of the remaining words in each short sentence;

S232: Match the words in each short sentence in S231 with the feature word pairs in the leaf nodes under each disaster damage category node of the disaster damage knowledge base, and record the feature word pair when the feature word pair of a certain disaster damage category is completely satisfied. Index (w ₁ ) and Index (w ² ) of the word in the sentence, mark the sentence as a candidate sentence of the disaster damage category.

S233: Determine whether other words in the candidate sentence can match the negative words in the disaster damage information identification knowledge base, and if they match, record the index position Index(N) of the negative words.

S234: If and only if Index(N)<Index(w ₁ ) or Index(N)<Index(w ₂ ), the candidate sentence belongs to the corresponding disaster damage category.

In some embodiments of the present disclosure, in the step S23, the wind disaster knowledge base extracts the emotional feedback information of the disaster audience, including the following steps:

S231': Segment the text to be processed according to _punctuation to form a short sentence set {S ₁ , S ₂ , . Index position information in each phrase.

S232': Match the words in each short sentence in S231 with various emotional feature words, negative words and feature words in the leaf nodes under the degree word node in the public emotion feedback knowledge base respectively, and match the matched feature words Record its index position information Index(E), Index(N), Index(D) and scores Score(E), Score(N), Score(D).

S233': when the index value of the negative word and the degree word is less than the index value of the sentiment word, then multiply Score(N) and Score(D) with the corresponding sentiment value Score(E) respectively, and then sum the final scores , forming the score Score(S _i ) for the short sentence.

S234': Set a threshold sequence, and put each text score Score(S) into the threshold sequence, so as to obtain the sentiment category of the text.

In some embodiments of the present disclosure, the multi-dimensional disaster-related information in step S3 is to use the disaster information extraction knowledge base constructed in step S2 to extract disaster damage information and public sentiment contained in Internet text data related to a designated disaster event Feedback information, combined with the timestamp information and the location information contained in the Internet text, to construct a four-dimensional disaster array [loss, emotion, time, location].

In some embodiments of the present disclosure, in step S3, monitoring and analysis of the disaster-affected area is performed according to the multi-dimensional disaster-related information, including spreading the constructed four-dimensional disaster situation array [loss, emotion, time, location] on a vector map and making Space-time analysis, so as to conduct detailed monitoring and real-time analysis of the entire process of disaster occurrence, and the final results are used for real-time disaster monitoring, disaster damage assessment and analysis, disaster rescue feedback and subsequent impact assessment.

(3) Beneficial effects

It can be seen from the above technical solutions that the disclosed method for monitoring and analyzing disaster situations for extracting multi-dimensional disaster-related information from the Internet has at least one of the following beneficial effects:

(1) Accurately extract the disaster loss information contained in Internet platforms such as Weibo and the emotional feedback information of the audience during the disaster process through the constructed disaster information extraction knowledge base, and combine the time and space dimensions for joint analysis to describe the disaster situation in detail. progress, and assist in real-time disaster monitoring, disaster damage assessment and analysis, disaster rescue feedback, and follow-up impact assessment;

(2) Through the feature word multi-dimensional expansion algorithm, it overcomes the shortcomings of sparse context features of short texts, and the difficulty in extracting and classifying fine-grained disaster loss information, thereby realizing the automatic mining of disaster loss information contained in Internet texts, and efficiently assisting disaster reduction work. conduct.

Description of drawings

FIG. 1 is a schematic flowchart of a disaster monitoring and analysis method for extracting Internet multi-dimensional disaster-related information according to an embodiment of the present disclosure.

FIG. 2 is a schematic structural diagram of a disaster information extraction knowledge base according to an embodiment of the present disclosure.

FIG. 3 is a schematic flowchart of a method for vertically and horizontally expanding each feature word in a knowledge base by a feature word multi-dimensional expansion algorithm FWME according to an embodiment of the present disclosure.

FIG. 4 is a schematic diagram of the spatial distribution of various types of disaster damage during the typhoon transit process according to an embodiment of the present disclosure.

FIG. 5 is a schematic diagram of a spatiotemporal distribution sequence and an emotional feedback change sequence of various types of disaster damage during a typhoon transit process according to an embodiment of the present disclosure.

Detailed ways

The present disclosure provides a disaster monitoring and analysis method based on the joint analysis of multi-dimensional disaster-related information. A feature word multidimensional extension algorithm FWME (Feature Words Multidimensional Extension) is used to assist in the construction of a disaster information extraction knowledge base, so as to solve the problem of fine-grained multidimensional extension. The extraction accuracy of disaster-related information is not high and the existing disaster monitoring and analysis methods are insufficient, and the multi-dimensional disaster-related information contained in the Internet text is extracted through this knowledge base, and the change characteristics of disaster damage and the emotional changes of the disaster audience are analyzed by analyzing the time and space sequence. Features to assess disaster damage and government post-disaster relief activities.

The overall technical scheme adopted by the present invention includes:

S1. Real-time acquisition and preprocessing of multi-source data.

S2. A feature word multi-dimensional expansion algorithm FWME is proposed to assist in the construction of a disaster information extraction knowledge base to extract disaster loss information and the emotional feedback information of disaster audiences;

S3. Monitor and analyze the disaster area according to the multi-dimensional disaster-related information.

The real-time acquisition of multi-source data in S1 includes data from Internet platforms such as forum posts, WeChat public accounts, news, Sina Weibo, etc., using keywords related to the designated disaster situation as search conditions. Preprocessing includes text de-duplication, simple-to-traditional conversion, and full-width conversion. At the same time, the temporal and spatial position information of the text is separately extracted and saved to the database. The time and space location information of the text includes time stamp information, location information, and the like.

The multi-dimensional disaster information in the text is obtained by constructing a disaster information extraction knowledge base in S2, and the knowledge base stores a wealth of disaster damage characteristic word pair information used to represent various disasters and the emotional feedback characteristic words of the audience in the disaster. In addition, negative word information and degree word information used to accurately express semantic features are also stored. Each feature word in the knowledge base is enriched and supplemented by the feature word multidimensional expansion algorithm FWME proposed by the present invention.

The step S2 utilizes the feature word multidimensional expansion algorithm FWME algorithm to assist in the construction of a disaster information extraction knowledge base to extract disaster loss information and the emotional feedback information of the disaster audience, including:

S21. Build a knowledge base for disaster information extraction

The disaster information extraction knowledge base includes extracting the disaster loss information and the emotional feedback information of the disaster audience in the Internet text, and using the disaster damage and public emotion feature words contained in the text for identification and classification. For example, the characteristic words of disaster damage are “off-electricity”, public sentiment feedback words such as “happy” and “sad”, etc. In addition, negative words and promotion words that have a negative effect on the semantics of the text are also considered.

The structure of the disaster information extraction knowledge base can be represented as a tree with a height of 4, wherein the second layer of the tree has two subtrees, the left subtree represents the disaster damage information identification knowledge base, and the lower node represents each Disaster damage category and negative word, each disaster damage category node includes multiple leaf nodes that are not fixed, and the entity information represented by each leaf node is independent of each other. Each leaf node stores multiple feature word pairs representing disaster events. Negative word node contains a series of negative word information, stored in its leaf nodes. In addition, the feature words and negative words in each leaf node dynamically reserve a storage space to save the index position information of the feature words or negative words in the text to be extracted, that is, the index of the disaster feature word Index (Wi) and the negative feature word Index Index(N). The right subtree of the second layer represents the public sentiment feedback knowledge base, and its structure is similar to that of the disaster damage information identification knowledge base. The second layer nodes represent sentiment categories, negative words and degree words, and the relevant nodes each have a leaf node to store the corresponding features. Thesaurus, in which each feature word dynamically reserves two storage spaces to save the score information and index information of the corresponding feature word in the text to be extracted, that is, the emotional feature word score Score (E) and index Index (E) ; Negative feature word score Score(N) and index Index(N); Degree feature word score Score(D) and index Index(D).

S22 , extracting various feature words in the knowledge base based on the feature word multidimensional expansion algorithm FWME algorithm to vertically and horizontally expand disaster information.

The FWME algorithm is used to extract various disaster-related feature words in the knowledge base of disaster information extraction, and the feature expansion is carried out from the vertical and horizontal dimensions respectively. A large number of Internet disaster-related texts are used as corpus (including text data provided by forum posts, WeChat public accounts, news websites, Sina Weibo, etc.) to calculate the similarity of various feature words, and select similar words that meet the threshold (and the same words to be expanded). The feature words have the same part of speech) as the vertical supplementary words of the feature words to be expanded in the knowledge base. Horizontal expansion is based on the vertical expansion, using the synonymous relationship between the existing feature words to use the words that meet the synonymous conditions as the synonymous supplementary words of each feature word in the knowledge base, where the synonyms are calculated. Take "Synonym Cilin" as the corpus.

S23, extracting disaster loss information and emotional feedback information of disaster audience groups.

In S2, the disaster loss information is extracted through the disaster situation information extraction knowledge base, including the following steps:

S231: The text to be processed is segmented according to punctuation to form a short sentence set {S1, S2, ...Si, _i >=1}, Si represents the short sentences formed after the text to be processed is segmented, and each short sentence is divided into words and removed. Stop words, and record the index position information of the remaining words in each short sentence.

S232: Match the words in each short sentence in (S231) with the feature word pairs in the leaf nodes under each disaster damage class node of the disaster damage knowledge base, and when the feature word pair of a certain disaster damage class is completely satisfied, At the same time, the index positions Index(w1) and Index(w2) of the word in the sentence are recorded, and the sentence is marked as a candidate for the disaster damage category.

S233: Then judge whether other words in the candidate sentence can match the negative words in the disaster damage knowledge base, and if they match, record the index position Index(N) of the negative words.

S234: If and only if Index(N)<Index(w1) or Index(N)<Index(w2), the candidate sentence belongs to the corresponding disaster damage category.

The disaster information extraction knowledge base in the S2 extracts the emotional feedback information of the disaster audience, including the following steps:

S231': Segment the text to be processed according to punctuation to form a short sentence set {S1, S2, ...Si, i>=1}, segment each short sentence, remove stop words, and record the remaining words in each short sentence. The index position information in the sentence.

S232': Match the words in each short sentence in S231' with the various emotional feature words, negative words and the feature words in the leaf nodes under the degree word node in the public emotion feedback knowledge base respectively, and compare the matched features. The word records its index position information Index(E), Index(N), Index(D) and scores Score(E), Score(N), Score(D).

S233': when the index value of the negative word and the degree word is less than the index value of the sentiment word, then multiply Score(N) and Score(D) with the corresponding sentiment value Score(E) respectively, and then sum the final scores , forming the Score(Si) of the short sentence.

S234': The final score Score(S) of the entire sentence is the summation of the scores of each short sentence ∑ _i≥1 Score(S _i ).

A threshold sequence is set, and each text score Score(S) is put into the threshold sequence, so as to obtain the sentiment category of the text.

The multi-dimensional disaster-related information in the step S3 is to use the disaster information extraction knowledge base constructed in S2 to extract the disaster damage information and public sentiment feedback information contained in the Internet text data related to the designated disaster event, and combine the time of the Internet text. Stamp information and implied location information to construct a four-dimensional disaster array [loss, emotion, time, location].

In the described step S3, monitoring and analyzing the disaster-affected area according to the multi-dimensional disaster-related information is to spread the constructed four-dimensional disaster situation array [loss, emotion, time, location] on the vector map and make a time-space analysis, so as to analyze the disaster situation. The whole process is monitored in detail and analyzed in real time, and the final results can be used for disaster damage assessment and analysis, disaster rescue feedback, disaster follow-up impact assessment, etc.

The present disclosure proposes a disaster monitoring and analysis method for extracting multi-dimensional disaster-related information from the Internet. The disaster information extraction knowledge base constructed by the present disclosure can accurately extract the disaster loss information contained in the microblog and the emotional feedback information of the audience during the disaster process. Combined with the time and space dimensions for joint analysis, the disaster progress process is described in detail, and it is used for real-time disaster monitoring, disaster damage assessment and analysis, disaster rescue feedback, and subsequent impact assessment.

In order to make the objectives, technical solutions and advantages of the present disclosure clearer, the present disclosure will be further described in detail below with reference to the specific embodiments and the accompanying drawings.

Certain embodiments of the present disclosure will be described more fully hereinafter with reference to the accompanying drawings, some but not all embodiments of which are shown. Indeed, various embodiments of the present disclosure may be embodied in many different forms and should not be construed as limited to the embodiments set forth in this number; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements.

In the first exemplary embodiment of the present disclosure, a disaster monitoring and analysis method for extracting Internet multi-dimensional disaster-related information is provided. Taking the typhoon "Hato" in Zhuhai City on August 23, 2017 as an example, the real-time fine-grained monitoring and analysis of the area affected by the typhoon is carried out. The feature word multidimensional extension algorithm FWME (Feature Words Multidimensional Extension) assists the construction of a disaster information extraction knowledge base to extract the disaster loss information contained in the Internet text and the emotional feedback information of the disaster audience. Finally, combined with the time and space position information of the text, a joint analysis is carried out to assist in disaster damage assessment, disaster rescue and rescue feedback.

FIG. 1 is a schematic flowchart of a disaster monitoring and analysis method for extracting Internet multi-dimensional disaster-related information according to an embodiment of the present disclosure. As can be seen from the figure, the method includes the following processes:

S1. Real-time acquisition and preprocessing of multi-source data.

S2. Propose a feature word multidimensional extension algorithm FWME (Feature Words MultidimensionalExtension) to assist in the construction of a knowledge base for typhoon disaster information extraction to extract the typhoon disaster loss information and the emotional feedback information of the typhoon disaster audience;

S3. Monitor and analyze the wind-affected area according to the multi-dimensional disaster-related information.

Real-time acquisition and preprocessing of the S1 multi-source data. Obtain the Internet multi-source data in the monitoring and analysis area, and the data comes from forums and posts, WeChat public accounts, news, Sina Weibo and other platforms. Then preprocess all the data, including de-duplication, simple-to-traditional conversion, and full-half-width conversion, etc. At the same time, parse out the time information and location information of the text upload contained in it, and store it in a separate field.

In the step S2, the typhoon disaster information extraction knowledge base is constructed to obtain the typhoon disaster loss information contained in the text and the emotional feedback information of the typhoon disaster audience groups. The feature word pairs used to represent different disasters and the emotional feedback feature words of the audience in the disaster stored in the knowledge base are enriched and supplemented by the feature word multidimensional expansion algorithm FWME proposed by the present invention. In addition, the knowledge base also includes Some negative words and degree words used to express semantic features precisely.

FIG. 2 is a schematic structural diagram of a disaster information extraction knowledge base according to an embodiment of the present disclosure. As shown in FIG. 2, its structure can be represented as a tree with a height of 4, wherein the second layer of the tree has two subtrees, the left subtree Represents the knowledge base of disaster damage information identification, the lower node represents each disaster damage category and negative words, each disaster damage category node contains multiple leaf nodes that are not fixed, and the entity information represented by each leaf node is independent of each other. Each leaf node stores multiple feature word pairs representing disaster events. Negative word node contains a series of negative word information, stored in its leaf nodes. In addition, the feature words and negative words in each leaf node dynamically reserve a storage space to save the index position information of the feature words or negative words in the text to be extracted, that is, the index of the disaster feature word Index (Wi) and the negative feature word Index Index(N). The right subtree of the second layer represents the public sentiment feedback knowledge base, and its structure is similar to that of the disaster damage information identification knowledge base. The second layer nodes represent sentiment categories, negative words and degree words, and the relevant nodes each have a leaf node to store the corresponding features. Thesaurus, in which each feature word dynamically reserves two storage spaces to store the score information and index information of the corresponding feature word in the text to be extracted, that is, the emotional feature word score Score(E) and index Index(E); Negative feature word score Score(N) and index Index(N); degree feature word score Score(D) and index Index(D).

In this embodiment, the categories of disaster damage are divided into 11 categories, as shown in Table 1

Table 1

serial number Disaster damage category 1 traffic congestion 2 vehicle destruction 3 forest damage 4 casualties 5 water supply impact 6 building damage 7 business impact 8 communication impact 9 Power supply impact 10 Damage to electrical facilities 11 Damaged infrastructure

In this embodiment, the emotion categories are divided into three categories, as shown in Table 2

Table 2

serial number public sentiment 1 positive emotions 2 neutral mood 3 negative emotion

Set the score of each feature word in the public sentiment feedback knowledge base, in which the score of each positive sentiment feature word is Score(Positive)=1, the corresponding score of the negative sentiment feature word is Score(Negative)=-1, and the negative word score is is Score(Negation)=-1, and the degree word score is Score(Degree)=1.5.

FIG. 3 is a flow chart of the method for vertically and horizontally expanding each feature word in the knowledge base by the feature word multi-dimensional expansion algorithm FWME in S2.

The vertical expansion is to use the word vector model integrated in the FWME algorithm, take the multi-source Internet disaster-related texts as the corpus, calculate the similarity of the extended feature words, and take the similar words within the threshold as the vertical supplementary words of the knowledge base. . The words that satisfy the similarity threshold and have the same part of speech are selected as the vertical supplementary words of the feature word to be expanded.

The threshold is the 4 words with the closest similarity to the target word.

Longitudinal supplementation increases the depth of words, such as "tree" in the same context as "telephone pole", "street lamp", etc., and "down" in the same context as "broken", "blown down" and so on. as shown in Table 3.

table 3

The horizontal expansion is based on the vertical expansion, for the existing feature words, using the synonymous relationship between the words, the word set that satisfies the synonymy conditions is used as the supplement of each feature word in the knowledge base. For example, in the synonym calculation, the "Synonym Cilin" is used as the corpus, and the word that satisfies the synonymy conditions between words is taken as the synonym expansion word of the word.

The synonym condition is set to take the position of the existing feature word in the "Synonym Word Forest" as the benchmark, and obtain all the words in the atomic word set where the word is located as its synonym expansion word.

Horizontal expansion, expand the synonyms of feature words, such as the synonym expansion of "railing" as shown in Table 4.

Table 4

The same method is used for multi-dimensional expansion of sentiment feature words to enrich sentiment feature words.

Through feature word expansion, the disaster damage feature words in the knowledge base of wind disaster information extraction and the emotional feedback feature words of disaster audience groups are enriched.

In S2, the disaster loss information is extracted from the knowledge base of wind disaster information extraction, including the following steps:

(1) The text to be processed is segmented according to punctuation to form a short sentence set {S ₁ , S ₂ , ... S _i , i >= 1}, and each short sentence is divided into words and stop words are removed, and the remaining words are recorded at the same time. Index position information in each phrase.

(2) Match the words in each short sentence in (1) with the feature word pairs in the leaf nodes under each disaster damage category node of the disaster damage knowledge base respectively, when the feature word pairs of a certain disaster damage category are completely satisfied , and simultaneously record the index positions Index(w ₁ ) and Index(w ₂ ) of the word in the sentence, and mark the sentence as a candidate for the disaster category.

(3) Then judge whether other words in the candidate sentence can match the negative words in the disaster damage knowledge base, and if they can match, record the index position Index(N) of the negative words.

(4) If and only if Index(N)<Index(W ₁ ) or Index(N)<Index(W ₂ ), the candidate sentence belongs to the corresponding disaster category.

The S2 stroke disaster information extraction knowledge base extracts the emotional feedback information of the disaster audience, including the following steps:

(1) The text to be processed is segmented according to punctuation to form a short sentence set {S ₁ , S ₂ , ... S _i , i >= 1}, and each short sentence is divided into words and stop words are removed, and the remaining words are recorded at the same time. Index position information in each phrase.

(2) Match the words in each short sentence in (1) with the positive emotions, negative emotions, negative words in the public sentiment feedback knowledge base, and the characteristic words in the leaf nodes under the degree word node, respectively. The feature word records its index position information Index(Positive), Index(Negative), Index(Negation), Index(Degree) and scores Score(Positive), Score(Negative), Score(Negation), Score(Degree).

(3) When the index value of the negative word and the degree word is less than the index value of the sentiment word, the Score(Negation) and Score(Degree) are respectively multiplied with the corresponding sentiment value, and then the final positive and negative scores are summed to form Score(S _i ) for the phrase.

(4) The final whole sentence score Score(S) is the sum of the scores of each short sentence.

Score(S)=∑ _i≥1 Score(S _i ).

In this embodiment, the emotion threshold is set to 1 bit, and the value is 0;

When Score(S)<0, the sentiment value of the sentence is negative. Score(S)>0, the sentiment value of the sentence is positive. Score(S)=0, the sentiment value of the sentence is neutral.

In the S3, monitoring and analysis of the disaster area is carried out according to the multi-dimensional disaster-related information, and a four-dimensional information array [loss, emotion, time, location] is constructed from time information, spatial location information, disaster damage information and public sentiment feedback information, and is displayed on the map. Spread out, and make a sequence comparison according to the progress of the disaster to understand the disaster in detail.

Figure 4 shows a schematic diagram of the spatial distribution of various types of disaster damage during the typhoon transit process. It can be seen from Figure 4 that the Xiangzhou District suffered a lot of disasters, and the disaster damage was mainly in terms of water supply, power supply and traffic impact. Combined with the specific location information, rescue can be carried out for each disaster damage event.

Figure 5 shows a schematic diagram of the time-space distribution sequence and emotional feedback change sequence of various types of disaster damage during the typhoon transit process. It can be seen from Figure 5 that with the movement of the typhoon (the typhoon moves to the northwest), the disaster damage information in each transit area gradually increases, and different areas show different damage distribution and emotional feedback before and after the typhoon transits, such as 9:00 a.m. -Before the typhoon made landfall at 12:00, the main types of disaster damage were traffic impacts. During this period, there were many public trips and return trips, traffic was congested, and emotional attitudes were mostly negative. After the typhoon crosses the border, such as from 18:00 to 24:00, the relevant disaster damage information gradually decreases, and the positive public sentiment increases, especially in Xiangzhou District (the most affected by the typhoon). Large, public sentiment turned positive.

The present invention firstly uses Internet multi-source data as the corpus for disaster monitoring and analysis. Secondly, the feature word multi-dimensional expansion algorithm FWME is proposed to assist in constructing the knowledge base of typhoon disaster information extraction to extract the information of typhoon disaster loss and the emotional feedback information of the typhoon disaster audience. Finally, combined with time and space location information, a multi-sequence joint analysis of the disaster situation is carried out. Achieving real-time and comprehensive disaster monitoring in disaster-stricken areas to assist the implementation of disaster reduction and relief activities.

So far, the embodiments of the present disclosure have been described in detail with reference to the accompanying drawings. It should be noted that, in the accompanying drawings or the text of the description, the implementations that are not shown or described are in the form known to those of ordinary skill in the technical field, and are not described in detail. In addition, the above definitions of various elements and methods are not limited to various specific structures, shapes or manners mentioned in the embodiments, and those of ordinary skill in the art can simply modify or replace them.

It should also be noted that the directional terms mentioned in the embodiments, such as "up", "down", "front", "rear", "left", "right", etc., only refer to the directions of the drawings, not used to limit the scope of protection of the present disclosure. Throughout the drawings, the same elements are denoted by the same or similar reference numbers. Conventional structures or constructions will be omitted when it may lead to obscuring the understanding of the present disclosure.

Moreover, the shapes and sizes of the components in the figures do not reflect the actual size and proportion, but merely illustrate the contents of the embodiments of the present disclosure. Furthermore, in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim.

Furthermore, the word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements.

Furthermore, unless the steps are specifically described or must occur sequentially, the order of the above steps is not limited to those listed above, and may be varied or rearranged according to the desired design. And the above embodiments can be mixed and matched with each other or with other embodiments based on the consideration of design and reliability, that is, the technical features in different embodiments can be freely combined to form more embodiments.

The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used with teaching based on this. The structure required to construct such a system is apparent from the above description. Furthermore, this disclosure is not directed to any particular programming language. It is to be understood that various programming languages can be used to implement the disclosures described herein and that the descriptions of specific languages above are intended to disclose the best mode of the disclosure.

The present disclosure can be implemented by means of hardware comprising several different elements, as well as by means of a suitably programmed computer. Various component embodiments of the present disclosure may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the related apparatus according to the embodiments of the present disclosure. The present disclosure can also be implemented as apparatus or apparatus programs (eg, computer programs and computer program products) for performing some or all of the methods described herein. Such a program implementing the present disclosure may be stored on a computer-readable medium, or may be in the form of one or more signals. Such signals may be downloaded from Internet sites, or provided on carrier signals, or in any other form.

Those skilled in the art will understand that the modules in the device in the embodiment can be adaptively changed and arranged in one or more devices different from the embodiment. The modules or units or components in the embodiments may be combined into one module or unit or component, and further they may be divided into multiple sub-modules or sub-units or sub-assemblies. All features disclosed in this specification (including accompanying claims, abstract and drawings) and any method so disclosed may be employed in any combination unless at least some of such features and/or procedures or elements are mutually exclusive. All processes or units of equipment are combined. Each feature disclosed in this specification (including accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise. Also, in a unit claim enumerating several means, several of these means can be embodied by one and the same item of hardware.

Similarly, it will be appreciated that in the above description of exemplary embodiments of the disclosure, various features of the disclosure are sometimes grouped together into a single embodiment, figure, or its description. However, this method of disclosure should not be interpreted as reflecting an intention that the claimed disclosure requires more features than are expressly recited in each claim. Rather, as the following claims reflect, disclosed aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the Detailed Description are hereby expressly incorporated into this Detailed Description, with each claim standing on its own as a separate embodiment of the present disclosure.

The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present disclosure in detail. It should be understood that the above-mentioned specific embodiments are only specific embodiments of the present disclosure, and are not intended to limit the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure should be included within the protection scope of the present disclosure.

Claims (9)

1. A disaster monitoring and analyzing method for extracting multi-dimensional disaster-related information of the Internet comprises the following steps:

s1: real-time acquisition and pretreatment of multi-source data;

s2: disaster situation information extraction knowledge base is constructed in an auxiliary mode through a feature word multidimensional expansion algorithm FWME to extract disaster loss information and emotion feedback information of disaster audience groups, and the disaster loss information and the emotion feedback information are used for obtaining multidimensional disaster-related information; the method comprises the following steps: extracting various feature words in a knowledge base based on feature word multi-dimensional expansion algorithm FWME longitudinal and transverse expansion disaster situation information;

s3: performing combined monitoring and analysis on the disaster area according to the multidimensional disaster-related information;

the FWME algorithm is used for performing feature expansion on various feature words in the disaster information extraction knowledge base from a longitudinal dimension and a transverse dimension respectively, wherein,

the longitudinal expansion comprises the steps of utilizing a word vector model integrated in a FWME algorithm, taking Internet disaster-related texts as linguistic data, carrying out similarity calculation on various feature words, and taking the similar words meeting a threshold value as longitudinal supplementary words of feature words to be expanded in a knowledge base;

and the transverse expansion comprises the steps of utilizing the synonymy relation among the words and phrases for the existing characteristic words on the basis of longitudinal expansion, and taking the words meeting the synonymy condition as synonymy supplementary words of all the characteristic words in the knowledge base.

2. The disaster monitoring and analyzing method as claimed in claim 1, wherein the multi-source data in step S1 is obtained from a plurality of internet platforms by using keywords related to the designated disaster through a search engine; the preprocessing comprises text deduplication, complex and simple conversion and full half-angle conversion, and time stamp information and position information of text data are extracted and stored separately.

3. The disaster monitoring and analyzing method as claimed in claim 1, wherein in step S2, a disaster information extraction knowledge base is constructed with the help of a feature word multidimensional extension algorithm FWME to extract disaster damage information and emotional feedback information of disaster audience groups, which includes:

s21: constructing a disaster information extraction knowledge base;

s23: disaster damage information and emotional feedback information of disaster audience groups are extracted.

4. The disaster monitoring and analyzing method according to claim 3, wherein,

in step S21, the disaster information extraction knowledge base includes a disaster information recognition knowledge base and a public emotion feedback knowledge base, and the disaster information and public emotion feature words included in the text are used for recognition and classification, while negative words counteracting the text semantics and degree words contributing to the text semantics are considered.

5. The disaster monitoring and analyzing method according to claim 3, wherein,

the disaster information extraction knowledge base is structurally represented as a tree with a height of 4 layers, wherein the top layer of the tree is the disaster information extraction knowledge base, the 2 nd layer comprises two subtrees, wherein:

the left sub-tree of the layer 2 represents a disaster damage information identification knowledge base, the left sub-tree corresponds to the layer 3 node and represents each disaster damage category and a negative word, each disaster damage category node comprises a plurality of unfixed leaf nodes, and entity information represented by each leaf node is independent; each leaf node of the fourth layer stores a plurality of feature word pairs which represent the disaster damage events and correspond to each node of the 3 rd layer, wherein the feature words comprise feature words which represent disaster damage objects and feature words of disaster damage actions and are used for identifying the disaster damage events; the negative word node comprises a series of negative word information and is stored in the leaf node of the fourth layer below the negative word node; dynamically reserving a storage space for the feature words and the negative words in each leaf node to store Index position information of the feature words or the negative words in the text to be extracted, namely the disaster damage feature word Index (W)_i) And negative feature word Index (N);

the right subtree at the layer 2 represents a public emotion feedback knowledge base, the nodes at the layer 2 represent emotion characteristic words, negative characteristic words and degree characteristic words, each related node is provided with a leaf node and stores a corresponding characteristic word base, and each characteristic word dynamically reserves two storage spaces for storing Score information and Index information of the corresponding characteristic word in the text to be extracted, wherein the Score information comprises an emotion characteristic word Score (E) and an Index (E) thereof; negative feature word Score (N) and its Index (N); the degree feature word Score (D) and its Index (D).

6. The disaster monitoring and analyzing method according to claim 3, wherein,

in step S23, the extracting disaster damage information based on the knowledge base includes the following steps:

s231: text to be processed forms a short sentence set by punctuating punctuation and punctuating sentences { S₁，S₂，...S_i，i＞＝1}，S_iThe short sentences formed after the text to be processed is broken are represented, each short sentence is divided into words and stop words are removed, and index position information of the rest words in each short sentence is recorded;

s232: matching the words in each short sentence in S231 with the feature word pairs in the leaf nodes under each disaster type node of the disaster damage knowledge base, and recording the Index position Index (w) of the word in the sentence when the feature word pairs of a certain disaster type are completely met₁) And Index (w)²) Marking the sentence as a candidate sentence of the disaster damage category;

s233: judging whether other words in the candidate sentence can be matched with negative words in the disaster damage information identification knowledge base or not, and if so, recording the Index position Index (N) of the negative words;

s234: if and only if Index (N) < Index (w)₁) Or Index (N) < Index (w)₂) Then, the candidate sentence belongs to the corresponding damage category.

7. The disaster monitoring and analyzing method according to claim 3, wherein,

in step S23, the knowledge base extracts emotional feedback information of the disaster audience group, and includes the following steps:

s231': text to be processed forms a short sentence set by punctuating punctuation and punctuating sentences { S₁，S₂，...S_iI > -1 }, and dividing words and removing stop words for each short sentence, and recording index position information of the rest words in each short sentence;

s232': matching the words in the short sentences in the S231 with the characteristic words in the leaf nodes under the nodes of various emotion characteristic words, negative words and degree words in the public emotion feedback knowledge base respectively, and recording Index position information Index (E), Index (N), Index (D) and scores Score (E), Score (N) and Score (D) of the matched characteristic words;

s233': when the index values of the negative words and the degree words are less than the index values of the emotion words, respectively multiplying Score (N) and Score (D) by corresponding emotion value Score (E), and then summing the final scores to form Score (S) of the short sentence_i)；

S234': setting a threshold sequence, and putting each text Score (S) into the threshold sequence, thereby obtaining the emotion category of the text.

8. The disaster monitoring and analyzing method of claim 1, wherein the multidimensional disaster-related information in step S3 is a four-dimensional disaster array [ loss, observation, time, location ] constructed by extracting disaster damage information and public emotional feedback information included in internet text data related to a specified disaster event using the disaster information extraction knowledge base constructed in step S2, and simultaneously combining timestamp information and included location information of internet text.

9. The disaster monitoring and analyzing method of claim 8, wherein the step S3 of monitoring and analyzing the disaster-affected area according to the multidimensional disaster-related information includes spreading the constructed four-dimensional disaster array [ loss, observation, time, location ] on a vector map and performing space-time analysis, so as to perform detailed monitoring and real-time analysis on the entire process of disaster occurrence, and the final result is used for disaster real-time monitoring, disaster damage assessment analysis, disaster rescue feedback and subsequent impact assessment.

Images