JP2023504961A - Typhoon incident information convergence method - Google Patents

Abstract

A method for converging typhoon incident information according to the present invention searches for a stage or a posture in which the process of typhoon incidents progresses by selecting, organizing, and matching typhoon incident information that has different origins in social media. In particular, it provides an information basis for regularization. A typhoon incident information convergence method according to the present invention includes a first step of collecting message documents related to typhoon incidents on social media, extracting typhoon incident information, and converting them into a structured information tuple format. , the second step of converging the target information based on the similarity of the multi-features, the third step of converging the state information based on the spatio-temporal features, the convergence of the state process information, and the result of converging the target information, Select the spatio-temporal node information that satisfies the time and position range requirements, converge the state information for each of these spatio-temporal nodes, sort the convergence results of multiple state information, and obtain the convergence result of the process information in which dynamic features appear. , mainly including the fourth step.

Description

The present invention relates to the field of big data mining, and in particular to a typhoon event information convergence method.

Since typhoons have a very negative and destructive impact on the natural ecology, socio-economy, and even on the sustainable development of humans, it is difficult to obtain relevant information on the process of typhoon events in a timely manner. , has become an important basis and reference for disaster incident response. Currently, in an environment driven by big data, social media has great prospects for application in managing disasters due to its efficient update frequency, spread routes from numerous sources and deep involvement. , gradually become a new means of acquiring typhoon incident information. However, social media itself has characteristics such as extremely fragmented information, various formats of expression, and diversified granularity of information, in addition to the fact that documents are short. The disorganized social media information not only makes it difficult to show the progress of the typhoon event, but also prevents users from effectively predicting the typhoon event process.

The information convergence method improves the rationality of the information structure and optimizes the access efficiency by effectively expressing the information resources so as to meet the needs and convenience of users to acquire effective information resources. . Information convergence methods for disaster events mainly include statistical methods, thematic model methods and knowledge element methods. (1) The statistical method uses statistical features such as word frequency, TF-IDF, N-gram, and mutual information to calculate the weight of the keywords in the information cell, select the most representative keyword, and It converges based on The method is simple and subjectively understandable, but the accuracy of keyword screening is not high and generally requires a second screening with auxiliary information. (2) A probabilistic theme model expresses the subject in an information cell using the probability distribution of the subject words, assuming that each document has a potential distribution for all the subject words. However, the effectiveness of this method is determined by the number of subjects, and in real social media, different subjects are constantly changing dynamically. The same message on social media may contain content from multiple themes, and thus there is a great deal of contention about the scope of interpretation of the subject words. (3) Knowledge elements define ethical relationships and hierarchical structures between different concepts, and common forms of knowledge elements include subjects, semantic networks, and related data. Convergence by knowledge elements involves constructing a conceptual model that expresses the composition of disaster events based on the knowledge element theory, sorting and structuring information again based on the semantic relationships defined in the model, and structuring the information. characteristics and their relationships.

Currently, statistical and thematic model methods are the most popular forms of disaster incident information convergence. However, the results converged by these two methods have a large granularity of information, and are usually nothing more than collections of various types of information related to disaster events. On the other hand, the convergence method by knowledge elements can decompose and reconstruct primitive resources based on the concept system in the disaster field, and obtain deep convergence results with a certain knowledge structure. However, at present, modeling of typhoon event knowledge often focuses on the hierarchical structure and relational relationships of each concept in typhoon events, but often overlooks the description and representation of the dynamic processes of typhoon events. In view of the scattered social media resources, which are large in volume and complex in categorization, it is necessary to create an information convergence method to sort and match typhoon incident information based on the incident progressing process.

The present invention lays a foundation for regulating information in examining the development stage and posture of the process of typhoon incidents by sorting, organizing and matching typhoon incident information with scattered sources in social media, and It is an object of the present invention to provide a typhoon incident information convergence method that helps improve the level of service provided by social media resources in managing incident responses.

In order to achieve the above object, the present invention provides the following technical means.
The typhoon incident information convergence method is
A first step of collecting message documents related to typhoon incidents on social media, extracting typhoon incident information, and converting them into a structured information tuple format;

A step of target information convergence based on the similarity of multi-features, in which the similarity between target names is used to determine whether the information tuples correspond to the same target, and the information tuples representing the same target are converged. a second step in which it is necessary to

It is a step of state information convergence based on spatio-temporal features, and from the result of converging target information, attribute values and action values that satisfy the requirements of a single time and location condition are selected. a third step in which the obtained attribute values and action values together constitute a result of converging the target information in a particular spatio-temporal state;

A step of process information convergence based on state, in which spatio-temporal node information that satisfies the requirements of time and position range is selected from the result of converging target information, and for each of these spatio-temporal nodes, while converging state information, Mainly includes a fourth step of sorting the converged results of the plurality of state information to form a converged result of process information representing dynamic features.

Preferably, in the first step, the typhoon event information includes target name, time information, location information, attribute information and action information.

Preferably, in the second step, it is necessary to further converge attribute items and action items of the same category for different information tuples representing the same object.

Preferably, in the first step, extracting typhoon incident information includes at least two parts: identification of information elements and association of information elements.
The identification of information elements identifies the constituent objects in the typhoon event, constructs a classification system, and extracts the names and characteristic information representing different classified objects from the social media documents, provided that the characteristic information includes time, location, and attributes. and conduct. The attribute information may be further divided into an attribute item indicating the classification of the attribute and an attribute value indicating the data or data volume of the attribute of the classification. Action information is similar to attribute information.

The association of information elements associates the objects and names of feature information in the same social media document to form an information tuple in the format On=<T, L, A, B>. However, On is a target name, T is time information, L is position information, A is attribute information, and B is action information.

Preferably, the second step uses the similarity of word embedding to determine the similarity between the object name, the attribute item and the action item, and includes the following steps.
Step S1 extracts data of all social media documents.

Step S2 trains word embedding using the Skip-gram model with the extracted result as a training data set.

In step S3, given that the target names are On1 and On2, the attribute items are A1 and A2, and the action items are B1 and B2, On1, On2, A1, A2, B1, For B2, get the word embeddings E(On1), E(On2), E(A1), E(A2), E(B1), E(B2).

Step S4 uses the cosine similarity to calculate similarity values between E(On1) and E(On2), E(A1) and E(A2), E(B1) and E(B2), respectively. Calculate simn, sima and simb. If simn≧εn, sima≧εa, simb≧εb (εn, εa, εb are thresholds), then On1 and On2, A1 and A2, B1 and B2 are the same object name, attribute item and action item. It turns out that it is possible to converge the information correspondingly.

Preferably, in the fourth step, sorting the results of converging the plurality of state information includes the following steps.
Step A1 sorts according to forward order or reverse order according to the state time information.

Step A2 sorts according to the method of large size or small size according to the position information of the states.

In step A3, the state attribute information and the action information may be sorted based on the magnitude or class of the feature value, or may be sorted based on the degree of similarity with the user's convergence conditions.

The following technical effects can be obtained using the above technical means.
The present invention constructs a typhoon event process information convergence method based on social media, in identifying different target information tuples related to typhoon events in social media documents, from "object-state-process" respectively: Multi-order convergence modes are discussed. First, based on the similarity of multidimensional features in the object layer, we converge the scattered feature information of the same object, and then converge the attribute information and action information of the object that satisfy the spatio-temporal features specified in the state layer. Then, the spatio-temporal granularity of the information is unified, and finally, in the process layer, the states are sorted according to the spatio-temporal relationship to achieve the effect of regularizing and structuring the information. Such a convergence mode should be based on social media, where information is distributed, granular, and non-regularized, but it is also possible to consider the dynamic progress of typhoon events, and to consider the dynamic progression of typhoon events. At the nodes, the feature information of different objects can be obtained and the regularization information representing the process features of the typhoon event is formed. In practical applications, it can play an important role in meeting the needs of government agencies to perform incident response tasks and public awareness throughout society.

It is the convergence mode of process information of multi-magnitude typhoon events. It is a spatio-temporal semantic means created in social media. It is an example of the result of extracting typhoon incident information from social media. It is an organizational structure and an example about the result of having converged object information. Figure 10 is an organizational structure and example of the results of converging state information; There are different stages of converging process information. Organizational structure and examples of the results of converging process information.

The present invention will now be described in detail with reference to the drawings and specific embodiments.
Embodiment The present invention discloses a method for converging process information of a typhoon event based on social media, which includes the following steps.
The first step is to collect message documents related to typhoon incidents on social media, extract typhoon incident information including target name, time information, location information, attribute information and action information, and convert it into a structured information tuple format. Convert.

The second step is target information convergence based on multi-feature similarity. It is necessary to determine whether the information tuples correspond to the same target based on the degree of similarity between the target names, and to converge the information tuples expressing the same target. For different information tuples representing the same object, it is necessary to further converge attribute items and action items of the same category.

The third step is state information convergence based on spatiotemporal features. Attribute values and action values that satisfy the requirements of a single time and location condition are selected from the result of converging the target information. Construct the result of converging the target information in the state.

The fourth step is state-based process information convergence. Select spatio-temporal node information that satisfies the time and position range requirements from the results of converging the target information, converge state information for each of these spatio-temporal nodes, sort the results of converging multiple states, and dynamically Form a convergence result of the characterized process information.

As preferred technical means, extracting typhoon event information in the first step includes:
1. Identify the constituent objects in the typhoon event, create a classification system, and extract the names and characteristic information representing different classified objects from the social media documents, where the characteristic information includes time, location, attributes and actions. . The attribute information may be further divided into an attribute item indicating the classification of the attribute and an attribute value indicating the data or data volume of the attribute of the classification. Action information is similar to attribute information.

2. In the same social media document, for the feature information, associate the object that appears with the name, and form an information tuple in the format On=<T, L, A, B>. However, On is a target name, T is time information, L is position information, A is attribute information, and B is action information.

As a preferred technical measure, constituent objects in a typhoon event are divided into subject objects and object objects. A cyclone is the subject object in an event as a disaster-causing element, and any other object destroyed, acted upon, or affected by the cyclone is the object object in the event. Different properties in the object subject may be classified respectively, mainly including categories such as people, infrastructure, transportation facilities and social activities. What should be explained can be divided into more detailed categories according to actual needs, referring to the classification methods in related fields for different subjects (Table 1).

Table 1 Major target categories during a typhoon event Target category Target name Subject target Cyclone Object target Person Mankind Thing Infrastructure Building/Outer wall decoration/Signboard/…
Communication facility
power supply facility
water supply facility
Gas supply facility
Urban greening/trees/gardens/…
Transportation facilities General roads/bridges/tunnels/…
highway
plane
train/motor vehicle/high-speed rail
car
Steamship/ferry Social activity Commercial area/shopping mall Factory
Kindergarten/Elementary school/High school/University
Sightseeing spots/ecological gardens/amusement parks/…
…

As preferred technical measures, extracting names and characteristic information representing different categorized objects in social media documents includes: a.
Step S1 builds a corpus representing information of typhoon events in social media documents, and the represented contents include name, time, location, attributes and action information elements representing different categorized objects.

Step S2 builds a temporal information extraction model according to the notation corpus and based on the conditional random field model to automatically identify the temporal information in the social media document.

Step S3 builds a location information extraction model according to the notation corpus and based on the deep belief network to automatically identify the location information in the social media document.

Step S4 summarizes the rule model of object name, attribute information and action information, including trigger dictionary and grammatical mode, according to the notation corpus, and automatically identifies the object name, attribute information and action information in the social media document.
As a preferred technical measure, associating each information element extracted from social media includes the following steps.

Step S1 constructs spatio-temporal semantic means. Characters, vocabulary, phrases, subsentences, sentences, paragraphs, etc. are each linguistic units in a document, and different linguistic units form the basic structure of the document through their semantic relationships. By some linguistic units, or by combining different linguistic units, complete semantic connotations can be expressed, that is, semantic means can be expressed. In this method, this word meaning means is defined as a spatio-temporal meaning means because the word meaning means includes temporal information and spatial information, and when the contents are discussed in the word meaning means, the existing spatio-temporal features are clearly expressed.

Analyze social media documents containing typhoon incidents. The distribution of spatio-temporal semantic means can be broadly divided into (1) target information describing only the same time and position, and such documents occupy the majority of social media documents. (2) The target information has the same writing completion time but different positions, and the number of such documents is relatively small. (3) Target information that compares multiple times and locations, which falls under integrated reporting and can be divided into three categories: low number of such documents;

Spatio-temporal information can be used to track changes in target features in tracked documents. Therefore, the method divides the social media documents into different spatio-temporal semantic means (FIG. 2) based on the extracted spatio-temporal information. The grounds for spatio-temporal semantic means are divided according to the position of the spatio-temporal information in the document. Specifically, it includes:
(1) For the first type case, the whole document is divided into one spatio-temporal semantic means because it contains only one time and position information.

(2) For the second and third cases, first divide the document into multiple time means based on the time information. When a plurality of pieces of position information exist in the time means, further division is performed based on the position information, and the time information in the time means is shared for the divided spatio-temporal semantic means.

Step S2 is an association rule between the object name and feature information. Based on dividing the social media document into multiple spatio-temporal semantic means, the identified object names and each kind of characteristic information are distributed in different means. Therefore, the organization can be structured by the means to which each information element belongs. Each spatio-temporal semantic means associates different information elements according to the following steps.

(1) Associating feature trigger words with feature values. Together, the feature trigger word and the feature value constitute the feature information of interest. In this case, only the attribute feature and the action feature are meant, the feature trigger word indicates the attribute item and the action item, and the feature value indicates the attribute value and the action value. In order to follow the discipline of proximity in expressing feature trigger words and feature values, we form the construct "feature trigger word feature value". By statistically counting the word frequencies of the previous three vocabularies in attribute values, the feature trigger words appear more frequently than 99%. Therefore, we associate the feature value with the trigger word of the feature whose position is closest first.

(2) Associate attributes, action information and object names. The basic phraseological practice of Chinese generally refers to the name of the object and then discusses, respectively, the various characteristics that the object has. Therefore, the attribute information and the action information are each associated with the target name whose position is closest to the same spatio-temporal semantic means.

(3) Associate the target name with time and location information. For the spatio-temporal semantic means in which the target name is located, the time information and the position information are associated with the target name, respectively.
Object names and various feature information for which a relational relationship is constructed in order are filled according to a tuple format of On=<T, L, A, B> (FIG. 3). What should be explained is that representing a typhoon event in one spatio-temporal semantic means may be limited to only one aspect, and one of attributes and actions may be missing when constructing the target information tuple. .

As preferred technical means, converging the target information in the second step includes the following convergence.
1, convergence based on object name; Assuming that the target name of the convergence condition is N, the degree of similarity simn between the On name and N is determined sequentially. If simn≧εn, where εn is the similarity threshold of the objects, then we know the same object and merge the information tuples of the same object.

A word embedding similarity method is adopted as a method for determining the degree of similarity of target names. The word embedding similarity method is based on training a word embedding model using a Skip-gram model. It determines whether the different vectors match directions in multi-dimensional space, and uses cosine similarity to determine magnitude.
For example, O (typhoon) = < 1:45 on August 10, 2019, Wenling City, Zhejiang Province, wind speed, 16 classes, landfall>, O (tropical cyclone) = < 20:50 on August 11, 2019, Shandong Qingdao, province, wind speed, class 9, landing> is an information tuple extracted from social media. If the target name of the convergence condition is "typhoon", the degree of similarity is judged for the target names "typhoon" and "tropical cyclone" in the information tuples, respectively, and it is assumed that both of these terms express the cyclone itself. Let the two item information tuples be the convergence result.

2. Convergence that combines the features of interest. When converging information tuples of the same target, feature information of attributes and actions with the same classification will appear in multiple items, and target information that matches specific features is further converged. Based on the result of converging the target names, the target attribute feature A and the action feature B in the convergence conditions are set. For attribute feature convergence, the similarity sima between the On attribute item and A is determined using the word embedding similarity method. When sima≧εa (where εa is the attribute similarity threshold), the attribute items are the same and the information converges, and after convergence, each attribute value and spatio-temporal feature can be retained as they are, Otherwise, it expresses different attribute terms in the same object, and it turns out that the attribute terms do not converge.

Converging action features determines the similarity simb of On action items and B using the word embedding similarity method. If simb≧εb (εb is the action similarity threshold), the manifestation action item is the same, the information converges, and similarly after convergence, each action information and spatiotemporal feature can be retained as it is, Otherwise, it represents different action items in the same object, and it turns out that the action items do not converge.
For example, based on the O (typhoon) and O (tropical cyclone) target information tuples described above, the attribute feature information of "wind speed" related to the typhoon is further converged. O (typhoon) and O (tropical cyclone) each have an attribute item "wind speed" that matches the similarity threshold. 16 class> and <20, 50, August 11, 2019, Qingdao City, Shandong Province, wind speed, 9 class> are the convergence results of the target feature.

3. Structuring the information of the target convergence result. The organizational format of the target information convergence result is shown in FIG. However, O (N) is the target of convergence, Al is the attribute item of the converged target, als is the specific attribute value, Bn is the action item of the converged target, bnu is the specific action value, <T, S> is the time and place at which the attribute value or action value occurred. In the first place, the scattered pieces of information are associated with the expressed object, and the same attribute item and action item in the object are jointed, and each attribute and action item has different characteristics expressed in multiple spatio-temporal conditions. I know it contains a value.

As a preferred technical means, converging state information in the third step includes: a.
1. To unify the space-time standards. The space-time frame is the basis on which states exist, and converging state information requires building a unified space-time reference. The time reference used herein is the Gregorian calendar, the time is Beijing time, and the spatial reference is the CGCS2000 coordinate system.

2. To normalize spatio-temporal information. Time information and location information are grounds for whether or not their associated attribute information and action information are features that express the target state under specific spatio-temporal conditions. Time information is now normalized using the Western calendar, calendar time, and clock time according to people's daily habits. The time normalization format is 'YYYY-MM-DD HH, MM, SS' with the format 'Date+Time', for example, '2019-08-10 12, 00, 00'. The location information is converted into a normalized display format according to the uniform spatial standard, and includes representation contents such as place names, addresses and spatial coordinates. However, place names may refer to standard names, codes and classifications published by the state at a specific time. The classification of address elements included in the address and the combination form of the elements may refer to standard norms published by the state or industry. Spatial coordinates undergo coordinate transformation correspondingly according to the requirements of the spatial standard.

3. Convergence for states. Set the time feature t and the position feature l to be converged, converge the result O(N) based on the target layer information, and set T = t and S = l for each attribute item and action item in O(N) If there is a feature value (attribute value and action value) of exists, and if it exists, this feature value is taken as convergence information, otherwise S = l, T < t and the feature value closest to t If it exists, this feature value is used as convergence information. If not, continue to determine if there exists a feature value with S close to l, T<t and closest to t, and if so, similarly take this feature value as convergence information. Do not converge this attribute item or action item if it still does not exist. By traversing all attribute items and action items in O(N), from each attribute item and action item, at most one item selects a feature value that best matches the spatio-temporal feature. These attribute information and action information are converged and together form a converged result of the state information of the object under the specified spatiotemporal conditions.

For example, if a message is recorded on social media at 1:45 on August 10th, the cyclone wind speed reaches level 16 in Wenling City, Zhejiang Province, the cyclone state at (2,00, Wenling City) will be converged. However, since the information about the wind speed is not updated between 1:45 and 2:00, the attribute feature of one item of the state (2:00, Wenling City) in the cyclone target is “16 levels of wind speed”. and With such a convergence mechanism, for the convergence result at any space-time node obtained, the state information is not limited to the target features explicitly mentioned as belonging to the current space-time, and In addition, the latest progress to date in every feature of interest can be included to ensure a fully and completely convergent result.

4. Structuring of information resulting from state convergence. The resulting organizational format of converging state information is shown in FIG. where S is the state where the object O(N) exists at time t and position l, Al and als denote the attribute features of the state, Bn and bnu denote the action features of the state, <T, S > is the time and position at which the attribute and action feature occurred.

As a preferred technical means, in the fourth step, the process information convergence includes two parts: state sequence convergence and event process convergence. A process is a spatio-temporal connection of different states, and the dynamics of the process are expressed by changes in attribute information and action information in the state. A typhoon event involves the evolution process of multiple objects during the occurrence of the event, and the process of a typhoon event is composed together by different states of multiple objects. Therefore, when the process layer information is converged, we use the form of decomposing it by class, and abstract the connection between the state information and the process information into three stages: the target state, the state order, and the event process (Fig. 6). ). However, the object state is the convergence of the object's attribute information and action information in a certain space-time, and the state sequence records the evolution process of the same object, and it is necessary to converge different states for the same object. Yes, an event process is a process in which multiple objects have evolved together, and are composed together by a sequence of multiple states.

As a preferred technical means, converging the state hierarchy includes the following steps.
Step S1 sets the time range tr and the spatial range sr to be converged, converges the result O(N) based on the target information, and sequentially traverses all attribute items and action items in O(N). . For each attribute item and action item, determine whether the attribute value or action value of T⊆tr and S⊆sr exists, and form all <T, S> satisfying the tr given sr range as a spatio-temporal node set . For all space-time nodes in the set, the convergence results of multiple states are obtained by the method according to the third step.

Step S2 sorts the convergence results of all states, first according to the time information of the states, sorting in the form of forward or reverse order, and then according to the position information of the states, the magnitude Sorted in the form of from or from small, and finally, based on the state attribute information and action information, may be sorted by the magnitude or class of the feature value, or sorted by the degree of similarity with the user's convergence condition You may The ordering of states ordered by three-dimensional conditions is the result of converging the process of a single object.

Step S3 structures the information resulting from the convergence of the state order. The organizational format resulting from converging state hierarchy information is shown in FIG. where P is the upper process of the object O(N) over time range tr and space range sr, and S is the object state at the spatio-temporal node <tn, ln>.

The preferred technical means, converging the course of events, includes the following steps.
Step S1 sets the time range tr and the space range sr to be converged, converges the result O(Ns)-O(Nt) for the target information of multiple items, first, all the attribute items in O(Ns) and Traversal the action items to get <T, S> that match the tr and sr ranges. Then traverse O(Ns+1) repeatedly until all of O(Nt) have been traversed. Form all <T, S> that match the tr given sr range as a set of spatiotemporal nodes.

Step S2 needs to use the same sorting mechanism for multiple target state orders to ensure that the overall convergence results are consistent in order. Regarding the convergence results for the event process, comparing the state features at different time nodes before and after the process, we analyzed the movement of the spatial features and the differences in the attributes and action features, and clarified the dynamic process of the entire typhoon event (Fig. 7). to be recorded.

Although the preferred embodiments of the creation of the present invention have been described in detail above, the creation of the present invention is not limited to the above-described examples. Substitution is possible. All of these equivalent changes and substitutions are included within the scope of the claims of this application.

Claims (6)

A first step of collecting message documents related to typhoon incidents on social media, extracting typhoon incident information, and converting them into a structured information tuple format;
A step of target information convergence based on the similarity of multiple features, in which it is determined whether or not the information tuples correspond to the same target based on the similarity between the target names, and the information tuples expressing the same target are converged. a second step requiring
It is a step of state information convergence based on spatiotemporal features, and selects attribute values and action values that satisfy the requirements of a single time and location condition from the result of converging target information. a third step, wherein the attribute value and the action value together constitute a result of converging the target information in a particular spatio-temporal state;
In the step of process information convergence based on the state, spatio-temporal node information that satisfies the requirements of time and position range is selected from the result of converging the target information. A typhoon incident information convergence method mainly comprising a fourth step of sorting the state information convergence results of to form a process information convergence result in which dynamic features appear. The method of claim 1, wherein in the first step, the typhoon incident information includes object name, time information, location information, attribute information and action information. The typhoon incident information convergence according to claim 1, wherein in the second step, for different information tuples representing the same object, it is necessary to further converge attribute items and action items of the same category. Method. In the first step, extracting typhoon incident information includes at least two parts: identification of information elements and association of information elements;
Identification of information elements involves identifying structural objects in typhoon events, building a classification system, and extracting names and characteristic information representing different classification objects from social media documents. and actions, the attribute information may be further divided into attribute items indicating the classification of attributes, and attribute values indicating the data or data volume possessed by the category attributes, and the action information is the same as the attribute information. ,
Information element association associates feature information with its appearing object and name in the same social media document, forming an information tuple in the format On=<T, L, A, B>, where On is the object The typhoon incident information convergence method according to claim 1, characterized in that name, T is time information, L is location information, A is attribute information, and B is action information. In the second step, using the similarity of word embedding, the similarity between the target name, the attribute item and the action item is determined,
a step S1 of extracting data of all social media documents;
step S2 of training word embedding using the Skip-gram model with the extracted result as a training data set;
Let On1 and On2 be the target names, A1 and A2 be the attribute items, and B1 and B2 be the action items. a step S3 of obtaining embeddings E(On1), E(On2), E(A1), E(A2), E(B1), E(B2);
Using cosine similarity, calculate the similarity values simn, sima and simb between E(On1) given E(On2), E(A1) and E(A2), E(B1) and E(B2) respectively If simn≧εn, sima≧εa, and simb≧εb (where εn, εa, and εb are thresholds), then assertions On1 and On2, A1 and A2, and B1 and B2 have the same object name, attribute item, and The typhoon incident information convergence method according to claim 1, characterized by comprising step S4, which is an action item and finds that the information can be converged correspondingly. In the fourth step, sorting the results of converging a plurality of state information includes:
Step A1 of sorting according to the forward order or reverse order according to the time information of the state;
Step A2 of sorting according to the mode of large size or small size according to the position information of the state;
Based on the attribute information of the state and the action information, sorting may be performed according to the magnitude of the feature value or the grade, or may be sorted according to the degree of similarity with the user's convergence condition. The typhoon incident information convergence method according to claim 1.

Images