A mass incident early warning method and system

Abstract

The present invention provides a method and system for early warning of mass incidents, including: obtaining conflict and dispute incidents of various administrative departments; identifying the subjects involved in the conflict and dispute incidents to obtain a time series of the number of event subjects; building a prediction model based on the time series of the number of event subjects; obtaining a relative error and a standard deviation based on the prediction model; and issuing a mass incident early warning when the relative error exceeds the standard deviation and is a positive value. The present invention can greatly improve the accuracy of early warning by building a prediction model using the time series of the number of event subjects and then issuing an early warning for mass incidents based on the model.

Description

Technical Field

[0001] The present invention relates to the technical field of grassroots social governance, and in particular to a method and system for early warning of mass incidents. Background Art

[0002] The complaint and help platform has become an important channel for the public to express their interests and vent their emotions. Before most mass incidents occur, the public will provide feedback through the help and help platform system. Based on this, comprehensive analysis of public opinion information scattered in various department systems and discovery of risk factors that may trigger mass incidents have important theoretical and practical significance for achieving scientific, timely and efficient early warning of mass incidents. Summary of the invention

[0003] In order to overcome the deficiencies of the prior art, the present invention aims to provide a method and system for early warning of mass incidents.

[0004] The present invention also provides a method for early warning of mass incidents, comprising:

[0005] Step 1: Obtain conflict and dispute events in various administrative departments;

[0006] Step 2: Identify the parties involved in the conflict and dispute event to obtain a time series of the number of parties involved in the event;

[0007] Step 3: Build a prediction model based on the time series of the number of event subjects;

[0008] Step 4: Obtaining relative error and standard deviation according to the prediction model;

[0009] Step 5: When the relative error exceeds the standard deviation and is positive, a mass incident warning is issued.

[0010] Preferably, the step 2: identifying the parties involved in the conflict and dispute event to obtain a time series of the number of parties involved in the event, includes:

[0011] Step 2.1: Perform Chinese word segmentation on the full name of the organization of the subject involved to obtain the word segmentation result;

[0012] Step 2.2: The word segmentation results are labeled according to the institution name keywords, location, institution type, and institution suffix, and a first-order chain conditional random field model is used for learning to obtain the linear chain CRFs conditional probability of the state sequence;

[0013] Step 2.3: Use the linear chain CRFs conditional probability of the state sequence to perform named entity recognition on the event data, and extract and recognize the organization group name and organization name;

[0014] Step 2.4: Use Baidu's word segmentation algorithm to identify the address and name of the subject involved to obtain the corresponding place name and name;

[0015] Step 2.5: Construct a time series of the number of event subjects based on the place names, personal names, organization group names, and institution names of the involved subjects.

[0016] Preferably, the linear chain CRFs conditional probability calculation formula of the state sequence is:

[0017]

[0018] Among them, f k (y i-1 ,y i ,x,i) is the characteristic function,ω k is the weight of the kth feature function, i represents the position of the word currently being marked in the sequence, and Z(x) is the normalization factor.

[0019] Preferably, the step 3: constructing a prediction model according to the time series of the number of event subjects includes:

[0020] A prediction model is established based on the time series of the number of event entities using a simple moving average method; the prediction model is:

[0021]

[0022]

[0023] where x1,...,x T It represents the time series of the number of event entities, and N represents the number of items in the moving average.

[0024] The present invention also provides a mass incident early warning system, comprising:

[0025] The conflict and dispute event acquisition module is used to obtain the conflict and dispute events of various administrative departments;

[0026] A subject identification module is used to identify the subjects involved in the conflict and dispute event to obtain a time series of the number of subject numbers of the event;

[0027] A prediction model building module is used to build a prediction model based on the time series of the number of event subjects;

[0028] A difference calculation module, used to obtain a relative error and a standard deviation according to the prediction model;

[0029] The early warning module is used to issue a mass incident early warning when the relative error exceeds the standard deviation and is a positive value.

[0030] Preferably, the subject identification module includes:

[0031] A word segmentation unit is used to perform Chinese word segmentation on the full name of the organization of the subject involved to obtain a word segmentation result;

[0032] A labeling unit is used to label the word segmentation results according to the institution name keywords, location, institution type, and institution suffix, and use a first-order chain conditional random field model to learn and obtain the linear chain CRFs conditional probability of the state sequence;

[0033] The first recognition unit is used to perform named entity recognition on the event data using the linear chain CRFs conditional probability of the state sequence, and extract and recognize the organization group name and the organization name;

[0034] The second recognition unit is used to use Baidu's word segmentation algorithm to respectively recognize the address and name of the subject involved to obtain the corresponding place name and name;

[0035] The time series construction unit is used to construct a time series of the number of event subjects based on the place names, personal names, organization group names and institution names of the subjects involved.

[0036] Preferably, the linear chain CRFs conditional probability calculation formula of the state sequence is:

[0037]

[0038] Among them, f k (y i-1 ,y i ,x,i) is the characteristic function,ω k is the weight of the kth feature function, i represents the position of the word currently being marked in the sequence, and Z(x) is the normalization factor.

[0039] Preferably, the prediction model building module includes:

[0040] The prediction model building unit is used to establish a prediction model using a simple moving average method according to the time series of the number of event subjects; the prediction model is:

[0041]

[0042] where x1,...,x T It represents the time series of the number of event entities, and N represents the number of items in the moving average.

[0043] The present invention also provides a computer-readable storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps in the above-mentioned method for early warning of mass incidents are implemented.

[0044] The beneficial effect of a group incident early warning method and system provided by the present invention is that: compared with the prior art, the present invention constructs a prediction model by utilizing the time series of the number of event subjects, and then issues early warning for group incidents based on the model, which can greatly improve the accuracy of early warning.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A flow chart of a method for early warning of mass incidents provided by an embodiment of the present invention is shown;

[0048] Figure 2 A schematic diagram of a mass incident early warning system provided by an embodiment of the present invention is shown;

[0049] Figure 3 The figure shows the result of analyzing the number of related events of a certain company using the simple moving average method provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0050] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0051] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0052] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] See also Figure 1 The present invention also provides a method for early warning of mass incidents, comprising:

[0054] Step 1: Obtain conflict and dispute events in various administrative departments;

[0055] Step 2: Identify the parties involved in the conflict and dispute event to obtain a time series of the number of parties involved in the event;

[0056] Furthermore, the step 2 comprises:

[0057] Step 2.1: Perform Chinese word segmentation on the full name of the organization of the subject involved to obtain the word segmentation result;

[0058] Step 2.2: The word segmentation results are labeled according to the institution name keywords, location, institution type, and institution suffix, and a first-order chain conditional random field model is used for learning to obtain the linear chain CRFs conditional probability of the state sequence;

[0059] Step 2.3: Use the linear chain CRFs conditional probability of the state sequence to perform named entity recognition on the event data, and extract and recognize the organization group name and organization name;

[0060] It is described below in conjunction with specific embodiments:

[0061] The identification of organizations is also based on the organization name database of the target area and matched with the existing structured data. In reality, many event descriptions do not contain structured data organization information, and it is more common to use the organization's name to describe the event. Therefore, abbreviation identification is required.

[0062] First, the abbreviation database is automatically generated based on the organization name in the organization name database. According to the organization name keyword (Key-word), location (Loc), organization type (Type), and organization suffix (Suffix), the following eight rules are used to generate a set of candidate abbreviations.

[0063] 1)loc+keyword;

[0064] 2)loc+keyword+type;

[0065] 3)loc+keyword+type+suffix;

[0066] 4)loc+keyword+suffix;

[0067] 5) keyword;

[0068] 6) keyword+type;

[0069] 7) keyword+type+suffix;

[0070] 8) keyword+suffix.

[0071] The present invention performs Chinese word segmentation on the full name of the organization, annotates the word segmentation results according to the organization name keywords, location, organization type, and organization suffix, and then uses a first-order chain conditional random field model (Linear-chain CRFs) for learning.

[0072] Let the input unlabeled word sequence be X = x1, x2, ..., x n , the label sequence is Y = y1, y2, ..., y n The present invention takes "Zhejiang Shenghua Yunfeng New Materials Co., Ltd." as an example, the observation word sequence is represented as X = (Zhejiang, Shenghua, Yunfeng, New Materials, Shares, Limited, Company), and the tag sequence Y = (loc, keyword, keyword, type, suffix, suffix, suffix). Under a given input sequence, the linear chain CRFs conditional probability of its state sequence is:

[0073]

[0074] Z(x) is the normalization factor:

[0075]

[0076] Among them, f k is the characteristic function, with a value of 0 or 1, ω k is the weight of the corresponding characteristic function. k Take the entire observation sequence X and the label y i and i-1 As an independent variable, the context information can be fully utilized as a feature, and non-independent features can be used, thereby enhancing the expressive power of the conditional random field model. Compared with the conditional maximum entropy model, due to the introduction of y in the feature function i-1 , so that the label sequence y can be globally normalized and the global optimal solution can be obtained.

[0077] Here are some examples:

[0078] f1(y i-1 ,y i ,X,i)=1if y i =keyword and y i-1 =loc; otherwise, 0.

[0079] If the weight ω1 of this feature is a large positive number, then this feature indicates that when y i-1 =loc, the present invention prefers to mark it as keyword.

[0080] f2(y i-1 ,y i ,X,i)=1, if i=1, and y i =loc; otherwise, it is 0.

[0081] Similarly, if the weight ω2 of this feature is a large positive number, then it is more likely to label the first participle in the full name of the organization as loc.

[0082] f3(y i-1 ,y i ,X,i)=1, if y i-1 =type and y i =suffix; otherwise, it is 0.

[0083] Similarly, if the weight ω3 of this feature is positive, it indicates that type tends to be followed by suffix.

[0084] Conditional random fields have the ability to express long-distance dependencies and overlapping features, and can solve the problem of "labeling bias" that is difficult to avoid in maximum entropy models. Therefore, in the same sequence labeling task, the performance of conditional random fields is better than that of maximum entropy models. The modeling and use of conditional random fields can be summarized as solving two basic problems:

[0085] I Learning problem: Given a training set Ask for The probability of occurrence is the largest model parameter set Λ.

[0086] II Decoding problem: Given Λ, X, find the most likely tag sequence corresponding to X.

[0087]

[0088] The present invention uses the open source tool CRF++, in which the learning problem adopts the quasi-Newton method BFGS algorithm to optimize the objective function and iteratively learns and solves the training set; and the decoding problem adopts the Viterbi algorithm to realize the labeling of the input sequence.

[0089] Secondly, word segmentation technology is used to perform named entity recognition on the event data, extracting and identifying the organization group name (nt) and organization name (ORG);

[0090] Finally, the extracted organization group name (nt) and organization name (ORG) are checked to see if they are the organization abbreviations in the abbreviation library. If so, the organization involved in the incident will be given the standardized organization name in the abbreviation library; if not, the original identified organization name will be retained.

[0091] Step 2.4: Use Baidu's word segmentation algorithm to identify the address and name of the subject involved to obtain the corresponding place name and name;

[0092] It should be noted that, since the mass events targeted for prediction and warning are within a specific time and space range, the place names extracted by the named entity recognition method can be standardized based on the place names and addresses of the target area. The present invention can perform place name standardization through the information of the local standard place name and address library. First, Baidu word segmentation technology is used to perform named entity recognition on the event data to extract words identified as place names (ns and LOC); secondly, it is determined whether the words identified as place names (ns and LOC) are place names in the place name and address library. If so, the place name of the event is taken as the standardized place name in the standard place name and address library; if not, the originally identified place name is retained. In the case of unstructured data, the name recognition of people (nr and PER) is extracted using Baidu word segmentation technology.

[0093] Step 2.5: Construct a time series of the number of event subjects based on the place names, personal names, organization group names, and institution names of the involved subjects.

[0094] A time series is a sequence of data that is arranged in chronological order and is interrelated. For example: Figure 3 It shows the number of events containing "a certain company" counted by natural week from June 2020 to December 2021, and arranged in chronological order to form a time series.

[0095] Time series analysis is an analysis technique that mines this dependency relationship and predicts the value at a certain moment in the future based on the analysis results. After identifying the subject information of event data from different departments, the present invention further establishes a quantitative model through time series analysis of related events in different time windows on the event subject to determine whether the event subject has the possibility of potential mass incidents.

[0096] Step 3: Build a prediction model based on the time series of the number of event subjects;

[0097] Before each new mass incident occurs, the public has fully expressed their interest demands through institutionalized channels. Moreover, within the T time series before the occurrence of the mass incident, the information of these demands will show a change in the amount of information of the event subject from nothing to something or from less to more as the date of the mass incident approaches. And the present invention mainly constructs a prediction model based on the change in the amount of information of the event subject.

[0098] Step 4: Obtain the relative error and the standard deviation according to the prediction model;

[0099] Step 5: When the relative error exceeds the standard deviation and is positive, a warning of a mass incident is issued.

[0100] Further, steps 3 - 5 include:

[0101] Based on the observed sequence x1,..., x of the number of event subjects identified T , take the number of terms N < T for the moving average, and calculate the simple moving average value. The calculation formula is:

[0102]

[0103] Assume that under normal conditions, the trend of the number of event subject information is fluctuating up and down at a certain level. Thus, a prediction model is established using the simple moving average method as:

[0104]

[0105] Figure 3 In it, the dashed part is the prediction result in the case of N = 6.

[0106] The relative error of the prediction is:

[0107]

[0108] The standard deviation of the prediction is:

[0109]

[0110] By comparing the relative error value of the prediction with the standard deviation, if the error exceeds the standard deviation and is positive, there may be a potential mass incident.

[0111] The present invention constructs a prediction model by using the time series of the number of event subjects, and then warns of mass incidents based on this, which can greatly improve the warning accuracy.

[0112] Please refer to Figure 2 , the present invention also provides a mass incident warning system, including:

[0113] The conflict and dispute event acquisition module is used to obtain the conflict and dispute events of various administrative departments;

[0114] A subject identification module is used to identify the subjects involved in the conflict and dispute event to obtain a time series of the number of subject numbers of the event;

[0115] A prediction model building module is used to build a prediction model based on the time series of the number of event subjects;

[0116] A difference calculation module, used to obtain a relative error and a standard deviation according to the prediction model;

[0117] The early warning module is used to issue a mass incident early warning when the relative error exceeds the standard deviation and is a positive value.

[0118] Preferably, the subject identification module includes:

[0119] A word segmentation unit is used to perform Chinese word segmentation on the full name of the organization of the subject involved to obtain a word segmentation result;

[0120] A labeling unit is used to label the word segmentation results according to the institution name keywords, location, institution type, and institution suffix, and use a first-order chain conditional random field model to learn and obtain the linear chain CRFs conditional probability of the state sequence;

[0121] The first recognition unit is used to perform named entity recognition on the event data using the linear chain CRFs conditional probability of the state sequence, and extract and recognize the organization group name and the organization name;

[0122] The second recognition unit is used to use Baidu's word segmentation algorithm to respectively recognize the address and name of the subject involved to obtain the corresponding place name and name;

[0123] The time series construction unit is used to construct a time series of the number of event subjects based on the place names, personal names, organization group names and institution names of the subjects involved.

[0124] Preferably, the linear chain CRFs conditional probability calculation formula of the state sequence is:

[0125]

[0126] Among them, f k is the characteristic function, ω k is the weight of the corresponding feature function, i represents the position of the word currently to be marked in the sequence, and Z(x) is the normalization factor.

[0127] Preferably, the prediction model building module includes:

[0128] The prediction model building unit is used to establish a prediction model using a simple moving average method according to the time series of the number of event subjects; the prediction model is:

[0129]

[0130] where x1,...,x T It represents the time series of the number of event entities, and N represents the number of items in the moving average.

[0131] The present invention also provides a computer-readable storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps in the above-mentioned method for early warning of mass incidents are implemented.

[0132] Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the embodiment of the present invention are the same as the beneficial effects of the mass incident early warning method in the above technical solution, and are not described in detail here.

[0133] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technical solution that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for early warning of mass incidents, characterized in that: include: Step 1: Obtain conflict and dispute events in various administrative departments; Step 2: Identify the parties involved in the conflict and dispute event to obtain a time series of the number of parties involved in the event; The step 2: identifying the parties involved in the conflict and dispute event to obtain a time series of the number of parties involved in the event, including: Step 2.1: Perform Chinese word segmentation on the full name of the organization of the subject involved to obtain the word segmentation result; Step 2.2: The word segmentation results are labeled according to the institution name keywords, location, institution type, and institution suffix, and the first-order chain conditional random field model is used for learning to obtain the linear chain CRFs conditional probability of the state sequence; wherein, the linear chain CRFs conditional probability calculation formula of the state sequence is: Among them, f k (y i-1 ,y i ,X,i) is the characteristic function,ω k is the weight of the kth feature function, i represents the position of the word to be marked in the sequence, and Z(x) is the normalization factor; Step 2.3: Use the linear chain CRFs conditional probability of the state sequence to perform named entity recognition on the event data, and extract and recognize the organization group name and organization name; Step 2.4: Use Baidu's word segmentation algorithm to identify the address and name of the subject involved to obtain the corresponding place name and name; Step 2.5: Construct a time series of the number of event subjects based on the place names, personal names, organization group names, and institution names of the involved subjects; Step 3: Build a prediction model based on the time series of the number of event subjects; The step 3: constructing a prediction model based on the time series of the number of event subjects, including: A prediction model is established based on the time series of the number of event entities using a simple moving average method; the prediction model is: where x1,...,x T represents the time series of the number of event entities, and N represents the number of items in the moving average; Step 4: Obtaining relative error and standard deviation according to the prediction model; The relative error is: The standard deviation is: Step 5: When the relative error exceeds the standard deviation and is positive, a mass incident warning is issued.

2. A mass incident early warning system, characterized in that: include: The conflict and dispute event acquisition module is used to obtain the conflict and dispute events of various administrative departments; A subject identification module is used to identify the subjects involved in the conflict and dispute event to obtain a time series of the number of subject numbers of the event; The subject identification module includes: A word segmentation unit is used to perform Chinese word segmentation on the full name of the organization of the subject involved to obtain a word segmentation result; The tagging unit is used to tag the word segmentation results according to the institution name keywords, location, institution type, and institution suffix, and adopt a first-order chain conditional random field model to learn and obtain the linear chain CRFs conditional probability of the state sequence; wherein the linear chain CRFs conditional probability calculation formula of the state sequence is: Among them, f k (y i-1 ,y i ,x,i) is the characteristic function,ω k is the weight of the kth feature function, i represents the position of the word to be marked in the sequence, and Z(x) is the normalization factor; The first recognition unit is used to perform named entity recognition on the event data using the linear chain CRFs conditional probability of the state sequence, and extract and recognize the organization group name and the organization name; The second recognition unit is used to use Baidu's word segmentation algorithm to respectively recognize the address and name of the subject involved to obtain the corresponding place name and name; A time series construction unit is used to construct a time series of the number of event subjects based on the place names, personal names, organization group names, and institution names of the subjects involved; A prediction model building module is used to build a prediction model based on the time series of the number of event subjects; The prediction model building module includes: The prediction model building unit is used to establish a prediction model using a simple moving average method according to the time series of the number of event subjects; the prediction model is: where x1,...,x T represents the time series of the number of event entities, and N represents the number of items in the moving average; A difference calculation module, used to obtain a relative error and a standard deviation according to the prediction model; The relative error is: The standard deviation is: The early warning module is used to issue a mass incident early warning when the relative error exceeds the standard deviation and is a positive value.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for early warning of mass incidents as claimed in claim 1 are implemented.

Images