CN115203365A - Social event processing method applied to comprehensive treatment field - Google Patents

Abstract

The invention relates to a social event processing method applied to the field of comprehensive treatment. The method solves the problem that the urban social event processing efficiency is low in the prior art. The method comprises the steps of S1, adopting an RE2 model to report event content, carrying out similarity matching with an event content text in a case base, meeting events of a threshold matching condition, and entering S3 to carry out allocation processing by referring to the events matched with the case base; s2, identifying the event which does not meet the threshold matching condition according to a primary component/event open type classification model; s3, transmitting the event from the event in the S1 or the event judged as the first-level component/event class in the S2 to an event acceptance center, and manufacturing a unified interface and a data structure of the access event; and comparing the similarity of the received event and the portrait in the secondary component/event management portrait library by using an RE2 model, and ensuring the accuracy of event division. The invention has the advantages that: the social event processing efficiency is effectively improved.

Description

A social event processing method applied in the field of comprehensive treatment

technical field

The invention relates to the technical field of social governance, in particular to a social event processing method applied to the field of comprehensive governance.

Background technique

The incident handling process in the field of comprehensive management generally includes the acceptance of incidents, response to handling results, return visits and satisfaction surveys. The scope of incidents handled includes infrastructure, city appearance, market supervision, street security, community management, rural management and other social aspects. , the departments handling incidents include municipal supervisors, economic and information, administrative law enforcement, insurance, communications and other departments and units, as well as gas, water, telecommunications, mobile and other companies, and even in some cases, reported social incidents require cross-departmental coordination, and In the case of multi-department collaborative processing, it is necessary to clarify the main responsible department, and the main responsible department will follow up and supervise the processing status of the incident.

At present, the handling of various social events has the following problems:

1. The incident reporting channel is not smooth, and I do not know how to report the problem after finding the problem;

2. The business acceptance and disposal procedures are not standardized, each department fights its own way, and there is no unified method for assigning and accepting incidents, which leads to the frequent occurrence of mutual shirk in the process of incident handling and ending without a problem;

3. There is no classification and classification of events, resulting in the inability to clarify the priorities between events when dealing with multiple events;

4. There is a lack of necessary incident supervision and feedback, the person in charge does not know the progress of the incident handling, and the public does not understand the handling result of the incident.

Most of the solutions for dealing with social problems in the above-mentioned existing technologies focus on a specific link in the process, and the process is relatively simple, without classifying, grading, and diverting events, and at this stage, many relevant departments have their own unique set of procedures. The diversity of the system brings the incompatibility of events, which affects the real-time problem solving and reduces the efficiency of social event processing.

SUMMARY OF THE INVENTION

The purpose of the present invention is to address the above problems, and provide a social event processing method applied in the field of comprehensive treatment with reasonable design and high processing efficiency.

In order to achieve the above object, the present invention adopts the following technical solutions: a social event processing method applied to the comprehensive treatment field, characterized in that the method specifically comprises the following steps:

S1. Use the RE2 model to report the event content and perform similarity matching with the event content text in the case database. For events that meet the threshold matching conditions, refer to the events matched by the case database and enter S3 for dispatch processing; report the event through the RE2 model and match the case with the case. Cases in the database are assigned by comparison and comparison, so as to determine the event category, which can effectively improve the accuracy of data.

S2. Events that do not meet the threshold matching conditions are identified according to the first-level component/event open classification model, that is, the Open-set Text Classification model;

S3. Events from events in S1 or events determined as first-level components/events in S2 are transmitted to the event acceptance center to create a unified interface and data structure for accessing events; the unified interface and data structure can enable the reception of social events to form Unified and easy to manage.

S4. According to the primary component/event category of the current event, call up all secondary components/event management portraits under the current primary component/event category from the secondary component/event management portrait library, and use the RE2 model to detect the current event The similarity between the content and the problem description text in the secondary component/event management; the RE2 model is used to compare the similarity of the problem description text between the received event and the portrait in the secondary component/event management portrait library to ensure the accuracy of event division .

S5. Create a virtual network for the event, and according to the secondary component/event management portrait, add the responsible department to the grid as a grid member, and judge whether the current event is a people's livelihood event according to the two-class algorithm model, and if it is a people's livelihood event, then Extract the party information from the current event by means of named entity extraction and add it to the virtual grid;

S6. Return visit and satisfaction survey on the event processing results. Satisfaction survey is conducive to optimizing the way of handling social events, and storing the way of handling social events with high satisfaction is convenient for reference.

In the above-mentioned social event processing method applied in the field of comprehensive treatment, in step S1, the establishment of the RE2 model specifically includes the following steps:

S11. The input includes the current event event1 and the case library, and the following steps are performed for each event content event2 of the case library:

S12. Convert the text event1 and event2 into vectors in the form of embedding;

S13. Send the embedding vector sequence into 3 blocks for continuous processing. Each block has the same structure and consists of an encoder layer, an alignment layer and a fusion layer. The encoder encoding layer calculates the context features of the sequence, and the alignment layer The correlation feature calculation is performed on the text event1 and event2, and the output of the fusion fusion layer is the input of the next block block or the input of the pooling pooling layer;

S14. The pooling pooling layer converts the output result of the block block into a fixed-length vector;

S15, after calculating the features of event1 and event2, a binary classification model is constructed in the prediction layer. Category 0 indicates that event1 and event2 are not similar, and category 1 indicates that event1 and event2 are similar.

In the above-mentioned social event processing method applied in the field of comprehensive treatment, in step S2, there are 5 first-level component categories in the first-level component/event open classification model, which are: public facilities, road traffic Facilities, city appearance and environment implementation, landscaping facilities and expansion components; there are 10 first-level event categories, which are: city appearance and environment, publicity and advertising, construction management, street order, street security, market supervision, emergencies, community Management, rural management and expansion events, and the total number of secondary components/event categories is 235-245; the secondary component/event management portrait library is established for the secondary components/events respectively. The content of the secondary component/event management portrait library includes: First-level component/event category, responsible department, problem description First-level disposal requirements.

In the above-mentioned social event processing method applied to the field of comprehensive treatment, in step S3, a first-level component/event classification model is established through the Open-set Text Classification algorithm, and the number of categories of the first-level component/event classification model is 12 +1,12 is a well-defined and fixed classification category, which includes the above-mentioned 10 first-level event categories and 5 first-level component categories, of which 1 is an open category, which includes new, potential component/event categories; unified The interface and data structure include unified interface request method, data encryption method, department area code and callback method.

In the above-mentioned social event processing method applied in the field of comprehensive treatment, the Open-set TextClassification algorithm is built based on CNN, and during the test, the last layer of CNN extracts the sentence features of the input class to obtain the Activation Vector or AV, and then Calculate the first k distance values from the Activation Vector of each test sample to each known class, and record the Activation Vector of the first k samples as k-ClassActivation Vectors or k-CAV, and calculate the test sample AV The distance from k-CAV, the sum of k probability values of each training class is recorded as total closed set probability, and finally OSP=1-total closed set probability is calculated, and the maximum total closed set probability of all training classes is compared with OSP to determine The test class is an unknown class or a known class.

In the above-mentioned social event processing method applied in the field of comprehensive treatment, in step S4, the responsible department of the portrait that satisfies the maximum threshold condition is the recommended processing department for the event. Personnel can determine whether to create a secondary component/event management portrait for this type of event.

In the above-mentioned social event processing method applied to the field of comprehensive treatment, in step S5, the establishment of the two-classification algorithm model specifically includes the following steps:

S51, the content other than the text information in the event text forms a classification training corpus S, and manually annotates the predictor variables of the corpus S, denoting that y _i =+1 represents a people's livelihood event, and y _i =-1 represents a non-people's livelihood event;

S52. Divide the sentence set S into Strain and Stest according to 8:2, and calculate each sentence in Strain and Stest according to the following formula:

S53. Combine all sentence vectors in Strain to obtain a matrix X, and calculate the first principal component u of X by TruncatedSVD;

S54, perform the following calculation on each sentence vector Vs in Strain and Stest: V _s =V _s -u*u ^T *v _s ;

和

其中y_i∈{+1，-1}，且 i∈(1，2，…，m)∪(1，2，…，n)，通过支持向量机SVM分类算法和贝叶斯参数优化器训练并评估模型。S55, for

and

where y _i ∈ {+1, -1}, and i ∈ (1, 2,…,m)∪(1,2,…,n), trained by SVM classification algorithm and Bayesian parameter optimizer and evaluate the model.

In the above-mentioned social event processing method applied in the field of comprehensive treatment, in step S6, the results of the satisfaction survey are divided into: very satisfied, satisfied, basically satisfied, dissatisfied and very dissatisfied, and the satisfaction result is Very satisfied and satisfied events are included in the event processing case library, and future events will quickly recommend relevant collaborative departments and processing procedures to users based on the historical data of the event processing case library.

In the above-mentioned social incident processing method applied in the field of comprehensive treatment, the incident acceptance center includes the citizen service hotline, petition and other relevant industry management information system incident sources, regulates the data structure of accessible incidents, and allows receiving basic The extra field outside the field and provides a unified way to call back the event handling process. Provide a variety of social incident acceptance methods, which facilitates timely feedback of social incidents and enables various sources to process information synchronously.

In the above-mentioned social event processing method applied in the field of comprehensive governance, the virtual network sets the first person in charge of the event or the responsible unit as the grid length. With the flow of social events and the increase of personnel in the processing department, the grid automatically Growth, the grid leader is responsible for coordinating related resources and tracking the eventual resolution of the incident. For incidents involving people's livelihood issues, the parties involved in the incident are added to the virtual grid to understand and monitor the progress of the entire incident through PC and mobile terminals, and evaluate after the incident is resolved.

Compared with the existing technology, the present invention has the advantages of reasonable design, high efficiency of social event processing, and strengthening the lead coordination of social event problems that need to be solved collaboratively by multiple departments, and defines a secondary classification system for urban components and events. , defines the components and event management portraits of sub-categories, and recommends methods based on the combination of RE2 and manual assignment; an event acceptance center is established to accept events from different departmental systems, and at the same time as a grass-roots system to coordinate with higher-level applications The channel for resources; two concepts, virtual grid and command room, are designed, and the person in charge of the incident and the people involved in the people's livelihood incident are regarded as grid members. For the events that need to be handled collaboratively across departments, an online command room is launched and the relevant departments are invited. Enter the command room for multi-span collaborative processing; according to the return visit and satisfaction survey of the processing results, save the processing process of the favorable event into the case model database, and when the event occurs in the future, it will quickly recommend the relevant collaborative departments to users based on the historical data of the case database. The processing flow effectively improves the processing efficiency of social events.

Description of drawings

Figure 1 is a flow chart of the present invention.

Detailed ways

The present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

As shown in Figure 1, a social event processing method applied in the field of comprehensive treatment, the method specifically includes the following steps:

S1. Use the RE2 model to report the event content and perform similarity matching with the event content text in the case library. Events that meet the threshold matching conditions enter S3 for dispatch processing with reference to the events matched by the case library;

S2. Events that do not meet the threshold matching conditions are identified according to the first-level component/event open classification model, that is, the Open-set Text Classification model;

S3. The event from the event in S1 or the event determined as a first-level component/event class in S2 is transmitted to the event acceptance center, and a unified interface and data structure for accessing the event are created;

S4. According to the primary component/event category of the current event, call up all secondary components/event management portraits under the current primary component/event category from the secondary component/event management portrait library, and use the RE2 model to detect the current event Similarity between content and problem description text in secondary component/incident management;

S5. Create a virtual network for the event, and according to the secondary component/event management portrait, add the responsible department to the grid as a grid member, and judge whether the current event is a people's livelihood event according to the two-class algorithm model, and if it is a people's livelihood event, then Extract the party information from the current event by means of named entity extraction and add it to the virtual grid;

S6. Return visit and satisfaction survey on the event processing results.

In step S1, the establishment of the RE2 model specifically includes the following steps:

S11. The input includes the current event event1 and the case library, and the following steps are performed for each event content event2 of the case library:

S12. Convert the text event1 and event2 into vectors in the form of embedding;

S13. Send the embedding vector sequence into 3 blocks for continuous processing. Each block has the same structure and consists of an encoder layer, an alignment layer and a fusion layer. The encoder encoding layer calculates the context features of the sequence, and the alignment layer The correlation feature calculation is performed on the text event1 and event2, and the output of the fusion fusion layer is the input of the next block block or the input of the pooling pooling layer;

S14. The pooling pooling layer converts the output result of the block block into a fixed-length vector;

S15, after calculating the features of event1 and event2, a binary classification model is constructed in the prediction layer. Category 0 indicates that event1 and event2 are not similar, and category 1 indicates that event1 and event2 are similar.

In step S2, there are 5 first-level component categories in the first-level component/event open classification model, which are: public facilities, road traffic facilities, city appearance environment implementation, landscaping facilities, and extended components; first-level event categories There are 10, which are: city appearance and environment, publicity and advertising, construction management, street order, street security, market supervision, emergencies, community management, rural management and expansion events, and the total number of secondary components/event categories is 235-245; establish a second-level component/event management portrait library for the second-level component/event. The content of the second-level component/event management portrait library includes: the first-level component/event category, responsible department, and problem description. First-level disposal requirements . The first-level components include urban infrastructure in the public area of urban management, and the first-level components include two levels, the first level is a major category, and the second level is a sub-category, each major category is composed of several sub-categories; and The first-level event is related to human behaviors, or the phenomenon and behavior that the urban appearance, environment and order are affected or destroyed due to natural factors. Small class composition.

In step S3, a first-level component/event classification model is established by the Open-set Text Classification algorithm. The number of categories of the first-level component/event classification model is 12+1, and 12 is a clear and fixed classification category, which includes the above 10 1 first-level event category and 5 first-level component categories, of which 1 is an open category, which includes new and potential component/event categories; unified interface and data structure includes unified interface request method, data encryption method, department area Encoding and callback methods. With the continuous expansion of the scale of urban development and the emergence of new things, part of the management content needs to be expanded and improved. The principle of expansion is: according to the method of immutable major categories and extensible minor categories, new components and events will be added in the under the Extended Parts and Extended Events categories.

Further, the Open-set Text Classification algorithm is based on CNN, and when testing, the last layer of CNN extracts the sentence features of the input class to obtain the Activation Vector, ie AV, and then calculates the Activation Vector of each test sample to each Know the first k distance values with the closest class distance, and record the Activation Vector of the first k samples as k-Class Activation Vectors, namely k-CAV, calculate the distance between the test sample AV and k-CAV, and the k of each training class. The sum of the probability values is recorded as the total closed set probability. Finally, OSP=1-total closed set probability is calculated, and the maximum total closed set probability of all training classes is compared with the OSP to determine whether the test class is an unknown class or a known class.

In step S4, the responsible department for the portrait that satisfies the maximum threshold condition is the recommended processing department for the event. If the threshold condition cannot be met, it will be manually assigned, and the staff can determine whether to create a secondary component/event management portrait for this type of event.

In step S5, the establishment of the binary classification algorithm model specifically includes the following steps:

S51. The content other than the text information in the event text forms a classification training corpus S, and manually annotates the predictor variables of the corpus S, denoting that y _i =+1 represents a people's livelihood event, and y _i =-1 represents a non-people's livelihood event;

S52. Divide the sentence set S into Strain and Stest according to 8:2, and calculate each sentence in Strain and Stest according to the following formula:

S53. Combine all sentence vectors in Strain to obtain a matrix X, and calculate the first principal component u of X by TruncatedSVD;

S54, perform the following calculation on each sentence vector Vs in Strain and Stest: v _s =v _s -u*u ^T *v _s ;

和

其中y_i∈{+1，-1}，且i∈(1，2，…，m)∪(1，2，…，n)，通过支持向量机SVM分类算法和贝叶斯参数优化器训练并评估模型。此外，本发明提供一种扩展方法，允许授权的第三方系统接入事件处理方法中，便于不同层级的处理部门、不同的事件类型定制化地丰富本部门的处理流程。为实现该目的，本发明设计了一套扩展接口及接入认证方案：S55, for

and

where y _i ∈ {+1, -1}, and i ∈ (1, 2,…,m)∪(1,2,…,n), trained by SVM classification algorithm and Bayesian parameter optimizer and evaluate the model. In addition, the present invention provides an extension method that allows an authorized third-party system to access the event processing method, which is convenient for processing departments at different levels and different event types to customize and enrich the processing flow of the department. In order to achieve this purpose, the present invention designs a set of extension interface and access authentication scheme:

1. The system provides framework specifications in accordance with the unified interface style, unified interface calling, and unified authority control. Third parties can seamlessly access their own business systems by developing and docking according to the specifications;

2. The system adopts the form of front-end and back-end separation. The front-end provides js script plug-ins. After the plug-in is introduced on the third-party page, it will automatically standardize the page style, insert the embedded point, and access the back-end interface; the developed front-end page is embedded into the system After that, a dynamic login token can be obtained during the opening process;

3. The system provides back-end sdk of commonly used multiple development languages, encapsulates the security encryption logic and open interface, and the third party only needs to configure the host, key and secret key to easily access and call the interface provided by the system; The back-end of the third party can obtain the current logged-in user, permissions, and business information in real time according to the dynamic token obtained by the front-end, so as to continue the internal business process of the third party;

4. Provide encryption logic descriptions for development and access by development languages that do not provide sdk support;

Based on the above solutions, the present invention not only achieves high expansion flexibility, but also has relatively strict normativeness.

In detail, in step S6, the results of the satisfaction survey are divided into: very satisfied, satisfied, basically satisfied, dissatisfied, and very dissatisfied, and events whose satisfaction results are very satisfied and satisfied are included in the event handling case database. In the event of an event, it can quickly recommend relevant collaborative departments and processing procedures to users based on the historical data of the event processing case library. Assist users to quickly create an effective command room and solve problems as soon as possible; and the command room invites responsible units in the same administrative region or different administrative regions to establish a multi-department collaborative processing command center for cross-department or cross-administrative area processing events to reduce wrangling and shirk. phenomenon, promote the smooth resolution of social incidents, and improve the processing efficiency of social incidents; each social incident handled by the command room will be automatically classified by the system, and saved into the case database in combination with the processing process of the collaborative departments in the command room. The event will use the recommendation algorithm to quickly recommend relevant collaborative departments and processing procedures to users based on the historical data of the case library, helping users to quickly create an effective command room.

Furthermore, the event acceptance center includes the citizen service hotline, petition and other related industry management information system event sources, standardizes the data structure of accessible events, allows to receive additional fields beyond the basic fields, and provides a unified way to transfer events. The processing procedure is called back.

Preferably, the virtual network sets the first person in charge of the event or the unit in charge as the grid leader. With the flow of events and the increase of processing department personnel, the grid automatically grows, and the grid leader is responsible for coordinating related resources and tracking the eventual finality solve. For social events involving people's livelihood issues, the relevant parties of the social event are added to the virtual grid, and the progress of the entire event can be understood and monitored through the PC terminal and mobile terminal, and the evaluation will be made after the event is resolved.

To sum up, the principle of this embodiment is: by building an event acceptance center with a unified interface and data structure for accessing events to access social events, and intelligently recommend the processing department of the social event, which is convenient for staff to quickly distribution, and the actual distribution results will be used as training materials for reverse training of the recommendation algorithm to improve the accuracy of the recommendation;

Secondly, establish a virtual network in the process of social event processing, and initiate an online command room as required to apply for multi-department collaboration. Relevant personnel can participate in the collaborative processing process of the entire event through the PC terminal or mobile terminal, and the final processing result will be fed back in real time. For all relevant personnel on the incident processing process line, in order to meet the real-time and efficient communication in the command room, and to meet the smooth video and voice collaborative communication, build a dedicated hardware server for the command room;

Finally, after the social event processing is completed, a satisfaction survey is carried out, and the events with the satisfaction result of being very satisfied and satisfied are included in the event processing case database. The recommendation algorithm will be automatically trained according to new cases on a regular basis, and new models will be generated to improve the event processing method. With the highest recommendation accuracy, social events that appear in the future can quickly recommend relevant collaborative departments and processing procedures to users based on the historical data of the event processing case database.

The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which the present invention pertains can make various modifications or additions to the described specific embodiments or substitute in similar manners, but will not deviate from the spirit of the present invention or go beyond the definitions of the appended claims range.

Claims (10)

1. A social event processing method applied to the field of comprehensive treatment is characterized by specifically comprising the following steps:

s1, reporting event content by adopting an RE2 model, performing similarity matching with an event content text in a case library, and entering S3 for allocation processing by referring to an event matched with the case library when the event meets a threshold matching condition;

s2, identifying the event which does not meet the threshold matching condition according to a primary component/event Open type Classification model, namely an Open-set Text Classification model;

s3, transmitting the event from the event in the S1 or the event judged as the first-level component/event class in the S2 to an event acceptance center, and manufacturing a unified interface and a data structure of the access event;

s4, calling out all secondary component/event management images under the current primary component/event type from a secondary component/event management image library according to the primary component/event type of the current event, and detecting the similarity between the content of the current event and a problem description text in secondary component/event management by adopting an RE2 model;

s5, creating a virtual network for the event, managing the portrait according to the secondary component/event, adding the responsibility department into the grid as a grid member, judging whether the current event is a civil event or not according to a two-classification algorithm model, and if the current event belongs to the civil event, extracting information of a party from the current event by adopting a named entity extraction mode and adding the information into the virtual grid;

and S6, event processing result return visit and satisfaction survey.

2. The social event processing method applied to the comprehensive treatment field according to claim 1, wherein in the step S1, the establishing of the RE2 model specifically comprises the following steps:

s11, inputting a current event1 and a case base, and executing the following steps for each event content event2 of the case base:

s12, converting the texts event1 and event2 into vectors in an embedding form;

s13, the embedding vector sequences are sent into 3 block blocks to be continuously processed, each block has the same structure and consists of an encoder layer, an alignment layer and a fusion layer, the encoder layer calculates the context characteristics of the sequences, the alignment layer calculates the correlation characteristics of the texts event1 and event2, and the output of the fusion layer is the input of the next block or the input of the posing pooling layer;

s14, converting the result output by the block into a vector with a fixed length by the posing pooling layer;

and S15, after calculating features of the event1 and the event2, constructing a two-classification model at a prediction layer, wherein the category 0 indicates that the event1 is not similar to the event2, and the category 1 indicates that the event1 is similar to the event 2.

3. The method for processing social events applied to the comprehensive treatment field according to claim 1, wherein in step S2, there are 5 primary component categories in the primary component/event open-type classification model, which are respectively: public facilities, road transportation facilities, city appearance environment enforcement, landscaping facilities, and expansion components; there are 10 primary event categories, which are: city appearance environment, advertising, construction management, street order, street security, market supervision, emergency, cell management, rural management and extended events, and the total number of second-level components/event categories is 235-245; establishing a secondary component/event management representation library for the secondary component/event respectively, the contents of the secondary component/event management representation library comprising: class of belonging primary component/event, department of responsibility, problem description primary disposition requirement.

4. The social event processing method applied to the comprehensive field according to claim 1, wherein in step S3, a primary component/event Classification model is established through an Open-set Text Classification algorithm, the category number of the primary component/event Classification model is 12+1,12 is a definite and fixed Classification category, which includes the above 10 primary event categories and 5 primary component categories, wherein 1 is an Open category, which includes new and potential component/event categories; the unified interface and data structure comprises a unified interface request mode, a data encryption mode, a department area coding mode and a callback mode.

5. The social event processing method applied to the field of comprehensive treatment as claimed in claim 4, wherein the Open-set Text Classification algorithm is built based on CNN, and when a test is performed, the last layer of CNN extracts sentence features of an input Class to obtain an Activation Vector, namely AV, then calculates the distance from the Activation Vector of each test sample to the first k nearest distance of each known Class, records the Activation Vectors of the first k samples as k-Class Activation Vectors, namely k-CAV, calculates the distance between the test sample AV and the k-CAV, records the sum of k probability values of each training Class as total closed set probability, finally calculates OSP =1-total closed probability, and compares the maximum total closed set probability of all training classes with OSP to determine that the test Class is an unknown Class or a known Class.

6. The method as claimed in claim 5, wherein in step S4, the responsibility department of the image satisfying the maximum threshold condition is the recommended processing department of the event, and if the threshold condition cannot be satisfied, the responsibility department is manually assigned, so that the staff can determine whether to create the secondary part/event management image of the event.

7. The social event processing method applied to the comprehensive treatment field according to claim 6, wherein in the step S5, the establishing of the two-classification algorithm model specifically comprises the following steps:

s51, forming a classified training corpus S by contents except the text information in the event text, manually marking the predictive variable of the corpus S, and recording y _i =1 denotes a civil event, y _i = -1 denotes non-civil event;

s52, according to the sentence set S, the sentence set S is set according to the following steps of 8:2 into Strain and Stest, calculated for each sentence in Strain and Stest according to the following formula:

s53, combining all sentence vectors in the string to obtain a matrix X, and calculating a first principal component u of the X through a Trun catedSVD;

s54, performing the following calculation on each sentence vector Vs in the string and the Stest: v. of _s ＝v _s -u*u ^T *v _s ；

S55, for

And

wherein y is _i Belongs to { +1, -1}, and i belongs to (1, 2, \8230;, m) U (1, 2, \8230;, n), and the model is trained and evaluated through a Support Vector Machine (SVM) classification algorithm and a Bayes parameter optimizer.

8. The social event processing method applied to the comprehensive treatment field according to claim 1, wherein in step S6, the result of the satisfaction survey is divided into: very satisfactory, substantially satisfactory, unsatisfactory and very unsatisfactory, with the satisfaction result being very satisfactory and the events being satisfactory being included in a case base, with future occurrences of the events being based on

And quickly recommending related cooperative departments and processing flows to the user by the historical data of the case base.

9. The method as claimed in claim 8, wherein the event processing center includes a citizen service hotline, a petition and other event sources of related industry management information system, and the data structure of the accessible event is standardized, and the method allows receiving additional fields outside the basic field and provides a uniform way to call back the event processing procedure.

10. The social event processing method applied to the comprehensive treatment field as claimed in claim 9, wherein the virtual network sets a first responsible person or responsible unit of the event as a grid length, the grid automatically grows as the number of the event circulation and the personnel of the treatment department increases, and the grid length is responsible for coordinating related resources and tracking the final solution of the event. For the event related to the civil problem, the event related parties are added into the virtual grid, the progress of the whole event is known and monitored in a mode of a PC (personal computer) end and a mobile end, and evaluation is carried out after the event is solved.

Images