CN114328907A - Natural language processing method for early warning risk upgrade event - Google Patents

Abstract

The invention provides a natural language processing method for early warning of risk upgrade events, which comprises the following steps: s1, establishing a data warehouse, and accessing a data source to the data warehouse; s2, based on natural language processing, extracting and processing event information in the data source through an event analysis related model; and S3, inputting the result value of the event analysis related model into the early warning analysis model, and carrying out risk value analysis and early warning of risk upgrading event upgrading. According to the method, various elements in the event are scientifically and efficiently analyzed through a natural language processing technology, a risk analysis model is obtained through massive event data training, the risk of converting a certain event into a high-risk case can be accurately obtained, and related workers can intervene in the event in advance, resolve contradiction disputes and prevent the high-risk case from occurring.

Description

Natural language processing method for early warning of risk escalation events

technical field

The invention belongs to the technical field of early warning risk escalation events, and particularly relates to a natural language processing method for early warning risk escalation events.

Background technique

At present, with the development of society and economy, civil cases caused by family and marital disputes, neighbor disputes, economic compensation disputes, labor disputes, and trivial disputes in daily work and life are becoming more and more frequent. If not properly guided, with the accumulation of contradictions, high-risk cases will often be triggered, causing huge losses to people's lives and property. In the past, there was no intelligent and efficient method to analyze the disputes and civil cases that have occurred, and draw the risk of the incident escalating into a high-risk case.

Natural language processing integrates linguistics, computer science, and mathematics. It can perform entity extraction, automatic summarization, text classification, etc. on text. In civil incidents, through natural language processing, the incident can be classified, and the relevant personnel, personnel and property losses, personnel emotions and other elements of the incident can be extracted. Based on this, a scientific and accurate analysis can be made on whether the incident will be further escalated. At the same time, by tracking The entire life cycle of events to refine the algorithm model. There is no method in the prior art to analyze disputes and civil cases by means of natural language processing, and obtain a method for escalating the event into a high-risk case.

SUMMARY OF THE INVENTION

The purpose of the present invention is to address the above problems, and provide a natural language processing method for early warning of risk escalation events, which can more accurately determine the risk of a certain event turning into a high-risk case, facilitate the resolution of conflicts and disputes, and prevent the occurrence of high-risk cases. .

To achieve the above object, the present invention has adopted the following technical solutions:

A natural language processing method for early warning of risk escalation events, including the following steps:

S1. Establish a data warehouse and connect the data source to the data warehouse;

S2. Based on natural language processing, extract and process the event information in the data source through the event analysis related model;

S3. Input the result value of the event analysis-related model into the early warning analysis model, and perform risk escalation event escalation risk value analysis and early warning. The present invention scientifically and efficiently analyzes various elements in events through natural language processing methods, combined with a risk analysis model trained with massive event data, can give early warning to risk escalation events, and relevant personnel can intervene in events in advance to resolve conflicts. Disputes, preventing and reducing the occurrence of high-risk cases.

Further, S1 specifically includes:

S101. Access a multi-channel data source;

S102. Clean the data according to the unified format, generate unified structured data, and write it into the database. The present invention obtains event data information from various channels, and combines training with massive event data to obtain a risk analysis model, and the generated structured data is stored in the same database or distributed in multiple databases.

Further, data sources include event data in various fields, including: citizen hotlines, hot events, non-police police information, various help hotlines, people's mediation data, and conflict mediation center data. The data resource library established by the present invention accesses event data in various fields, and provides data basis and basis for subsequent event classification and analysis.

Further, the event analysis related models include an event classification model that can dynamically configure event types according to business needs, an entity extraction model for extracting event locations, organizations, and personnel, and a property loss model for calculating property losses. A casualty model for , and a popular sentiment model for calculating the intensity of popular sentiment.

Further, S2 includes:

S201, classifying events according to the event information, dividing the events into multiple major categories by using an event classification model, and setting multiple subcategories under each major category;

S202, sending the classified results into the entity extraction model for entity extraction, and extracting the features of texts and categories that conform to comprehensive social governance;

S203. Carry out preset keyword collision according to the different contents of the entity extraction result, calculate the score through the property loss model and the casualty model, enter the full text into the public sentiment model, and obtain the intense public sentiment of the risk escalation event through analysis and calculation degree;

Among them, the corresponding model calculation results are passed down through interfaces between each model.

Further, the event classification classifies events through a hybrid algorithm combining the K-means algorithm and the Bayesian network: the K-means algorithm is used to train sample clusters, and a manual intervention interface is provided for manual dynamic intervention to configure the training results. As a result, events are classified using a Bayesian network. The present invention can manually intervene and dynamically configure the training results.

Further, before step S202, the entity extraction model is fed with corpus in advance: a large number of events related to comprehensive social governance are sorted and summarized, and then trained. Correction to obtain an entity extraction model that can extract features of texts and categories that fit social comprehensive governance.

Further, the escalation risk value analysis of the risk escalation event calculates the risk value according to the risk value algorithm, and calculates the risk value of civil-to-penalty by combining the results extracted from the event.

Further, after step S3, the method further includes adjusting the weight of each element to the risk value through historical data to optimize the risk value algorithm.

Further, the risk escalation event escalation risk value early warning provides early warning of the event escalation risk value by establishing a risk escalation event early warning trigger mechanism: a corresponding early warning risk value threshold is set for the event, and when the risk escalation event escalation risk value exceeds the risk threshold Alert information is sent to relevant personnel. The present invention notifies the relevant personnel in a convenient form for processing when the escalated risk value of the risk escalation event exceeds the risk threshold.

Compared with the prior art, the advantages of the present invention are:

The natural language processing method for early warning of a risk escalation event of the present invention can scientifically and efficiently analyze various elements in the event through the natural language processing technology, and the risk analysis model obtained by training with the massive event data can be obtained more accurately. When an incident becomes a risk of a high-risk case, relevant staff can intervene in the incident in advance, resolve conflicts and disputes, and prevent the occurrence of high-risk cases.

Other advantages, objects, and features of the present invention will appear in part from the description that follows, and in part will be appreciated by those skilled in the art from the study and practice of the invention.

Description of drawings

Figure 1 is a flow chart of the present invention.

Detailed ways

In order to make those skilled in the art better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

As shown in FIG. 1 , the natural language processing method for early warning of risk escalation events in this embodiment, and the early warning of risk escalation events by the natural language processing method, includes the following steps:

S1. Establish a data warehouse and connect the data source to the data warehouse;

S2. Based on natural language processing, extract and process the event information in the data source through the event analysis related model;

S3. Input the result value of the event analysis-related model algorithm into the early warning analysis model, and perform risk escalation event escalation risk value analysis and early warning. The data source in this embodiment includes event data. In this embodiment, the natural language processing method is used to scientifically and efficiently analyze various elements in the event, and the risk analysis model obtained by training with the massive event data can give an early warning of the risk escalation event, and the relevant personnel can intervene in the event in advance and resolve conflicts accordingly. Disputes and prevent the occurrence of high-risk cases.

This embodiment S1 specifically includes:

S101. Access to multi-channel data sources: In this embodiment, access event information is obtained from sources such as the city's big data centers, various operators, hotline data centers, and various basic management platforms;

S102. Clean the data according to a unified format, generate unified structured data, and write it into multiple databases.

The data repository established in this embodiment, that is, the data warehouse, can access event data in various fields, and the event data in various fields includes data from the city's citizen hotline, hot events, non-police police information, various help hotlines in the city, people's mediation data, and spear mediation data. Incident data obtained by the Resolution Center, etc. The data repository established in this embodiment is connected to event data in various fields, and can provide a data basis and basis for subsequent classification and analysis.

This embodiment establishes an event analysis related model based on natural language processing. The event analysis related model in this embodiment includes: 1. an event classification model, that is, an event type can be dynamically configured according to business needs; 2. an entity extraction model, which includes Event location, organization, personnel, etc.; 3. Property damage model; 4. Casualty model; According to the result values of the above five model algorithms, input to the early warning analysis model.

This embodiment S2 includes:

S201. Classify events according to the event information, divide the events into multiple major categories by using an event classification model, and set multiple minor categories under each major category. In this embodiment, the K-means algorithm and the Bayesian network are combined. The hybrid algorithm classifies events: the K-means algorithm is used to train the sample clustering, and the improved Bayesian network is used to classify the events according to the training results; in this embodiment, the training results can be dynamically configured through the automatic intervention of the result machine or manual intervention;

S202, sending the classified results into the entity extraction model for entity extraction, and extracting the features of texts and categories that conform to comprehensive social governance. This embodiment conforms to comprehensive social governance, that is, conforms to the provisions of comprehensive social management;

S203. Carry out preset keyword collision according to the different contents of the entity extraction result, calculate the score through the property loss model and the casualty model, and enter the full text into the public sentiment model. The public sentiment model uses lexical analysis, Chinese word vector representation, Analytical operations such as word sense similarity, Chinese DNN language model, dependency syntax analysis, and short text similarity are used to obtain the intensity of the public's emotional intensity of the event;

Among them, the calculation results of each model are passed down through the interface. That is, in this embodiment, the model calculation results are transmitted downward through interfaces between the aforementioned models.

The score involved in S203 in this embodiment is a customizable configuration rule. For example, whether there is casualty or not, different scores will be obtained, and whether there will be property damage and the amount of loss will obtain different scores. Full text refers to all the data, the entirety of the data. The calculation rule of emotional intensity is to divide the intensification into a hundred percent system according to the negative emotion machine learning algorithm, which is divided into three grades.

This embodiment adopts nlp as the feature extraction method for texts and categories of comprehensive social governance. Before step S202, the corpus is fed in advance: a large number of events related to comprehensive social governance are sorted and summarized, and then trained. During the training process, Each entity is marked and named, and the result is corrected, so as to design a feature extraction method that fits the text and category of comprehensive social governance.

The risk value analysis and early warning of risk escalation event escalation in this embodiment is based on a specific algorithm model, and the event is converted into a high-risk case risk value by combining the result information extracted from the event. Among them, the risk value calculation in this embodiment is based on the risk value algorithm, combined with the result information extracted from the event, to calculate the risk value of civil-to-penalty. The results extracted from the event include large categories, small categories, feature entities, difference rates of preset keywords, and the like.

At the same time, the risk escalation event escalation risk value early warning in this embodiment provides an early warning of the event escalation risk value by establishing a risk escalation event early warning trigger mechanism: the early warning is based on the early warning risk value thresholds set by various events. When the escalation risk value of an escalation event exceeds the risk threshold, an early warning is provided to the relevant departments. In this embodiment, relevant personnel may be notified for processing through convenient forms such as short messages and emails. The early warning risk value thresholds of various events are set in advance, and those skilled in the art can flexibly set them according to the actual situation.

In addition, in the long-term calculation process of this embodiment, the risk value algorithm can be optimized according to historical data, and the weight of each element to the risk value can be adjusted. Among them, the weight of each element to the risk value is adjusted through historical data to optimize the risk value algorithm, and each element is the result extracted from the above-mentioned event as the basis for the calculation of the risk value of civil-to-penalty, including each intermediate product.

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 natural language processing method for early warning of risk escalation events is characterized by comprising the following steps:

s1, establishing a data warehouse, and accessing a data source to the data warehouse;

s2, based on natural language processing, extracting and processing event information in the data source through an event analysis related model;

and S3, inputting the result value of the event analysis related model into the early warning analysis model, and carrying out risk value analysis and early warning of risk upgrading event upgrading.

2. The natural language processing method for early warning of a risk escalation event according to claim 1, wherein the S1 specifically includes:

s101, accessing a multi-channel data source;

and S102, cleaning the data according to the uniform format to generate uniform structured data, and writing the uniform structured data into a database.

3. The natural language processing method for early warning risk escalation event according to claim 1, wherein the data source includes event data of each field, specifically including: the system comprises citizen hot lines, hot spot events, non-police affair warning conditions, various help hot lines, citizen regulation data and contradiction regulation center data.

4. The natural language processing method for early warning of risk escalation events according to claim 1, wherein the event analysis correlation models include an event classification model for dynamically configurable event types according to business needs, an entity extraction model for extracting event locations, organizations, and people, a property loss model for calculating property losses, a casualty model for calculating people, and a people emotion model for calculating the emotional intensity of people.

5. The natural language processing method for early warning of a risk escalation event according to claim 4, wherein the S2 includes:

s201, classifying events according to event information, and dividing the events into a plurality of major classes through an event classification model, wherein a plurality of minor classes are arranged under each major class;

s202, sending the classified results into an entity extraction model for entity extraction, and extracting the characteristics of texts and categories which are in accordance with the social comprehensive treatment;

s203, performing preset keyword collision according to different entity extraction result contents, calculating a score through a property loss model and a casualty model, enabling a full text to enter a people emotion model, and obtaining the people emotion violence degree of a risk upgrading event through analysis and calculation;

and transmitting corresponding model calculation results downwards among the models in an interface mode.

6. The natural language processing method for early warning of risk escalation events according to claim 5, wherein the event classification classifies events by a hybrid algorithm combining K-means algorithm and Bayesian network:

training sample clustering through a K-means algorithm, providing an artificial intervention interface for artificial dynamic intervention to configure a training result, and classifying events by using a Bayesian network according to the training result.

7. The natural language processing method for early warning of risk escalation event according to claim 5, characterized in that, before step S202, the entity extraction model is corpus-fed in advance: and (3) arranging and summarizing a large number of events related to the social comprehensive treatment, then training, marking and naming each entity in the training process, and correcting according to the obtained result to obtain an entity extraction model capable of extracting the characteristics of texts and categories which are in accordance with the social comprehensive treatment.

8. The natural language processing method for early warning risk escalation events according to claim 1, wherein the risk escalation event escalation risk value analysis performs risk value calculation according to a risk value algorithm, and calculates a civil diversion risk value by combining the event extracted result.

9. The natural language processing method for early warning of a risk escalation event according to claim 8, further comprising, after step S3: and adjusting the weight of each element on the risk value through historical data to optimize the risk value algorithm.

10. The natural language processing method for early warning of a risk escalation event according to claim 1, wherein the risk escalation event escalation risk value early warning provides early warning of an event escalation risk value by establishing a risk escalation event early warning trigger mechanism:

and setting a corresponding early warning risk value threshold value for the event, and sending early warning information to related personnel when the risk upgrading event upgrading risk value exceeds the risk threshold value.

Images