KR20210083510A - Crime detection system through fake news decision and web monitoring and Method thereof - Google Patents

Abstract

The present invention groups monitoring selection targets into general web targets, dark web targets, and fact collection targets based on expert advice, puts reputable websites in data collection targets, efficiently collects and performs distributed aggregation and incremental indexing, generates learning data from the collected crime intelligence data, compares and analyzes web monitoring data and similar fact documents using natural language processing technology and artificial intelligence technology, and calculates the probability of authenticity of criminal intelligence. So, it improves the reliability of the criminal intelligence. A crime detection system includes a selection part, a collection part, a crime information extraction model generation part, a crime information extraction part and a determination part.

Description

Crime detection system through fake news decision and web monitoring and Method thereof

The present invention relates to a system and method for detecting criminal information through fake news detection and periodic web monitoring, and more particularly, to detect fake news and periodically monitor the web and analyze the collected data to automatically generate the latest criminal information. It is about extraction technology.

The dark web is an encrypted internet network that cannot be accessed with a general internet browser. The dark web was started to ensure security, anonymity and privacy from the surveillance and censorship of power, but it has become a hotbed of crime in the internet underworld such as counterfeit bills, dark arms trade, contract murder, counterfeiting of official documents, drug trade, and child pornography trade. .

Fake news disseminated through social networking services (SNS) or media companies is cleverly manipulated deceptive news, which is false information that is intentionally disseminated in the form of media reports for political or economic gain, and distorts or manipulates core contents. Most of the news contains provocative content that is not easy to confirm.

Since most of the dark web is a hotbed of crime, data collected from the dark web can be used as criminal intelligence, and since fake news is a public achievement hidden behind freedom of expression, the method to identify and combat fake news is difficult. need.

As a prior art related to a criminal intelligence detection system, Patent Document 1 proposes a dark web crime information analysis system and method for performing criminal profiling by analyzing information scanned in the dark web.

The above technology suggests a system for analyzing criminal information acquired through the dark web, and in view of the increasingly sophisticated and intelligent real criminal methods, there is a problem that only dark web criminal information is insufficient. It is necessary to calculate the probability of authenticity of criminal intelligence by collecting, comparing, analyzing, and comprehensively judging web data and fact data.

Korean Patent No. 10-1852107

In order to solve the above problem, the present invention groups the monitoring selection target into general web target, dark web target, and fact collection target based on expert advice, and includes websites with public trust in data collection targets, and performs distributed collection and incremental indexing. efficiently collected through

It is intended to generate learning data from the collected crime information data, analyze web monitoring data and similar fact documents using natural language processing technology and artificial intelligence technology, calculate the probability of authenticity of crime information, and provide the final result to users.

The crime information detection system through web monitoring of the present invention to solve the above problems is a selection unit that selects a crime information detection monitoring target, a collection unit that collects crime information from the selected target, and examines the collected crime information. A crime intelligence extraction model generation unit that learns, corrects and deletes and re-learns, a crime information extraction unit through data learning and analysis, and a judgment unit to determine whether the crime information is true, wherein the crime information extraction model is based on the 6th principle It is characterized in that it is created with existing and new crime models with a schema by

The selection unit is characterized in that the selection target is grouped into a general web target, a dark web target, or a fact collection target.

The collection unit is characterized in that it prevents access blocking by distributed collection through a plurality of collectors, and is efficient collection by performing incremental indexing.

The crime information extraction model generation unit extracts crime information from the crime information data by generating learning data marked with an answer tag from the crime information data collected by a dedicated team including experts and writing an answer document for artificial intelligence learning, characterized by analysis.

The extraction unit is a step of analyzing the crime intelligence data and extracting the latest crime information. From the data collected by the collection unit and the data detected by the open case information search and similar document information search, the unsolved case through the criminal intelligence extraction model extracts crime information including, and analyzes morphemes using natural language processing and artificial intelligence for each collected data, detects unanalyzed words, and generates reports based on the extracted crime information, analyzed morphemes, and detected unanalyzed words. characterized by writing.

The determination unit is a step of determining whether the crime information is true, and searches for data similar to the web monitoring data in the fact DB, and searches for similar fact data to determine whether the extracted crime information is true, natural language processing technology and artificial intelligence Detecting fake news that analyzes web monitoring data and similar fact documents using technology and calculates the probability that the web monitoring data is false, and stores the evidence whenever the probability of fake news increases and provides the final result to the user characterized in that it includes providing.

As another embodiment of the present invention, the method of detecting crime information through web monitoring includes a selection step of selecting a crime information detection monitoring target, a collection step of collecting crime information from the selected target, examining the collected crime information, and learning A crime information extraction model generation step of modifying and deleting and re-learning, a crime information extraction step through data learning and analysis, and a judgment step of determining whether the crime information is true, wherein the crime information extraction model is based on the 6th principle It is characterized in that it is created with existing and new crime models as schemas, and is trained with the correct answer dataset.

The selection and collection step is to extract index words and inverse index words from fact data collections, general web collections, and dark web collections through the steps of setting the monitoring target site and setting the collection schedule, and performing incremental indexing. It is characterized in that it comprises a step.

The step of generating the crime information extraction model is composed of a crime information extraction model construction algorithm and a result verification and re-learning algorithm.

The crime intelligence extraction model construction algorithm is a step of securing existing and new crime intelligence data, a crime intelligence tag marking step, a correct data set creation step for learning, a crime intelligence extraction model learning step, and extraction of the crime information verified through the extraction model performance verification Including the step of storing the model, the result test and re-learning algorithm is a step of requesting verification to a user (expert), a step of generating training data verified through user (expert) verification, an addition step to the non-analyzed dictionary, and a crime intelligence It is characterized in that it includes a step of performing additional learning of the extraction model.

The extraction step consists of data analysis and the latest crime intelligence extraction algorithm, and an analysis performance algorithm for each document.

The latest crime information extraction algorithm includes a step of searching for open cases and similar documents, a step of grouping by web monitoring documents and similar documents of unsolved cases, a step of extracting crime information, an analysis execution step for each document, and an automatic report writing step, The analysis performing algorithm for each document is characterized in that it includes a document pre-processing step, a morphological analysis and an unanalyzed word detection step, an entity name analysis and an unanalyzed word detection step, a syntax analysis step, a semantic translation analysis step, and a document summary step.

The determination step is a data analysis step using natural language processing technology, a fact data search step similar to the corresponding data, and a fact data analysis step using natural language processing technology if there is similar data, and a comparison analysis step of the data currently being analyzed and the fact data and a final fake probability calculation step for crime intelligence data when there are no more similar fact documents remaining.

The present invention groups monitoring selection targets into general web targets, dark web targets, and fact collection targets, includes websites with public trust in data collection targets, and efficiently collects and collects crime intelligence data through distributed collection and incremental indexing. There is a remarkable effect to improve the reliability of criminal intelligence by generating learning data and calculating the probability of authenticity of crime information by comparatively analyzing web monitoring data and similar fact documents using natural language processing technology and artificial intelligence technology.

1 is a block diagram illustrating a crime information detection system according to an embodiment of the present invention.
2 is a flowchart illustrating a collection method for collecting crime information from a crime information detection target in the crime information detection method according to an embodiment of the present invention.
3 is a flowchart illustrating a method of constructing a crime information extraction model.
4 is a flowchart illustrating the latest method of extracting crime information and a method of performing analysis for each document.
5 is a flowchart illustrating a method of testing and re-learning as a result of analysis with a crime information extraction model.
6 is a flowchart illustrating a method for determining whether a crime information is true.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings and the contents described in the accompanying drawings, but the present invention is not limited or limited by the embodiments.

1 is a block diagram illustrating a crime information detection system according to an embodiment of the present invention, wherein the crime information detection system includes a selection unit 10 that selects a crime information detection monitoring target, and a collection unit that collects crime information from the selected target. The unit 20, the crime information extraction model generating unit 30 that verifies, learns, corrects and deletes the collected crime information and re-learns, the crime information extraction unit 40 through data learning and analysis, whether the crime information is true It includes a determination unit 50 for determining the.

The selection unit 10 selects a monitoring target based on expert advice, and groups the selection target into a general web target 110 , a dark web target 120 , and a fact collection target 130 .

In the present invention, the web (web) is used as a term encompassing both the general web and the dark web.

In the criminal intelligence extraction of the present invention, direct extraction by acquaintances, personal connections, or interrogation is excluded, but an online-offline connection is preferable in a way that is used for basic data or index word selection for extraction through the web.

An expert is a broad concept that includes cybercrime-related experts, information protection experts, and criminal psychologists, and offline experts such as information police and investigative police can also be included for online-offline connection.

In addition, experts include a dedicated team including experts, and monitoring targets by them include not only general web targets, dark web targets, and fact-collecting targets, but also websites with credibility such as large newspapers, blogs for mid-level journalists, and blogs for professional lawyers. do.

Recently, as the need for international cybersecurity and cybercrime prevention is increasing, the web target includes not only domestic but also overseas sites, and the expert group may include international crime experts.

In particular, the dark web is a website that cannot be accessed with a general internet browser and exists in an encrypted network and can only be accessed through a specific browser. In order to collect and monitor information on the web, procedures and technologies different from those of the general web must be used, so separate collection and monitoring is required.

Therefore, an expert who can access the dark web, scan and gather information may be required separately.

The feature of the collection unit 20 for collecting crime information from a selected target is to prevent access blocking through distributed collection through a plurality of collectors, and to efficiently collect by performing incremental indexing.

Distributed collection is a collection method that avoids blocking access from collection objects by distributing and accessing a plurality of collection objects to a plurality of collectors.

The collected data goes through frequent fact verification by a dedicated team including experts, and is classified and stored into a web monitoring data DB and a fact DB according to the source, and the web monitoring data DB is classified into a general web DB and a dark web DB and stored. do.

In the case of fact data collection, if the collected fact data is excessively vast and extensive, it may be subjected to frequent fact verification by a dedicated team including the expert after sampling.

Index words and inverse index words are extracted from the collected data.

An index is a method of searching memory or data in memory using an index (a list of data and its storage locations) to speed up the time it takes to find a desired item of data.

Inverse indexing is a technology that maps and stores the primary key of documents or values such as addresses and file names to keywords after morphological analysis and tuning of the target document. The advantage of inverting indexing is that you can find it very quickly. It requires additional work and memory than before, but the search is processed very quickly.

Incremental indexing is a method of retrieving a large amount of information such as big data and GPU (Graphics Processing Unit), and refers to an incremental search system that reflects only the addition, deletion, and change of the relevant document to the search service.

The crime information extraction model generation unit 30 is a crime information extraction model generation unit that tests, learns, corrects and deletes the collected crime information and re-learns, and extracts the correct answer tag from the crime information data collected by a dedicated team including experts. By generating the marked learning data and writing the correct answer document for artificial intelligence learning, the crime information is extracted and analyzed from the crime information data. Examples of artificial intelligence learning include machine learning through deep learning.

The function of the crime intelligence extraction model generation unit is to build a crime information extraction model, test the results, and re-learning.

Building a crime intelligence extraction model is the task of securing existing and new crime intelligence data, marking the crime information tag, and generating a correct data set for learning, and the generated model is verified by a dedicated team including criminal intelligence and crime information experts. is saved after

The criminal intelligence extraction model categorizes criminal cases and sets the time, place, crime target, crime method, and motive as a schema based on the 6th principle. When the document to be analyzed is input to the criminal intelligence extraction model, the model binds the extracted criminal information information in parentheses and tags along with the identifier what information the information corresponds to. For example, if a document with the content "I will wait for the eagle in front of Exit 4 of Hongik University Station around 10:00 pm on the 10th of this month. I will deliver it as scheduled." (Around 10 p.m./hour on Sunday) (in front/place of Exit 4 of Hongik University Station) (Eagle/the subject of the crime (drug term)) will be on standby. (Standby/criminal method (drug term)) will be delivered as scheduled." to derive

Criminal intelligence refers to materials necessary to produce criminal information as raw materials that have not yet been analyzed, which identify people, evidence, or cases, or are related to witnesses or evidence of criminal cases. It is the product of the analysis process to evaluate the intelligence that has been collected, and it refers to the conclusion or prediction of a crime phenomenon created by using scientific methods to solve the problem after integrating the relevant information into an intensive package, but it is usually used interchangeably. have.

Criminal intelligence is one of the contents of investigation intelligence, and it can be said that all matters that can be used as a reference in criminal investigation.

Since the built model does not have a long validity period, it must always be renewed by new environments and information. In the present invention, the renewal function is performed by itself by the result verification and re-learning process.

That is, one of the features of the present invention is that when a new crime information is added, the operation of reproducing and correcting the part that is in conflict with the existing crime information is automatically performed during the result verification and re-learning process during the construction of the criminal information extraction model of the present invention .

The result verification and re-learning process is characterized by including a step of requesting verification from an expert, a step of generating learning data verified through expert verification, an additional step to the unanalyzed dictionary, and a step of performing additional learning of the criminal intelligence extraction model.

The present invention is a virtuous cycle in which crime intelligence is extracted from crime intelligence data, and a report written based on it is delivered to an expert, and is regenerated as a correct answer document with errors corrected through expert verification and fed back as re-learning data for generating a criminal intelligence extraction model. This is done automatically in the stage of building a model for extracting criminal information.

The crime intelligence extraction unit 40 through data learning and analysis analyzes the crime information data and extracts the latest crime information. From the data collected by the collection unit and the data detected by the open case information search and similar document information search Crime intelligence is extracted through the crime intelligence extraction model, and morphemes are analyzed using natural language processing and artificial intelligence for each collected data, and unanalyzed words are detected. Create a report based on

The data learning and analysis is machine learning in which a computer program improves its information processing ability by itself through learning using data and processing experience, and among them, it corresponds to supervised learning in which mapping between input and output exists.

The present invention includes a source-specific training set for mapping between input and output. For example, it is to improve the detection ability of real crime information through supervised learning that learns fake news or fake crime information and sources of credible news or real crime information as input and output pairs, respectively.

For machine learning, artificial intelligence that mimics human learning ability can be used.

Similarity analysis algorithms such as Term Frequency-Inverse Document Frequency (TF-IDF) can be used when searching for documents similar to unsolved cases from the collected data. TF-IDF is an algorithm that calculates the similarity between sentences through word frequency and word specificity.

In order to perform similarity analysis using the TF-IDF algorithm, morphological analysis is first performed on the documents to be analyzed. TF and IDF values are calculated for the result of morphological analysis. The TF value is obtained from the number of occurrences of the word in the document, the DF value is obtained from the number of documents among the entire documents, and the IDF value is obtained by taking the reciprocal of the DF value. Finally, similarity analysis is performed using the TF-IDF value calculated by multiplying the obtained TF value and the IDF value.

In the process of analyzing crime intelligence data and extracting the latest crime information, a pre-established criminal intelligence extraction model is applied, and a virtuous cycle in which the criminal information extraction model is renewed by the newly extracted latest crime information continues.

For each collected data, natural language processing and artificial intelligence are used to analyze morphemes, detect unanalyzed words, and create a report based on the extracted criminal intelligence, the analyzed morphemes, and the detected unanalyzed words.

For reference, natural language processing is a computational technique for automating human language analysis and expression.

The written report is delivered to the user (expert), and after verification of the user (expert), it is regenerated as a correct answer document in which the error is corrected, and is fed back as re-learning data for generating the criminal intelligence extraction model, so that the renewal of the criminal information extraction model is performed. will continue to be done

The determination unit 50 is a step of determining whether fake news or crime information is true, and similar fact data that searches for data similar to web monitoring data in the fact DB, and determines whether the extracted news or crime information is true. Detecting fake news that analyzes web monitoring data and similar fact documents using search, natural language processing technology and artificial intelligence technology, and calculates the probability that the web monitoring data is false, and stores the evidence whenever the probability of being fake increases and providing the final result to the user.

Data similar to web monitoring data is searched for in the fact DB, and if similar documents do not exist, the probability that the web monitoring data is fake is very high. Request a review or re-analysis from the team, and if similar documents exist, compare and contrast the web monitoring data and similar documents using natural language processing technology and artificial intelligence technology to calculate the false probability of the web monitoring data. When detected, the fake probability is calculated according to different degrees and the document is stored as evidence.

If there are similar fact documents that are not continuously compared and contrasted, the fake probability is calculated according to the degree to which different contents are detected by additional comparison and comparison, or the same loop is repeated until there are no similar fact documents remaining after updating the calculated fake probability. It is repeated, and if there are no similar fact documents remaining, the final false probability of the criminal intelligence data is calculated.

Whenever the false probability increases in the above process, the data used as the basis is stored as the basis data and the final result is provided to the user.

On the other hand, the method of detecting crime information through web monitoring of the present invention includes a selection step of selecting a crime information detection monitoring target, a collection step of collecting crime information from the selected target, learning by examining the collected crime information, and modifying and deleting the information. It consists of a stage of generating a model for extracting criminal information by re-learning it, a stage of extracting criminal information through data learning and analysis, and a judgment stage of determining whether the information is true or not.

2 is a flowchart illustrating a collection method for selecting a crime information detection target and collecting crime information from the selected target in the crime information detection method according to an embodiment of the present invention. It includes a step of extracting index words and inverse index words from fact data collections, general web collections, and dark web collections, and performing incremental indexing through the steps of setting and setting the collection schedule.

In the verification stage of the collected fact data, if the amount of collected fact data is appropriate, it is sent as fact verification data to experts and the front-end team, and if it is excessive, it is sent as verification data after sampling.

After fact-testing by experts and front-end teams, index words and inverse index words are extracted from the collected data, and incremental indexing is performed for quick retrieval and stored in the fact DB.

The processing stage of the collected web data is first classified into the general web and the dark web. In the case of data collected from the general web, index words and inverse index words are extracted from the collected data, and incremental indexing is performed for quick search and stored in the fact DB. Save.

In the case of dark web data, index words and inverse index words are extracted by linking with hidden services, and incremental indexing is performed for quick search and stored in the fact DB.

The step of generating the crime information extraction model is composed of a crime information extraction model construction algorithm and a result verification and re-learning algorithm.

3 is a flowchart illustrating a method of constructing a crime intelligence extraction model. The steps of securing existing and new crime intelligence data, the step of marking the crime information tag, the creation of a correct answer dataset for learning, the step of learning the criminal intelligence extraction model, and verification of the performance of the extraction model It includes the step of storing the verified criminal intelligence extraction model through.

The criminal intelligence extraction model categorizes criminal cases and sets the time, place, crime target, crime method, and motive as a schema based on the 6th principle. When the document to be analyzed is input to the criminal intelligence extraction model, the model groups the extracted criminal information information in parentheses and tags along with the identifier what information the information corresponds to. For example, if a document with the content "I will wait for the eagle in front of Exit 4 of Hongik University Station around 10:00 pm on the 10th of this month. I will deliver it as scheduled." (Around 10 p.m./hour on Sunday) (in front/place of Exit 4 of Hongik University Station) (Eagle/the subject of the crime (drug term)) will be on standby. (Standby/criminal method (drug term)) will be delivered as scheduled." is derived and delivered to the user.

Securing the previously collected crime intelligence data and newly collected crime intelligence data, and tagging the crime intelligence tag on the crime intelligence data obtained by linking with the criminal intelligence tag set DB secured by consulting an expert in advance Create an answer data set.

The crime intelligence extraction model is trained with the generated correct answer dataset and the performance of the extraction model is tested. If the correct rate of the extraction model is higher than the reference value, the crime information extraction model is saved. Repeat.

The answer data set is written as a schema based on slang expressions and the 6th principle. As an example of the correct answer data set, it is written as a tag of the 6th principle, such as 'My brother (dispatcher) today (time) in front of Hongdae (place) meat (cannabis) wangchang (amount delivered) (delivery).' Often times.

The extraction step consists of data analysis and the latest crime intelligence extraction algorithm, and an analysis performance algorithm for each document.

4 is a flowchart illustrating the latest method of extracting crime information and a method of performing analysis for each document.

The latest crime information extraction method of FIG. 4 includes a step of searching for open cases and similar documents, a step of grouping by web monitoring documents and similar documents of unsolved cases, a step of extracting crime information, an analysis execution step for each document, and an automatic report writing step.

A similarity analysis algorithm such as TF-IDF can be used to search for open cases and similar documents. TF-IDF is an algorithm that calculates the similarity between sentences through word frequency and word specificity.

Unsolved cases and similar documents are searched from the data extracted from the open case DB and web (general web and dark web) monitoring data DB, and the crime information is collected using the criminal intelligence extraction model after grouping the web monitoring documents and the open case similar documents. extract This extraction includes the extraction of overall criminal information from the data collected in this cycle and the extraction of criminal information about cases similar to unsolved cases among the data collected in this cycle.

The method of performing analysis for each document of FIG. 4 includes a document preprocessing step, a morphological analysis and an unanalyzed word detection step, an entity name analysis and an unanalyzed word detection step, a syntax analysis step, a semantic area analysis step, and a document summary step. Natural language processing and artificial intelligence technologies can be used for data analysis. Document summary is summarized as information according to the six-fold principle based on the results of syntax analysis and semantic translation analysis.

An automatic report is created based on the crime intelligence data extracted as described above and the analysis results of the collected data. The preparation method is written according to a predefined report format by listing the crime intelligence data extracted from the document and the unanalyzed words detected. The prepared report is stored in the crime intelligence report DB, and after verification by the user (expert), it is used as the latest crime intelligence information support, unsolved crime intelligence information support, and learning data generation data.

5 is a flow chart showing a method for testing and re-learning as a result of analysis with a criminal intelligence extraction model. A user (expert) automatically reviews the verification-requested report, corrects the wrong part, regenerates it as a correct document, and enters the system It is a process of re-learning, and based on this, it is a process to derive more accurate results over time. This process includes the step of requesting verification from the user (expert), the step of generating the learning data verified through user (expert) verification, the step of adding the unanalyzed word to the dictionary, and the step of re-learning the crime information extraction model do.

In summary, criminal intelligence is extracted from the criminal intelligence data extracted by the criminal intelligence extraction model, and the report prepared based on this is delivered to the expert, verified by the expert, and then regenerated as an error-corrected correct document to create the criminal intelligence extraction model again. The virtuous cycle that is fed back as re-learning data for the purpose of the present invention is that it is made by itself in the stage of building the criminal intelligence extraction model of the present invention.

6 is a flowchart illustrating a method for determining whether a crime information is true. The data analysis step using natural language processing technology, the fact data search step similar to the corresponding data, and the fact data analysis step using natural language processing technology when similar data exists It includes a step of comparing and analyzing the data under analysis with the fact data, and a step of calculating the final fake probability for the crime intelligence data when there are similar fact documents remaining.

It analyzes the data of the web monitoring data DB using natural language processing technology, and uses a similarity analysis algorithm such as TF-IDF to search for fact data similar to the corresponding data.

Depending on the existence of similar data, if similar documents do not exist, the probability that the web monitoring data is fake is very high, so it is judged that the calculation of the fake probability is meaningless, and it is classified as 'pending judgment' and automatically requested to review or reanalyzed by experts and dedicated teams , and if there is a similar document, the web monitoring data and the similar document are compared and contrasted using natural language processing technology and artificial intelligence technology to calculate the false probability of the web monitoring data. After calculating the fake probability, the document is stored as evidence.

If there are similar fact documents that are not continuously compared and contrasted, the fake probability is calculated according to the degree to which different contents are detected by additional comparison and comparison, or the same loop is repeated until there are no similar fact documents remaining after updating the calculated fake probability. It is repeated, and if there are no similar fact documents remaining, the final false probability of the criminal intelligence data is calculated.

Whenever the false probability increases in the above process, the data used as the basis is stored as the basis data and the final result is provided to the user.

The identification of fake news is used to determine whether the crime information is true, and the fake news detector works in the same way as described above.

10; selection 20; collector
30; generating unit 40; extractor
50; judgment unit 110; general web
120; Dark Web 130; Fact Gathering

Claims (11)

a selection unit that selects a crime intelligence detection and monitoring target;
a collection unit for collecting crime information from the target selected by the selection unit;
a crime information extraction model generating unit for re-learning by examining, learning, modifying and deleting the crime information collected by the collection unit;
A crime information extraction unit that learns data from the crime information extraction model and extracts crime information through analysis; and
Including a determination unit for determining whether the crime information extracted from the crime information extraction unit is true,
The crime intelligence extraction model is a schema based on the 6th principle, and is generated as an existing and new crime model, and is a crime intelligence detection system through web monitoring, characterized in that it is learned from the correct answer dataset. According to claim 1,
The selection unit is a crime intelligence detection system through web monitoring, characterized in that grouping the selection target into a general web target, a dark web target, and a fact collection target. According to claim 1,
The collection unit prevents access blocking by distributed collection through a plurality of collectors, and is an efficient collection by performing incremental indexing. According to claim 1,
The crime information extraction model generation unit generates learning data with an answer tag marked from the collected crime information data and creates an answer document for AI learning, characterized in that it extracts and analyzes the crime information from the crime information data Crime intelligence detection system through monitoring. According to claim 1,
The extraction unit,
Extracting crime information including unsolved cases from the data collected by the collection unit and the data detected by the open case information search and similar document information search through the crime information extraction model,
For each collected data, natural language processing and artificial intelligence are used to analyze morphemes and detect unanalyzed words.
Crime intelligence detection system through web monitoring, characterized in that it creates a report based on the extracted crime information, the analyzed morpheme, and the detected unanalyzed word. According to claim 1,
The judging unit,
Search for data similar to web monitoring data in the fact DB, and search for similar fact data to determine the truth of the extracted criminal intelligence,
Analyze web monitoring data and similar fact documents using natural language processing technology and artificial intelligence technology to detect fake news that calculates the probability that the web monitoring data is false,
Crime intelligence detection system through web monitoring, characterized in that it includes providing a result of storing the evidence and providing the final result to the user whenever the probability of a fake increases. a selection step of selecting a crime intelligence detection and monitoring target;
A collection step of collecting crime information from the selected target;
A crime information extraction model generation step of examining, learning, modifying, and deleting the collected criminal information and re-learning;
Criminal intelligence extraction step through data learning and analysis, and
Including a judgment step of determining whether the crime information is true,
The crime information extraction model is generated as a schema according to the 6th principle, and the crime information detection method through web monitoring, characterized in that it is learned from the correct answer dataset. 8. The method of claim 7,
The selection and collection step is
Extracting index words and inverse index words from fact data collections, general web collections, and dark web collections through the steps of setting the monitoring target site and setting the collection schedule, and performing incremental indexing. A method of detecting criminal intelligence through web monitoring. 8. The method of claim 7,
The crime intelligence extraction model generation step consists of a crime intelligence extraction model construction algorithm and result verification and re-learning algorithm,
The crime intelligence extraction model construction algorithm includes the steps of securing the existing and new crime intelligence data, the crime intelligence tag marking step, the generation of correct answer dataset for learning, the crime intelligence extraction model learning step, and the criminal intelligence verified through the extraction model performance verification storing the extraction model;
The result verification and re-learning algorithm, web monitoring, characterized in that it includes a verification request step, a step of generating learning data verified through user verification, a step of adding to the unanalyzed dictionary, and a step of performing additional learning of a crime intelligence extraction model A method of detecting criminal intelligence through 8. The method of claim 7,
The extraction step consists of data analysis and the latest crime intelligence extraction algorithm and an analysis execution algorithm for each document,
The latest crime intelligence extraction algorithm includes a step of searching for unsolved cases and similar documents, a step of grouping web monitoring documents and similar documents of unsolved cases, a step of extracting crime information, a step of analyzing each document, and an automatic report writing step,
The analysis performing algorithm for each document includes a document pre-processing step, a morphological analysis and an unanalyzed word detection step, an entity name analysis and an unanalyzed word detection step, a syntax analysis step, a semantic translation analysis step, and a document summary step. A method of detecting criminal intelligence through web monitoring. 8. The method of claim 7,
The determining step includes a data analysis step using natural language processing technology, a fact data search step similar to the corresponding data, and a fact data analysis step using natural language processing technology when similar data exists;
Comparative analysis stage between the data currently being analyzed and the fact data, and
Crime intelligence detection method through web monitoring, characterized in that it includes a final false probability calculation step for the crime intelligence data when there are no more similar fact documents remaining.

Images