Method and apparatus for automatic information classification of criminal investigation data

Abstract

An apparatus and a method for automatically classifying various types of data related to criminal investigations are provided. The apparatus for automatic information classification of criminal investigation data according to a preferred embodiment of the present invention comprises: a preprocessing unit for converting various formats of investigation data into a predetermined format of text and outputting digital evidence data text; and an investigation data information classification unit for classifying the type of crime, to which the corresponding data belongs, from the digital evidence data text output from the preprocessing unit. In an embodiment, the apparatus of the present invention may further comprise a result management and recording unit which builds a database, in a predetermined format, using the data analyzed by the investigation data information classification unit and the investigation data, and stores the database in an investigation information database.

Description

Method and device for automatic information classification of criminal investigation data

[0001] The present invention relates to a method and device for automatically classifying information of criminal investigation data, and more particularly, to a method and device for automatically classifying information of criminal investigation data by utilizing artificial intelligence technology to automatically classify various types of data related to criminal investigation, thereby enabling a user to easily search for desired information by filtering criminal investigation data from various viewpoints.

[0002]

[0003] During a criminal investigation, a large amount of relevant evidence, statements, CCTV footage, and other materials are generated. This vast amount of evidence must be carefully categorized by type to ensure proper use.

[0004] Currently, these are classified manually, which not only makes the classification process itself time-consuming and costly, but can also slow down the speed of criminal investigations.

[0005] Moreover, while criminal investigation data contains diverse types of information, its format is unstructured. Manually analyzing this unstructured data has numerous limitations.

[0006]

[0007] The present invention has been made in consideration of these points, and its purpose is to provide a method and device that can automatically classify various types of evidence related to criminal investigation using an artificial intelligence model.

[0008]

[0009] An automatic information classification device for criminal investigation data according to a preferred embodiment of the present invention comprises a preprocessing unit that converts various types of investigation data into a text format of a predetermined format and outputs digital evidence text, and an investigation data information classification unit that classifies the digital evidence text output from the preprocessing unit into which type of crime the data belongs.

[0010] The above preprocessing unit may include an OCR processing module that extracts text from image data using OCR (Optical Character Recognition) technology and saves the text as a text file in a predetermined format, an STT (Speech-to-Text) processing module that converts voice data into text using STT technology and saves the text as a text file in a predetermined format, and a Python library processing module that extracts necessary information from other types of data using libraries suitable for the corresponding format provided by the programming language Python and saves the extracted information as a text file in a predetermined format.

[0011] In one embodiment, the investigative data information classification unit calculates a probability value for each crime type and classifies the type with the highest probability value or higher as the highest crime type of the corresponding data.

[0012] In one embodiment, the investigative data information classification unit transforms the digital evidence text output from the preprocessing unit into a numerical form so that the machine can understand it, expresses the information of the text transformed into a numerical form in a vector form, and then uses a crime type classifier to calculate a probability value that the vector value belongs to each crime type.

[0013] In one embodiment, the investigative data classification unit classifies crime types using a model that additionally trains digital crime investigation data on a large-scale language model pre-trained using public security domain learning data.

[0014] The above crime type classifier may be trained using a large amount of fine-tuning data consisting of text and crime types for that text.

[0015] In one embodiment, the automatic information classification device for criminal investigation data of the present invention further includes a result management and recording unit that stores the data analyzed by the investigation data information classification unit and the investigation data in a database in a predetermined format in an investigation information database.

[0016] A method for automatically classifying information of criminal investigation data according to a preferred embodiment of the present invention comprises a preprocessing step of converting various types of investigation data into a text format of a predetermined format to generate digital evidence text, and an investigation data information classification step of classifying the digital evidence text generated in the preprocessing step to which type of crime the data belongs.

[0017] The above preprocessing step may include an OCR processing step of extracting text from image data using OCR (Optical Character Recognition) technology and saving it as a text file in a predetermined format, an STT processing step of converting voice data into text using STT (Speech-to-Text) technology and saving it as a text file in a predetermined format, and a Python library processing step of extracting necessary information from data in other formats using libraries suitable for the corresponding format provided by the programming language Python and converting it into a text file in a predetermined format and saving it.

[0018] In one embodiment, the investigative data information classification step includes a step of calculating a probability value for each crime type and classifying the type with the highest probability value or higher as the highest type of the corresponding data.

[0019] In one embodiment, the investigative data information classification step includes a step of converting digital evidence text output from a preprocessing unit into a numerical form so that a machine can understand it, a step of expressing information of the text converted into a numerical form in a vector form, and a step of using a crime type classifier to calculate a probability value that the vector value belongs to each crime type.

[0020] In one embodiment, the automatic information classification method of criminal investigation data of the present invention includes a step of pre-training a large-scale language model using security domain learning data, and a step of additionally training digital criminal investigation data on the pre-trained large-scale language model.

[0021] In one embodiment, the automatic information classification method of criminal investigation data of the present invention further includes a step of creating a crime type classifier by fine-tuning the learned large-scale language model using a large amount of fine-tuning data consisting of text and crime types for the text.

[0022] The automatic information classification method of criminal investigation data of the present invention may further include a result management and recording step of storing the data analyzed by the investigation data information classification unit and the investigation data in a database in a predetermined format in an investigation information database.

[0023] In one embodiment, the investigative information database stores link information that can access investigative data, an app from which the investigative data was extracted, the data type of the investigative data, link information that can access a digital evidence text file generated in a preprocessing process, and crime type information.

[0024] The above format can be used in the JSON (Java Script Object Notation) file format.

[0025]

[0026] According to the present invention, by automatically classifying various types of evidence related to criminal investigations using an artificial intelligence model, processing time and effort can be significantly reduced compared to manually classifying the data. Furthermore, by quickly classifying large amounts of data, the investigation process can be accelerated.

[0027] In addition, according to the present invention, key information required for a criminal investigation can be accurately discovered and classified, thereby enabling the rapid discovery of evidence that plays a key role in the investigation, thereby increasing the efficiency of the investigation.

[0028] In addition, according to the present invention, objective analysis is performed without human subjectivity, thereby increasing the consistency and reliability of the results.

[0029] Furthermore, the present invention allows for efficient use of data even when reanalysis is required in the future. Because the classified data is organized and structured, it is easily utilized for trend analysis or additional research.

[0030] Furthermore, the present invention allows investigators to focus their resources on more specialized tasks, freeing up resources previously devoted to manual classification. Furthermore, the quality of investigations can be improved by allocating resources to tasks requiring subjective human judgment.

[0031] These effects can help reduce time and costs in criminal investigations and lead to more accurate and efficient investigations.

[0032]

[0033] FIG. 1 is a block diagram showing the configuration of an automatic information classification device for criminal investigation data according to a preferred embodiment of the present invention.

[0034] FIG. 2 is a flowchart showing the operation flow of a method for automatically classifying information of criminal investigation data according to one embodiment of the present invention.

[0035] Figure 3 is a conceptual diagram showing how investigative data in various formats is converted into digital evidence text in the data preprocessing stage.

[0036] Figure 4 is an example of digital evidence text.

[0037] Figure 5 shows the investigation data information classification process of the investigation data information classification unit according to one embodiment of the present invention.

[0038] Figure 6 shows an example of the types of crimes classified in one embodiment of the present invention.

[0039] Figure 7 is a conceptual diagram for explaining the pre-training process of Pol-Robust-BERT according to one embodiment of the present invention.

[0040] Figure 8 is an example of data used for pre-training of the Pol-Robust-BERT model.

[0041] Figure 9 shows sample data used for fine-tuning the Pol-Robust-BERT model.

[0042] Figure 10 shows an example of an investigative information database.

[0043] Figure 11 shows an example of the result of searching the investigative information database.

[0044] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings.

[0045] FIG. 1 is a block diagram showing the configuration of an automatic information classification device for criminal investigation data according to a preferred embodiment of the present invention.

[0046] An automatic information classification device (100) for criminal investigation data according to a preferred embodiment of the present invention comprises a preprocessing unit (110) that converts various types of criminal investigation data (M) (hereinafter referred to as “investigation data”) into text format, an investigation data information classification unit (120) that reads evidence text data that has undergone a preprocessing process and classifies which type of crime the data belongs to, and a result management and recording unit (130) that systematically organizes the automatically classified analysis results and stores them in an investigation information database (D).

[0047] Investigative data (M) is comprised of various formats, including text, images, audio files, and video files. The preprocessing unit (110) converts various types of data into a common text format, thereby ensuring consistent subsequent processing.

[0048] The investigation data classification unit (120) can be implemented using a large language model (LLM) trained with public security domain data and fine-tuned with digital crime investigation data, and equipped with specialized capabilities for classifying criminal investigation data. The investigation data classification unit (120) preprocesses various types of investigation data (M) and converts them into text-format data. The input data is analyzed to determine which crime type the data belongs to. For example, the unit reads criminal investigation data and determines whether the data falls under the category of breach of trust, fraud, embezzlement, or robbery.

[0049] The Results Management and Recording Unit (130) stores the data analyzed by the Investigation Data Classification Unit (120) and the original investigation data (M) in a database format in the Investigation Information Database (D). This database may include the original results, the analysis results themselves, and recorded metadata. By utilizing the database to store and manage the investigation data, efficient retrieval and reuse of investigation results or data is possible.

[0050]

[0051] FIG. 2 is a flowchart showing the operation flow of a method for automatically classifying information of criminal investigation data according to one embodiment of the present invention.

[0052] A method for automatically classifying information of criminal investigation data according to one embodiment of the present invention comprises a preprocessing step (S110) of converting various types of criminal investigation data into text format, and a criminal investigation data information classification step (S120) of reading evidence text data that has undergone the preprocessing process and classifying which type of crime the data belongs to.

[0053] In the preprocessing stage, the device of the present invention extracts text from image data using Optical Character Recognition (OCR) technology, and converts voice data into text using Speech-to-Text (STT) technology. This allows for the expression of various types of data in a common format for analysis.

[0054] In the criminal investigation data classification stage, a pre-trained language model is used to classify the criminal investigation data. Since the present invention utilizes textual information from criminal investigation data, a pre-trained language model specialized for the security domain, such as the Pol-BERT_small model, is used. Pol-Robust-BERT reads text data and classifies it into the crime type it belongs to.

[0055] The automatic information classification method for criminal investigation data of the present invention may further include a results management and recording step (S130) for systematically organizing and storing the automatically classified analysis results in a database. This allows users to filter criminal investigation data from various perspectives and easily retrieve desired information.

[0056]

[0057] Next, we will explain each process in more detail.

[0058]

[0059] 1. Data preprocessing

[0060]

[0061] In the data preprocessing step (S110), as illustrated in Fig. 3, various types of investigative data (M) are converted into digital evidence text (T) expressed in a unified format. For example, image data (Mi) is extracted from the image using OCR (Optical Character Recognition) technology in the OCR processing module (111) and saved as a text file in a predetermined format, and audio data (Mv) such as recording files and video files is converted into text using STT (Speech-to-Text) technology in the STT processing module (112) and saved as a text file in a predetermined format.

[0062] In addition, data (Mo) in other formats such as Excel, Doc, CSV (Comma-Separated Values) can be used to extract text by utilizing libraries provided by the programming language Python. That is, the Python library processing module (113) extracts necessary information from data (Mo) such as Excel files, Word files, CSV files, etc. by utilizing a library appropriate for the format, and converts it into a text file in a predetermined format.

[0063] For example, the above "predetermined format" may be a JSON (Java Script Object Notation) file format. FIG. 4 illustrates an example of digital evidence text (T) expressed in the form of a JSON file. In FIG. 4, the file name of the investigative material (M) is recorded in <'src':>, and the data type of the relevant material is recorded in <'src_type':>. The data type may include, for example, text ('text'), image ('image'), audio ('audio'), etc. <'text:'> records the text extracted from the investigative material file. The example in FIG. 4 shows a case where the texts "Sillim Station" and "CCTV" are recognized in the A.png file of the image type.

[0064] In this way, the results of data preprocessing are always expressed in a unified format, so they can be consistently analyzed in subsequent steps.

[0065]

[0066] 2. Classification of investigative data information

[0067]

[0068] Figure 5 illustrates an investigation data information classification process of an investigation data information classification unit (120) according to one embodiment of the present invention. The investigation data information classification unit (120) uses a large language model to classify the type of investigation data from digital evidence text (T) expressed in a unified format. Examples of large language models that can be used include OpenAI's GPT (Generative Pre-trained Transformer) series, Google's BERT (Bidirectional Encoder Representations from Transformers) model, ELECTRA, ALBERT, RoBERTa, etc., and the present invention is not limited to a specific large language model.

[0069] The Pol-Robust-BERT model (122) illustrated in Figure 5 is an artificial intelligence model that has language comprehension and criminal investigation data classification capabilities through a fine-tuning process utilizing a pre-training model. The pre-training and fine-tuning processes of the Pol-Robust-BERT model (122) will be described later.

[0070] The investigation data information classification unit (120) transforms the digital evidence text (T) into a numerical form (121) so that the machine can understand it. Pol-Robust-BERT (122) understands the information of the text transformed into a numerical form and converts it into a vector form (CLS 0,1,2,...,n ) is expressed as a. The FC (Fully Connected) Layer (123) uses a classification model to calculate and output the probability value (S) that the vector value generated through Pol-Robust-BERT belongs to each crime type. The investigation data information classification unit (120) can classify the type with the highest probability value or higher as the highest type of the corresponding data. For example, among the crime types with a probability value of 0.6 or higher, a predetermined number of types with the highest probability can be classified as the crime type of the corresponding data. In addition to the crime type, the classified information may further include important features of the criminal investigation data. Fig. 6 shows an example of a crime type classified in one embodiment of the present invention.

[0071]

[0072] Fig. 7 is a conceptual diagram for explaining the pre-learning process of Pol-Robust-BERT (122) according to one embodiment of the present invention.

[0073] The BERT language model is first pretrained using security domain training data (L1) (S141). The large-scale language model that has undergone the pretraining process is called the "Pol-BERT_small model." This is a pretrained language model specialized for the security domain. Therefore, the Pol-BERT_small model fundamentally possesses the linguistic understanding capabilities of the security domain. However, since it is not specialized for the digital crime investigation domain, the Pol-BERT-small model is additionally trained with digital crime investigation data (L2) (S142), ultimately generating the Pol-Robust-BERT model (122).

[0074] Figure 8 illustrates sample data from the pre-training process. Figure 8 (a) shows the original data, while (b) shows data with key words masked. The masked data, as shown in (b), is input into a large-scale language model. The large-scale language model is trained to ensure that the word "throw" appears more frequently in the [MASK]-processed portion.

[0075] A crime type classifier is created by fine-tuning the Pol-Robust-BERT model (122) generated through the steps of FIG. 7 using criminal investigation data. FIG. 9 shows sample data used for fine-tuning the Pol-Robust-BERT model. The Pol-Robust-BERT model (122) is trained to analyze the text of a conversation and classify the conversation as “economic embezzlement.” In the example of FIG. 9, the input value of the model is the text of the conversation, and the label information is “economic embezzlement.” By training the Pol-Robust-BERT model (122) using a large amount of fine-tuning data consisting of text and the crime type for the text, the Pol-Robust-BERT model (122) acquires specialized capabilities for classifying criminal investigation data.

[0076]

[0077] 3. Results management and recording

[0078]

[0079] The results management and record unit (130) systematically organizes the analysis results from the investigation data information classification unit (120) and stores them in the investigation information database (D). In addition, the results management and record unit (130) links the digital evidence text (T) converted into a unified format by the preprocessing unit (110) and the original data (M) before conversion to the analysis results and stores them in the investigation information database (D).

[0080] In this way, the investigative information database (D) systematically stores information related to the analyzed results. This may include the original data, the analysis results themselves, and recorded metadata. By utilizing the database to store and manage results in this way, efficient retrieval and reuse of results is possible.

[0081] In other words, users can easily filter criminal investigation data from various perspectives to find the information they need, and they can also access the data that formed the basis of the analyzed results. This allows users to record the data that formed the basis of the analyzed results—the original data, datasets, and documents used in the investigation—so that the source of the data can be verified later, thereby helping to increase the reliability and verifiability of the results.

[0082] Metadata can be entered directly by the investigator or generated during the analysis phase. Metadata generated during the analysis phase includes information about the analysis time and date.

[0083]

[0084] Figure 10 shows an example of an investigation information database (D). As can be seen in Figure 10, the investigation information database (D) includes link information that can access the investigation data that was the subject of the analysis, i.e., the original data file, the app from which the original data was extracted, the data type of the original data file, link information that can access the digital evidence text (T) file generated during the preprocessing process, crime type information obtained by analyzing the preprocessing data, and information on the time of creation and time of analysis as meta information.

[0085] The investigative information database (D) allows users to easily search and retrieve desired information. Furthermore, users can utilize various criteria and filters to retrieve results. For example, data can be filtered and retrieved based on various factors, such as the crime type classified in the investigative data classification section (120), as well as the text contained in the digital evidence text (T) file, such as drug type, transaction method (cash, cryptocurrency), and the severity of the crime. This allows users to analyze results from various perspectives, identify crime patterns, and establish effective investigative directions.

[0086] Users can also search the investigation information database (D) by case. Figure 11 shows an example of the search results, showing the investigation data corresponding to case number 7935 and the crime types classified for each investigation data. Users can view the crime type and click on the desired investigation data number to view the contents of the corresponding source file.

[0087]

[0088] Hereinafter, the present invention has been described with several examples, but the present invention is not necessarily limited to these examples even though all components constituting the embodiments of the present invention have been described as being combined as one or operating in combination. Within the scope of the purpose of the present invention, all of the components may be selectively combined and operated one or more times. In addition, although all of the components may be implemented as individual independent hardware, some or all of the components may be selectively combined and implemented as a computer program having program modules that perform some or all of the functions combined in one or more hardware. The codes and code segments constituting the computer program can be easily inferred by those skilled in the art of the present invention. Such a computer program may be stored in a computer-readable medium and read and executed by a computer or processor, thereby implementing an embodiment of the present invention.

[0089] The terms "include," "comprise," or "have" as used herein, unless otherwise specifically stated, imply that the relevant component may be present, and therefore should be construed to include other components rather than to exclude other components.

[0090] The above description is merely an illustrative illustration of the technical idea of ​​the present invention, and those skilled in the art will appreciate that various modifications and variations can be made without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the technical idea of ​​the present invention, and the scope of the technical idea of ​​the present invention is not limited by these embodiments. The scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within a scope equivalent thereto should be interpreted as being included in the scope of the rights of the present invention.

[0091]

[0092] [Explanation of symbols]

[0093] 110 Preprocessing unit,

[0094] 120 Investigative Data Information Classification Department,

[0095] 130 Results Management and Record Book,

[0096] M investigative data,

[0097] D Investigative Information Database.

[0098]

[0099] CROSS-REFERENCE TO RELATED APPLICATION

[0100] This patent application claims priority under 35 USC §119(a) to Korean Patent Application No. 10-2023-0139780, filed October 18, 2023, the entire contents of which are incorporated herein by reference. Furthermore, this patent application claims priority in countries other than the United States for the same reasons, the entire contents of which are incorporated herein by reference.

Claims

1. A preprocessing unit that converts various types of investigative materials into a text format of a specified format and outputs digital evidence text; The investigative data information classification unit classifies the digital evidence text output from the preprocessing unit into which type of crime the data belongs. An automatic information classification device for criminal investigation data.

2. In paragraph 1, the preprocessing unit, An OCR processing module that extracts text from an image using OCR (Optical Character Recognition) technology and saves it as a text file in a specified format, An STT processing module that converts voice data into text using STT (Speech-to-Text) technology and saves it as a text file in a specified format, A Python library processing module that extracts the necessary information from other formats of data by utilizing libraries appropriate for the format provided by the programming language Python, converts it into a text file of a specified format, and saves it. An automatic information classification device for criminal investigation data, including:

3. In paragraph 1, the investigative data information classification department, For each crime type, the probability value of belonging to that crime type is calculated, and the type with the highest probability value or higher is classified as the highest type of crime in the data. An automatic information classification device for criminal investigation data.

4. In paragraph 3, the investigative data information classification department, The digital evidence text output from the preprocessing unit is converted into a numerical form so that the machine can understand it, and the information in the text converted into a numerical form is expressed in a vector form, and then a crime type classifier is used to calculate the probability value that the vector value belongs to each crime type. An automatic information classification device for criminal investigation data.

5. In paragraph 3, the investigative data information classification department, Using a model trained on digital crime investigation data to classify crime types, in addition to a large-scale language model pre-trained using public security domain learning data, An automatic information classification device for criminal investigation data.

6. In paragraph 5, the crime type classifier, It is trained using a large amount of fine-tuned data consisting of text and crime types for that text. An automatic information classification device for criminal investigation data.

7. In any one of paragraphs 1 to 6, The results management and record department stores the data analyzed in the above investigative data information classification department and the investigative data in a database in a prescribed format and stores them in the investigative information database. An automatic information classification device for criminal investigation data, which further provides:

8. In paragraph 7, The above investigative information database stores link information that can access investigative data, the app from which the investigative data was extracted, the data type of the investigative data, link information that can access the digital evidence text file generated during the preprocessing process, and information on the type of crime. An automatic information classification device for criminal investigation data.

9. In any one of paragraphs 1 to 6, The above format is a JSON (Java Script Object Notation) file format. An automatic information classification device for criminal investigation data.

10. A preprocessing step that converts various types of investigative materials into a text format of a specified format to create digital evidence text, An investigative data information classification step that classifies the digital evidence text generated in the preprocessing step into which type of crime the data belongs. A method for automatically classifying information of criminal investigation data.

11. In the 10th paragraph, the preprocessing step is: An OCR processing step that extracts text from an image using OCR (Optical Character Recognition) technology and saves it as a text file in a specified format, An STT processing step that converts voice data into text using STT (Speech-to-Text) technology and saves it as a text file in a specified format; A Python library processing step that extracts the necessary information from other formats of data by utilizing libraries appropriate for the format provided by the programming language Python, converts it into a text file of a specified format, and saves it. A method for automatically classifying information of criminal investigation data, including:

12. In paragraph 10, the investigative data information classification step is: For each crime type, a step is included in which the probability value of belonging to that crime type is calculated, and the type with the highest probability value or higher is classified as the highest type of the corresponding data. A method for automatic information classification of criminal investigation data.

13. In paragraph 12, the investigative data information classification step is: A step of converting the digital evidence text output from the preprocessing unit into numerical form so that the machine can understand it, A step of expressing information in text transformed into numeric form in vector form, A step of using a crime type classifier to calculate the probability that a vector value belongs to each crime type. A method for automatically classifying information of criminal investigation materials including .

14. In paragraph 12, A step for pre-training a large-scale language model using security domain learning data, Including a step of additionally training digital crime investigation data on a pre-trained large-scale language model. A method for automatic information classification of criminal investigation data.

15. In paragraph 14, A step of creating a crime type classifier by fine-tuning the learned large-scale language model using a large amount of fine-tuning data consisting of text and crime types for the text. A method for automatically classifying information of criminal investigation materials including:

16. In any one of paragraphs 10 to 15, The result management and recording stage of storing the data analyzed in the above investigation data information classification department and the above investigation data in a database in a prescribed format in the investigation information database. A method for automatically classifying information of criminal investigation data, further comprising:

17. In paragraph 16, The above investigative information database stores link information that can access investigative data, the app from which the investigative data was extracted, the data type of the investigative data, link information that can access the digital evidence text file generated during the preprocessing process, and information on the type of crime. A method for automatic information classification of criminal investigation data.

18. In any one of paragraphs 10 to 15, The above format is a JSON (Java Script Object Notation) file format. A method for automatic information classification of criminal investigation data.

Images