Method and apparatus for identifying and recognizing same person in messenger chat room

Abstract

Disclosed is a method for identifying and recognizing the same person in a messenger chat room. The method comprises the steps in which: an extractor tags words included in chat room information with named entity recognition tags and extracts the tagged named entity recognition tags; and a criminal investigation analyzer generates knowledge graphs having a subject (S)–predicate (P)–object (O) structure on the basis of the tagged named entity recognition tags, and identifies and recognizes, as the same person, words tagged with named entity recognition tags indicating a person in knowledge graphs, among the knowledge graphs, having a similarity of a specific threshold or greater.

Description

Method and device for identifying and recognizing the same person within a messenger chat room

[0001] The present invention relates to a technology for extracting necessary information and inferring it so that words referring to the same person, such as names, job titles, nicknames, and titles described in messenger chat rooms, among data extracted through mobile phone forensics from a suspect's digital device during a criminal investigation, can be identified and recognized as a single person.

[0002] In criminal investigations, the analysis of data extracted through mobile phone forensics has become a critically important factor in identifying criminal circumstances and proving charges. Analyzing the content of digital evidence obtained through mobile phone forensics, such as messenger chat rooms, emails, and text messages, is crucial for determining whether a crime is suspected, the motive for the crime, circumstances of prior conspiracy, the existence and relationship of joint crimes, and specific means of committing the crime; investigators devote a significant amount of time and effort to analyzing this data.

[0003] In the case of chat room analysis, which takes the most time during digital evidence analysis, if a single conventional chat room has been maintained for a long time, there are so many utterances that it is difficult for a person to read through the conversation content one by one to detect criminal suspicion.

[0004] To improve this, conventional technology is limited to viewing search results by entering keywords or examining only relevant conversations by directly specifying the time frame during which criminal activity is suspected. Furthermore, to examine the relationships between key criminal suspects, the process involves directly selecting individuals as nodes and defining their relationships to establish connection lines.

[0005] A problem with conventional technology is that during the process of identifying relationships between key suspects, if the analyst fails to recognize not only the subjects participating in the chat room (participants in one-on-one chats or group chat networks) but also the people mentioned in the conversation and their titles, they may perceive the same individual as a different person or fail to recognize them as a person at all; therefore, improvement is needed in this regard.

[0006] The objective of the present invention, which aims to solve the aforementioned problems, is to provide a method and device for identifying and recognizing the same person within a messenger chat room to improve investigation efficiency by automatically determining whether the person is the same through the analysis of words describing the main criminal suspect using various terms among the digital evidence obtained from the main criminal suspect during chat room analysis.

[0007] In addition, the present invention has another objective of providing a method and apparatus for identifying and recognizing the same person within a messenger chat room to determine whether words referred to by various titles, such as nicknames or job titles, are the same person by converting related core information into a knowledge graph in an SPO structure and determining their relevance to provide grounds for determining the same person.

[0008] In addition, another objective of the present invention is to provide a method and device for identifying and recognizing the same person within a messenger chat room for accurate information processing based on information finally determined by an investigator (analyst), by allowing the user to finally confirm the analyzed results.

[0009] A method for identifying and recognizing the same person within a messenger chat room according to one aspect of the present invention for achieving the above-mentioned purpose comprises the steps of: tagging words included in chat room information with named entity recognition tags and extracting the tagged named entity recognition tags; generating knowledge graphs with an S (subject) P (predicate) O (object) structure based on the tagged named entity recognition tags; and identifying and recognizing as the same person words tagged with the named entity recognition tags representing a person in knowledge graphs having a similarity of at least a specific threshold among the knowledge graphs.

[0010] A device for identifying and recognizing the same person within a messenger chat room according to another aspect of the present invention comprises: an extractor that tags words included in chat room information with named entity recognition tags and extracts the tagged named entity recognition tags; and a crime investigation analyzer that generates knowledge graphs with an S (subject) P (predicate) O (object) structure based on the tagged named entity recognition tags, and identifies and recognizes words tagged with the named entity recognition tags representing a person as the same person in knowledge graphs having a similarity of more than a specific threshold among the knowledge graphs.

[0011] According to the present invention, the time required to identify the same person among chat room contents containing very long conversations among the data extracted after digital forensic analysis and the manpower required for this can be reduced.

[0012] In addition, by utilizing predefined crime analysis schemas to extract information for extended named entities that require interpretation by crime, the accuracy of term extraction and analysis suitable for investigative purposes can be enhanced.

[0013] In addition, crime interpretation capabilities can be improved by transitioning from the existing method of finding the same person manually or through simple questions / searches to an automatic analysis / connection method.

[0014] FIG. 1 is a configuration diagram of a device for identifying and recognizing the same person in a messenger chat room according to an embodiment of the present invention.

[0015] Figure 2 is a diagram illustrating the crime attribute extended named entity recognition schema used to extract NER tags in the extended NER tag extractor of Figure 1.

[0016] Figure 3 is a diagram illustrating an extended named entity recognition (NER) tag extracted by the extended NER tag extractor of Figure 1.

[0017] Figure 4 is an exemplary configuration diagram of the crime investigation analyzer of Figure 1.

[0018] FIG. 5 is a flowchart of a method for identifying and recognizing the same person within a messenger chat room according to an embodiment of the present invention.

[0019] FIG. 6 is an exemplary configuration diagram of a computing device for implementing the device of FIG. 1 and / or the method of FIG. 5.

[0020] The terms used in this specification are used merely to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, terms such as “comprising” or “having” are intended to indicate the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0021]

[0022] FIG. 1 is a configuration diagram of a device for identifying and recognizing the same person in a messenger chat room according to an embodiment of the present invention.

[0023] Referring to FIG. 1, the device (100) according to an embodiment of the present invention is configured to identify and recognize the same person within a messenger chat room by using messenger chat room information obtained from a digital forensic device (10) as input.

[0024] Digital forensic equipment (10) consists of physical devices used to extract or image (replicate) data such as messenger chat room information, contact information, etc. from a digital device (e.g., smartphone, etc.) by a program that analyzes, recovers, or duplicates data or reconstructs an event. Since the present invention is not characterized by the structure and function of digital forensic equipment, a description thereof is replaced by known technology.

[0025] To identify and recognize the same person within a messenger chat room, the device (100) according to an embodiment of the present invention may include an extended Named Entity Recognition (NER) tag extractor (110), a storage device (120), a crime investigation analyzer (130), and a display device (140).

[0026] The NER tag (NER information) extractor (110) tags words included in chat room information input from digital forensic equipment (10) with named entity recognition tags and extracts the tagged named entity recognition tags. For example, the NER tag extractor (110) tags words (entities) included in chat room information into specific categories using a named entity recognition (NER) technique, which is a natural language processing (NLP) technique. At this time, the words (entities) are tagged using a named entity recognition tag that is predefined in a crime attribute extended named entity recognition schema stored in the storage device (120).

[0027] The crime investigation analyzer (130) generates knowledge graphs of an S (subject) P (predicate) O (object) structure based on the tagged named entity recognition tags extracted by the NER tag extractor (110). The crime investigation analyzer (130) can identify and recognize words tagged with the named entity recognition tags representing people as the same person in the knowledge graphs having a similarity of more than a certain threshold. Additionally, the crime investigation analyzer (130) can cluster the words tagged with the named entity recognition tags identified and recognized as the same person.

[0028] The display device (140) can display the clustered words identified and recognized as the same person by the crime investigation analyzer (130) to provide them to a user (e.g., investigator, etc.).

[0029]

[0030] Figure 2 is a diagram illustrating the crime attribute extended named entity recognition schema used to extract NER tags in the extended NER tag extractor of Figure 1.

[0031] Referring to Fig. 2, the crime attribute extended named entity recognition schema defines a named entity recognition schema classified into major category tags, medium category tags, and minor category tags.

[0032] For example, major category tags may include 'person', 'transaction', 'place', and 'action'. Medium category tags include 'name', 'title', 'age', 'transaction date and time', 'final transaction location', 'transaction item', 'transaction method', 'address', 'place name', 'building name', 'sale', 'plan', 'confirmation', and 'meeting'. Minor category tags include job title, alias, chat name, subject name, address, account, product, cash, coin, drugs, other, direct delivery, online transaction, coin transaction, courier, remittance, purchase, sale, and brokerage.

[0033] Here, 'Name', 'Title', and 'Age' are classified as sub-tags of the main category tag 'Person'; 'Transaction Date / Time', 'Final Transaction Location', 'Transaction Item', and 'Transaction Method' are classified as sub-tags of the main category tag 'Transaction'; and 'Address', 'Place Name', and 'Building Name' are classified as sub-tags of the main category tag 'Place'. The minor category tags include 'Job Title', 'Alias', 'Conversation Name', 'Address', 'Account', 'Product', 'Cash', 'Coin', 'Drugs', 'Direct Delivery', 'Online Transaction', 'Courier', 'Remittance', 'Coin Transaction', 'Drug Transaction', 'Courier', 'Remittance', 'Purchase', 'Sale', and 'Brokerage'. Here, 'Job Title', 'Alias', 'Computer Name', and 'Subject Name' are classified as sub-categories of the medium category tag 'Title'. 'Address or Account' is classified as a sub-category tag of the medium category tag 'Final Transaction Location'. 'Product, Cash, Coin, Drugs, Other' are classified as sub-categories of the medium category tag 'Traded Items'. 'Direct Delivery, Online Transaction, Coin Transaction' are classified as sub-categories of the medium category tag 'Transaction Method'. Here, 'Courier, Remittance' can be further classified as sub-attributes of 'Online Transaction'. 'Purchase, Sale, Brokerage' can be classified as sub-categories of the medium category tag 'Trading'.

[0034] Words included in the chat room information are tagged and extracted by an NER tag extractor into extended NER tags (extended NER tag information) as illustrated in Fig. 3, according to the crime attribute extended named entity recognition schema of Fig. 2. For example, in a chat room, the word 'self' (entity) can be tagged as Person (Major Category) - Title (Medium Category) - Alias ​​(Minor Category). In a chat room, 'going to work' can be tagged as Action (Major Category) - Other (Medium Category). In a chat room, the word 'colleague' (entity) can be tagged as Person (Major Category) - Title (Medium Category) - Other (Minor Category). In a chat room, the word 'Assistant Manager Park of Team 1' (entity) can be tagged as Person (Major Category) - Title (Medium Category) - Chat Name (Minor Category). In a chat room, the words (entities) 'Company' and 'Manager's Office' can be tagged as Location (Major Category) - Other (Medium Category).

[0035]

[0036] Figure 4 is an exemplary configuration diagram of the crime investigation analyzer of Figure 1.

[0037] Referring to FIG. 4, the crime investigation analyzer (130) generates knowledge graphs of an S (subject) P (predicate) O (object) structure based on the tagged named entity recognition tags extracted by the NER tag extractor (110) as described above, and can identify and recognize words tagged with the named entity recognition tags representing people as the same person in the knowledge graphs having a similarity of more than a certain threshold. Additionally, the crime investigation analyzer (130) clusters the words tagged with the named entity recognition tags that have been identified and recognized as the same person.

[0038] To this end, the crime investigation analyzer (130) may include a knowledge graph generator (132), a similarity calculator (134), and a clusterer (136).

[0039] The knowledge graph generator (132) analyzes words (entities) tagged with named entity recognition (NER) tags by the NER tag extractor (110) and classifies them into subjects (S), predicates (P), and objects (O), and connects the classified subjects (S), predicates (P), and objects (O) to generate knowledge graphs of an SPO structure.

[0040] To classify words (entities) into subjects (S), predicates (P), and objects (O), various techniques can be used, including syntactic parsing algorithms, semantic role labeling (SRL) algorithms, and deep learning-based SPO extraction algorithms.

[0041] A parsing algorithm is a process of analyzing the grammatical structure of a sentence containing words (entities) to identify the role of each word. In this process, the sentence can be represented in a tree form to extract components such as the subject (S), predicate (P), and object (O).

[0042] The semantic role analysis algorithm is the process of identifying the semantic roles that the subject (S), predicate (P), and object (O) play in a sentence. While the aforementioned syntactic analysis focuses on grammatical structure, semantic role analysis focuses on the semantic role that each component plays in the sentence.

[0043] Deep learning-based SPO extraction algorithms use deep learning-based models, such as Transformers, to extract SPO structures. Examples of deep learning-based models include BERT (Bidirectional Encoder Representations from Transformers) and Attention Mechanisms.

[0044] The knowledge graphs generated by the knowledge graph generator (132) are implemented in the form of an SPO knowledge database and input into a similarity calculator (134).

[0045] The similarity calculator (134) calculates the similarity of knowledge graphs generated by the knowledge graph generator (132) and identifies and recognizes words (entities) tagged with named entity recognition tags representing people in knowledge graphs having a similarity greater than a certain threshold as the same person.

[0046] To this end, the similarity calculator (134) identifies and recognizes words (entities) tagged with named entity tags representing people as the same person by calculating the similarity between words (entities) tagged with named entity tags representing people in knowledge graphs and surrounding words (surrounding entities), that is, words (entities) tagged with named entity tags that do not represent people.

[0047] For example, the similarity calculator (134) calculates the similarity between the word tagged with the named entity tag that does not represent a person in the first knowledge graph and the word tagged with the named entity tag that does not represent a person in the second knowledge graph, and if the calculated similarity is greater than or equal to the specific threshold, the word tagged with the named entity tag that represents a person in the first knowledge graph and the word tagged with the named entity tag that represents a person in the second knowledge graph are identified and recognized as the same person.

[0048] To explain simply, assuming a first knowledge graph composed of S1P1O1 and a second knowledge graph composed of S2P2O2, S1 and It is assumed that S2 consists entirely of words tagged with named entity recognition tags representing people. For example, assuming that S1 is the word 'Hong Gil-dong' tagged as a person and S2 is also tagged as the word 'Dr. Hong' tagged as a person, to determine whether 'Hong Gil-dong' and 'Dr. Hong' are identical, the similarity between P1 (predicate), a surrounding word of S1 (subject) in the first knowledge graph, and P2 (predicate), a surrounding word of S2 (subject) in the second knowledge graph, is calculated, or the similarity between O1 (object), a surrounding word of S1 (subject) in the first knowledge graph, and O2 (object), a surrounding word of S2 (subject) in the second knowledge graph is calculated.

[0049] Typically, even if multiple conversers in a chat room use different words (entities) to refer to a specific person, it is highly likely that they will use similar words (entities) for the surrounding words (surrounding entities) that refer to that specific person, such as predicates (O) or objects (O). Based on this, in an embodiment of the present invention, when multiple conversers use different words (entities) to refer to a specific person, the word (entity) referring to that specific person is identified and recognized as the same person by calculating the similarity of the surrounding words. It should be noted here that the surrounding words are words (entities) tagged with named entity recognition tags that do not represent people, for example, predicates (O) and / or objects (O). If the object (O) is a word tagged with a named entity recognition tag representing a person, that object (O) is excluded from the similarity calculation.

[0050] Word embedding and distance metric methods can be used to calculate the similarity between predicates or objects. Word embedding and distance metric methods are algorithmic techniques that can be usefully applied to calculate the semantic similarity between two words or sentences.

[0051] Word embedding is a method of representing words as fixed-dimensional vectors, enabling the measurement of semantic similarity between words in a vector space. This allows for the calculation of similarity between predicates or objects. Word embedding methods include Word2Vec and GloVe (Global Vectors for Word Representation).

[0052] Word2Vec converts the meanings of words into high-dimensional vectors and learns that words used in similar contexts have close vector values. There are two models: CBOW (Continuous Bag of Words) and Skip-Gram. These models can measure the similarity between predicates or objects by calculating cosine similarity between embedded vectors. Cosine similarity measures similarity based on the angle between two vectors.

[0053] GloVe learns word vectors based on a co-occurrence matrix. It generates vectors by learning how often words appear together within a context. Similar to Word2Vec, the similarity between two words can be calculated using the cosine similarity between GloVe vectors.

[0054] The clusterer (136) clusters words tagged with named entity recognition tags identified and recognized as the same person. For clustering, the K-Means algorithm may be used.

[0055] Meanwhile, when the word identified and recognized as the same person is tagged with a named entity recognition tag such as 'person-name', that is, when it includes a real name, the clusterer (136) clusters by binding the word tagged with the named entity recognition tag (word including a real name) and the found contact number when a contact number matching the word tagged with the named entity recognition tag such as 'person-name' (word including a real name) is found in the contact information obtained from the digital forensic equipment (10 in FIG. 1).

[0056] If the word identified and recognized as the same person is tagged with a named entity recognition tag such as 'person-title-username'—that is, if it does not include a real name—it is clustered into a separate cluster distinct from the cluster formed by words bound to contact numbers. Of course, if a contact number stored as a title or username rather than a real name is found, that contact number is bound to the word tagged with a named entity recognition tag such as 'person-title-username' and clustered.

[0057] In this way, the clusterer (136) clusters words referring to a specific person bound to a contact number and words not bound to a contact number into different clusters, and provides these clusters to a user (e.g., an investigator) through a display device (140).

[0058] Clusters bound to contact information consist of words clearly associated with that contact, allowing them to clearly provide details such as a specific individual's name, nickname, position, and relationships. This greatly assists investigators in clearly identifying that person's identity. Conversely, clusters not bound to contact information are not directly linked to the contact but can provide surrounding information or context. This makes it easier to detect the possibility that the same person has been referred to by different names or aliases.

[0059] As described above, the device (100) according to an embodiment of the present invention enables automated same-person analysis through clustered data, thereby allowing for more accurate and faster identification of the same person in chaotic chat room information.

[0060]

[0061] FIG. 5 is a flowchart of a method for identifying and recognizing the same person within a messenger chat room according to an embodiment of the present invention.

[0062] Referring to FIG. 5, first, in S510, the extractor (110) tags the words included in the chat room information with named entity recognition tags, and the process of extracting the tagged named entity recognition tags is performed.

[0063] Next, in S520, the crime investigation analyzer (130) performs a process of generating knowledge graphs of the S(subject)P(predicate)O(object) structure based on the tagged named entity recognition tag.

[0064] Next, in S530, a process is performed in which a crime investigation analyzer (130) identifies and recognizes words tagged with the named entity recognition tag representing a person as the same person in knowledge graphs having a similarity level greater than a specific threshold (high similarity level) among the knowledge graphs.

[0065] In an embodiment, S510 includes a process of tagging the words using a predefined named entity recognition tag in a crime attribute extended named entity recognition schema.

[0066] In an embodiment, the named entity recognition tag includes a major category tag, a medium category tag, and a minor category tag; the major category tag includes person, transaction, place, and action; the medium category tag includes name, title, age, transaction date and time, transaction final destination, traded item, transaction method, address, place name, building name, sale, plan, confirmation, and meeting; and the minor category tag includes job title, alias, conversation name, address, account, product, cash, coin, drug, direct delivery, online transaction, courier, remittance, coin transaction, drug transaction, courier, remittance, purchase, sale, and brokerage.

[0067] In an embodiment, S520 includes a knowledge graph generator (132) that analyzes words tagged with the named entity recognition tag and classifies them into a subject (S), a predicate (P), and an object (O), and a process of generating the knowledge graph by connecting the classified subject (S), predicate (P), and object (O).

[0068] In an embodiment, S530 includes a similarity calculator (134) that calculates the similarity between a word tagged with the named entity recognition tag that does not represent a person in a first knowledge graph and a word tagged with the named entity recognition tag that does not represent a person in a second knowledge graph, and if the calculated similarity is greater than or equal to the specific threshold, the word tagged with the named entity recognition tag that represents a person in the first knowledge graph and the word tagged with the named entity recognition tag that represents a person in the second knowledge graph are identified and recognized as the same person.

[0069] In an embodiment, the process of calculating the similarity includes, when the word tagged with the named entity recognition tag representing the person is the subject (S), the process of calculating the similarity between the words tagged with the named entity recognition tag that do not represent the person included in the predicate (P) or the object (O).

[0070] After S530, the clusterer may further perform the process of clustering words tagged with the named entity recognition tag identified and recognized as the same person, and the display device may further perform the process of displaying the clustered words to provide them to the user.

[0071]

[0072] FIG. 6 is an exemplary configuration diagram of a computing device for implementing the device of FIG. 1 and / or the method of FIG. 5.

[0073] Referring to FIG. 6, the computing device (600) includes a processor (610), memory (620), input / output devices (I / O Devices) (630), a power supply (640), a communication device (650), and a storage device (660).

[0074] The processor (610) includes a central processing unit (CPU) and / or a graphics processing unit (GPU) of the computing device (600) and processes data and executes instructions. The processor (610) may be the entity performing the steps described in the method of FIG. 6. Additionally, the processor (610) may control, execute, and manage the operation of the blocks illustrated in FIG. 1 and 4.

[0075] Memory (620) is a device that temporarily or / and permanently stores various program data, instructions, and / or intermediate data and result data generated at each step necessary for performing the steps described in the method of FIG. 6, and may include RAM (volatile memory), ROM (non-volatile memory), etc. Additionally, memory (620) may temporarily or / and permanently store intermediate data and result data generated in the blocks illustrated in FIG. 1 and 4.

[0076] The input / output device (630) may include a keyboard, mouse, touchscreen and / or display device (140).

[0077] The power supply unit (640) is a device that supplies power to the peripheral components (610–630, 650 and 660).

[0078] The communication device (650) may be a network interface for communicating with an external device or system and supports various forms of communication such as wired communication (USB, Ethernet) or wireless communication (Bluetooth, Wi-Fi, mobile communication).

[0079] The storage device (660) may be a non-volatile storage medium such as a hard disk or an SSD, and may be a device for storing programs and applications for performing the steps described in the method of FIG. 6, and an operating system that supports the execution of such programs and applications. The operating system supports the processor (610) to efficiently process the instructions required for the present invention.

[0080] The system bus (670) is a device that supports data communication of peripheral components (610–660).

[0081]

[0082] The embodiments disclosed in this specification should be considered for illustrative purposes rather than for limiting purposes. The scope of the invention is defined by the claims, not by the foregoing description, and all variations within the scope of the claims should be interpreted as being included in the invention.

[0083]

[0084] The present invention can be utilized in various industries to identify criminal circumstances and prove charges through the analysis of data extracted via mobile phone forensics in criminal investigations.

Claims

1. A step in which an extractor tags words included in chat room information with named entity recognition tags and extracts the tagged named entity recognition tags; and A crime investigation analyzer generates knowledge graphs with an S(subject)P(predicate)O(object) structure based on the tagged named entity recognition tags, and identifies and recognizes words tagged with the named entity recognition tags representing people as the same person in knowledge graphs having a similarity of more than a specific threshold among the knowledge graphs. A method for identifying and recognizing the same person within a messenger chat room, including 2. In Paragraph 1, The step of extracting the tagged named entity recognition tag is, A step of tagging the above words using predefined named entity recognition tags in the crime attribute extended named entity recognition schema. A method for identifying and recognizing the same person within a messenger chat room, including 3. In Paragraph 1, The above named entity recognition tag is, Includes major category tags, medium category tags, and minor category tags, The above major category tags include people, transactions, places, and actions, and The above medium category tags include name, title, age, transaction date and time, final transaction location, traded item, transaction method, address, place name, building name, sale, plan, confirmation, and meeting, and The above subcategory tags include job title, alias, username, address, account, product, cash, coin, drugs, direct delivery, online transaction, courier, remittance, coin trading, drug trading, courier, remittance, purchase, sale, and brokerage. A method for identifying and recognizing the same person within a messenger chat room, including 4. In Paragraph 1, The above identification and recognition step is, A knowledge graph generator analyzes words tagged with the named entity recognition tags and classifies them into subjects (S), predicates (P), and objects (O); and The knowledge graph generator generates the knowledge graph by connecting the classified subject (S), predicate (P), and object (O). A method for identifying and recognizing the same person within a messenger chat room, including 5. In Paragraph 1, The above identification and recognition step is, A similarity calculator calculates the similarity between a word tagged with the named entity recognition tag that does not represent a person in a first knowledge graph and a word tagged with the named entity recognition tag that does not represent a person in a second knowledge graph; If the calculated similarity is greater than or equal to the specific threshold, the step of identifying and recognizing the word tagged with the named entity recognition tag representing a person in the first knowledge graph and the word tagged with the named entity recognition tag representing a person in the second knowledge graph as the same person. A method for identifying and recognizing the same person within a messenger chat room, including 6. In Paragraph 5, The step of calculating the above similarity is, When the word tagged with the named entity recognition tag representing the person is the subject S, the step of calculating the similarity between the words tagged with the named entity recognition tag that do not represent the person included in the predicate (P) or the object (O). A method for identifying and recognizing the same person within a messenger chat room, including 7. In Paragraph 1, The above identification and recognition step is, A clusterer clusters words tagged with the named entity recognition tag identified and recognized as the same person; and A display device displays the clustered words to provide them to the user. A method for identifying and recognizing the same person within a messenger chat room, including 8. An extractor that tags words included in chat room information with named entity recognition tags and extracts the tagged named entity recognition tags; and A crime investigation analyzer that generates knowledge graphs with an S (subject) P (predicate) O (object) structure based on the aforementioned tagged named entity recognition tags, and identifies and recognizes words tagged with the aforementioned named entity recognition tags representing people as the same person in knowledge graphs having a similarity level above a specific threshold. A device for identifying and recognizing the same person within a messenger chat room, including

Images