Apparatus for detecting false information in document through content area analysis

Abstract

According to an embodiment of the present invention, an apparatus for detecting false information in a document by analyzing a content area of the document comprises: a document data reception unit for receiving document data from a user terminal used by a user; and an area recognition unit for distinguishing and recognizing a text area, a table area, and an image area in the received document data.

Description

Document False Information Detection Device via Content Area Analysis

[0001] The present invention relates to a device for detecting false information in a document through content area analysis, and more specifically, to a device for detecting false information in a document through content area analysis that extracts a content area within a document and analyzes the relationship between the type of the extracted content area and surrounding content areas to determine false information in the document.

[0002] "This research was conducted as a result of the University ICT Research Center program of the Ministry of Science and ICT and the Institute for Information & Communications Technology Planning & Evaluation." (RS-2024-00436936)

[0003] The rapid advancement of AI technology in recent years has led to an increase in the generation of disinformation and deepfake content. In particular, advanced analysis techniques are required to detect malicious information within complex and large-scale document data; however, existing technologies have limitations, such as focusing on a single content type or failing to support real-time responses for administrators regarding detected content. To address these issues, this invention describes a device for detecting false information in documents through content area analysis, which extracts all content areas within a document and analyzes the types of the extracted content areas and the relationships between surrounding content areas to identify false information within the document.

[0004] [Prior Literature]

[0005] Registered Patent 10-2340542

[0006] The present invention relates to a device for detecting false information in a document through content area analysis, and more specifically, to a device for detecting false information in a document through content area analysis that extracts a content area within a document and analyzes the relationship between the type of the extracted content area and surrounding content areas to determine false information in the document.

[0007] A document false information detection device that detects false information in a document by analyzing the content area of ​​a document comprises: a document data receiving unit that receives document data from a user terminal used by a user; an area recognition unit that distinguishes and recognizes text areas, table areas, and image areas in the received document data; a paragraph recognition unit that distinguishes between sentences and non-sentences based on terminal words in the recognized text area, recognizes an area containing at least two consecutive sentences without line breaks as a paragraph, and recognizes the remaining area excluding the recognized paragraph as a non-paragraph; a paragraph group generating unit that creates a paragraph group by grouping paragraphs, table areas, and image areas placed between an adjacent pair of non-paragraphs; a summary paragraph generating unit that generates a summary paragraph containing at least one sentence containing the content explained by each paragraph included in each paragraph group and the table area using a natural language processing algorithm; a summary paragraph similarity calculating unit that compares summary paragraphs using all possible combinations of two within each paragraph group, determines similarity, and calculates the mutual similarity between two summary paragraphs; and when the similarity calculated within each paragraph group is less than a set value, the corresponding pair of summary It includes a first false information detection unit that extracts paragraphs and, if there are duplicate summary paragraphs among the extracted summary paragraphs, detects a text area or table area corresponding to the duplicate summary paragraph as the first false information and transmits it to a user terminal.

[0008] A document false information detection device according to one embodiment of the present invention further includes a matching information transmission unit that transmits matching information to a user terminal by matching the detected first false information to the target paragraph group when, as a result of the paragraph similarity calculation unit comparing the summary paragraph of the detected first false information with each of the summary paragraphs in the paragraph group adjacent to the upper and lower sides containing the detected first false information and calculating the similarity, there exists a target paragraph group composed of summary paragraphs in which all values ​​are greater than or equal to a set value; and a document data update unit that updates document data by reflecting the input information and transmits the updated document data to the user terminal when the user responds to the matching information and inputs a location to place a text area or table area corresponding to the detected first false information within the target paragraph through the user terminal.

[0009] A document false information detection device according to one embodiment of the present invention further comprises: a first sentence extraction unit that extracts first sentences included in a summary paragraph when the summary paragraph included in each paragraph group is an explanatory paragraph; a second sentence extraction unit that analyzes the extracted first sentences to recognize words included in each of the sentences as a subject, object / complement, or predicate, and extracts a second sentence composed of at least two or more components, including the subject, that are identical or similar to the first sentence extracted through web crawling; and a second false information detection unit that compares the first sentence with a matching second sentence and detects the first sentence as non-false information if the two components, including the subject, are identical or similar and the remaining component, the target component, is different, and the word of the target component of the first sentence and the word of the target component of the second sentence are compatible, and detects the first sentence as second false information if they are incompatible and transmits it to a user terminal.

[0010] The second sentence extracted by the second sentence extraction unit is a sentence composed of at least two identical components, including the subject of the first sentence, or words in which the semantic relationship belongs to a superior-subordinate relationship, and the document data update unit updates the document data by deleting the text area or table area corresponding to the second false information and transmits the updated document data to the user terminal.

[0011] A document false information detection device according to one embodiment of the present invention further includes a metadata extraction unit that extracts metadata from image data of a recognized image area, and a third false information detection unit that analyzes the production date, time, modification history, and software used in the extracted metadata to determine whether deepfake technology has been applied to the image data of the recognized image area, and if it is determined that deepfake technology has been applied to the recognized image area, detects the image area as third false information and transmits it to a user terminal, and the document data update unit updates the document data by deleting the image area corresponding to the third false information and transmits the updated document data to the user terminal.

[0012] The present invention can improve the accuracy of document data by analyzing received document data to divide the document data into multiple paragraph groups, analyzing text areas and table areas included within each paragraph group, and determining that a specific paragraph or table is false information if it is determined to be incorrect information within the paragraph group, deleting the determined false information, or moving the false information to the paragraph group if the false information is compared with the content of the paragraph group adjacent to the paragraph group to which the determined false information belongs and a similarity level greater than a set value is obtained.

[0013] The present invention can improve the accuracy of document data by determining whether sentences in summary paragraphs included in each paragraph group are false through web crawling, rapidly deleting text areas or table areas corresponding to summary paragraphs determined to be false, and determining whether image data in recognized image areas is false through metadata, and rapidly deleting image areas determined to be false.

[0014] FIG. 1 is a block diagram of a system for detecting false information in a document through content area analysis according to one embodiment of the present invention.

[0015] FIG. 2 is a block diagram of a document false information detection device according to one embodiment of the present invention.

[0016] FIG. 3 is a block diagram of a false information detection unit according to one embodiment of the present invention.

[0017] FIG. 4 is a block diagram of a sentence extraction unit according to one embodiment of the present invention.

[0018] A document false information detection device according to an embodiment of the present invention, which detects false information in a document by analyzing the content area of ​​a document, comprises: a document data receiving unit that receives document data from a user terminal used by a user; an area recognition unit that distinguishes and recognizes a text area, a table area, and an image area in the received document data; a paragraph recognition unit that distinguishes between sentences and non-sentences based on a terminal word in the recognized text area, recognizes an area containing at least two consecutive sentences without a line break as a paragraph, and recognizes the remaining area excluding the recognized paragraph as a non-paragraph; a paragraph group generating unit that generates a paragraph group by grouping paragraphs, a table area, and an image area placed between an adjacent pair of non-paragraphs; a summary paragraph generating unit that generates a summary paragraph containing at least one sentence containing the content explained by each paragraph included in each paragraph group and the table area using a natural language processing algorithm; and a summary paragraph similarity calculating unit that compares summary paragraphs in all possible combinations of two within each paragraph group, determines similarity, and calculates the similarity between two summary paragraphs.

[0019] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0020] Throughout the specification, when a part is described as being "connected" to another part, this includes not only cases where they are "directly connected," but also cases where they are "electrically connected" with other elements interposed between them. Furthermore, when a part is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. The present invention will now be described in detail with reference to the attached drawings.

[0021] FIG. 1 is a block diagram of a document false information detection system (1000) through content area analysis according to one embodiment of the present invention.

[0022] Referring to FIG. 1, a system for detecting false information in a document through content area analysis according to one embodiment of the present invention (1000) may include a device for detecting false information in a document (200) connected to a user terminal (100) and a network (400).

[0023] The user terminal (100) may be a terminal used by a person who intends to identify false information from documents. For example, the user terminal (100) may be a terminal used by a person who reviews papers, reports, and marketing materials. The user may be a person who intends to improve the accuracy of documents by quickly determining false information in documents through the present invention.

[0024] The user terminal (100) may be a smartphone. However, it is not limited thereto, and the user terminal (100) may include electronic devices such as a general desktop computer, navigation system, laptop, digital broadcasting terminal, PDA (Personal Digital Assistants), PMP (Portable Multimedia Player), tablet PC, etc. The electronic device may have one or more general or special purpose processors, memory, storage, and / or networking components (wired or wireless).

[0025] The document false information detection device (200) receives document data from a user terminal (100), recognizes text areas, table areas, and image areas in the received document data, and analyzes the recognized areas to detect false information. The document false information detection terminal may be a server and may be implemented in the form of an application within the user terminal (100). The document false information detection will be explained in more detail in FIGS. 2 to 4.

[0026] The communication method of the network (400) is not limited and may include not only communication methods utilizing communication networks (e.g., mobile communication networks, wired online, wireless online, broadcasting networks) that the network (400) may include, but also short-range wireless communication between devices. For example, the network (400) may include one or more arbitrary networks (400) among networks (400) such as PAN (personal area network), LAN (local area network), CAN (campus area network), MAN (metropolitan area network), WAN (wide area network), BBN (broadband network), and online.

[0027] FIG. 2 is a block diagram of a document false information detection device (200) according to one embodiment of the present invention, FIG. 3 is a block diagram of a false information detection unit (207) according to one embodiment of the present invention, and FIG. 4 is a block diagram of a sentence extraction unit (210) according to one embodiment of the present invention.

[0028] Referring to FIGS. 2 to 4, a document false information detection device (200) according to one embodiment of the present invention may include a document data receiving unit (201), a region recognition unit (202), a paragraph recognition unit (203), a paragraph group generating unit (204), a summary paragraph generating unit (205), a summary paragraph similarity calculation unit (206), a false information detection unit (207), a matching information transmission unit (208), a document data update unit (209), a sentence extraction unit (210), and a metadata extraction unit (211).

[0029] The document data receiving unit (201) can receive document data from the user terminal (100). The present invention can analyze the received document data and detect false information in the document data.

[0030] The area recognition unit (202) can distinguish and recognize text areas, table areas, and image areas in the received document data. The area recognition unit (202) can distinguish and recognize text areas, table areas, and image areas in the received document data by referring to sample data for characters, sample data for tables, and sample data for images.

[0031] The paragraph recognition unit (203) distinguishes between sentences and non-sentences based on a terminal word in the recognized text area, recognizes an area containing at least two consecutive sentences without line breaks as a paragraph, and recognizes the remaining area excluding the recognized paragraph as a non-sentence. Since sentences have terminal words such as "~da," the paragraph recognition unit (203) can recognize an area containing at least two consecutive sentences as a paragraph by taking this into consideration.

[0032] The paragraph group generating unit (204) can generate a paragraph group by grouping paragraphs, table areas, and image areas placed between adjacent pairs of non-paragraphs. Paragraphs with similar content, tables, and images may be placed between non-sentences such as subheadings and adjacent non-sentences, and the paragraph group generating unit (204) can generate a paragraph group by grouping these paragraphs, tables, and images.

[0033] The summary paragraph generation unit (205) can generate a summary paragraph containing at least one sentence of the content explained by each paragraph included in each paragraph group and the table area using a Natural Language Processing Algorithm. As an example of the present invention, the summary paragraph generation unit (205) can generate a summary paragraph of the content explained by each paragraph included in each paragraph group and the table area using a known Natural Language Processing technique using a deep learning model algorithm. However, the method of generating the summary paragraph is not limited to this, and various known summarization methods may be utilized in the process of generating the summary paragraph.

[0034] The summary paragraph similarity calculation unit (206) can compare summary paragraphs in all possible combinations of 2 within each paragraph group and determine similarity to calculate the similarity between 2 summary paragraphs. For example, if there are 3 summary paragraphs A, B, and C within a paragraph group, the summary paragraph similarity calculation unit (206) can compare A and B, B and C, and A and C to calculate the similarity between A and B, B and C, and A and C.

[0035] The summary paragraph similarity calculation unit (206) can calculate the similarity between two summary paragraphs by quantifying the similarity between summary paragraphs in a manner such that the subject and predicate of the sentences included in the summary paragraphs are given the highest weight, and the object is given the next highest weight, thereby determining whether there is similarity between the summary paragraphs.

[0036] The false information detection unit (207) may include a first false information detection unit (271), a second false information detection unit (272), and a third false information detection unit (273).

[0037] The first false information detection unit (271) can extract a pair of summary paragraphs when the similarity calculated within each paragraph group is less than a set value, and if there is a duplicate summary paragraph among the extracted summary paragraphs, it can detect a text area or table area corresponding to the duplicate summary paragraph as the first false information and transmit it to the user terminal (100).

[0038] Generally, a paragraph group consists of paragraphs and table areas that have similar content. If any single paragraph or table area does not have any content similar to the remaining paragraphs or table areas, it can be considered that the paragraph is incorrectly placed in the paragraph group. The first false information detection unit (271) can detect duplicate summary paragraphs among the extracted summary paragraphs as first false information and transmit them to the user terminal (100) by taking these points into consideration. For example, if a paragraph group contains summary paragraphs A, B, and C, and A and B have a similarity greater than or equal to a set value, A and C have a similarity less than or equal to a set value, and B and C have a similarity less than a set value, the first false information detection unit (271) extracts A and C during the similarity calculation process for A and C and extracts B and C during the similarity calculation process for B and C. Since the extraction result shows that C is a duplicate summary paragraph, the text area or table area corresponding to C can be detected as first false information. As an example of the present invention, the set value may be determined by the administrator of the present invention.

[0039] The matching information transmission unit (208) can compare the summary paragraph of the first false information detected by the summary paragraph similarity calculation unit (206) with the summary paragraph of the first false information detected by the summary paragraph similarity calculation unit (206) and the summary paragraphs in the paragraph groups adjacent to the upper and lower sides, and if there exists a target paragraph group composed of summary paragraphs in which all values ​​are greater than or equal to a set value as a result of calculating the similarity, the matching information that matches the detected first false information to the target paragraph group can be transmitted to the user terminal (100).

[0040] Since the summary paragraph of the first false information is a paragraph of false information that does not match the content of the paragraph group of the summary paragraph, and the content of the first false information should originally be placed in another paragraph group adjacent to the paragraph group, but may have become false information because it was incorrectly placed in the paragraph group, the summary paragraph similarity calculation unit (206) can calculate the similarity by comparing the summary paragraph of the first false information with each of the summary paragraphs in the paragraph groups adjacent to the upper and lower sides of the paragraph group in the received document data, taking these points into consideration. In the case where there is a target paragraph group composed of summary paragraphs in which all values ​​are greater than or equal to the set value, it is reasonable for the first false information to be placed in the target paragraph group rather than in the group paragraph where it was originally placed. Therefore, the matching information transmission unit (208) can generate matching information that matches the detected first false information to the target paragraph group and transmit it to the user terminal (100) in order to deliver this information to the user.

[0041] The document data update unit (209) can update the document data by reflecting the input information and transmit the updated document data to the user terminal (100) when the user inputs a location to place a text area or table area corresponding to the first false information detected through the user terminal (100) in response to the matching information. However, if the previously described target paragraph group does not exist, or if the user checks the matching information and wishes to delete the first false information rather than place it in the target paragraph group, and therefore does not input a location to place the first false information, the document data update unit (209) can delete the first false information to update the document data and transmit the updated document data to the user terminal (100).

[0042] As such, the present invention can improve the accuracy of document data by analyzing received document data to divide the document data into multiple paragraph groups, analyzing text areas and table areas included within each paragraph group, and determining that a specific paragraph or table is false information if it is determined to be incorrect information within the paragraph group, deleting the determined false information, or moving the false information to the corresponding paragraph group if the false information is compared with the content of the paragraph group adjacent to the paragraph group to which the determined false information belongs and a similarity level exceeding a set value is obtained.

[0043] The sentence extraction unit (210) may further include a first sentence extraction unit (281) and a second sentence extraction unit (282).

[0044] The first sentence extraction unit (281) can extract first sentences included in a summary paragraph if the summary paragraph included in each paragraph group is an explanatory paragraph. The first sentence extraction unit (281) can extract first sentences from a summary paragraph if the summary paragraph included in each paragraph group is not a paragraph of assertive content but a paragraph of explanatory content.

[0045] The second sentence extraction unit (282) analyzes the extracted first sentences and recognizes the words included in each of the sentences as subjects, objects / complements, or predicates, and can extract a second sentence consisting of at least two or more identical or similar words, including the subject of the first sentence extracted through web crawling.

[0046] The second false information detection unit (272) can detect the first sentence as non-false information if the first sentence is compared with a second sentence that matches the first sentence, and the two components including the subject are composed of identical or similar words, and the remaining component, the target component, is composed of different words, and the words of the target component of the first sentence and the words of the target component of the second sentence are compatible words, and if the words are incompatible words, the first sentence is detected as second false information and transmitted to the user terminal (100).

[0047] The second sentence extracted by the second sentence extraction unit (282) may be a sentence in which at least two or more components, including the subject, are identical to the first sentence or are composed of words in which the semantic relationship belongs to a superior-subordinate relationship.

[0048] For example, if a sentence is found through crawling in which the subject is the same as the first sentence and the object is the same or consists of similar words belonging to a superior-subordinate relationship, the second sentence extraction unit (282) can extract the sentence as the second sentence.

[0049] And if the predicate of the first sentence consists of a word that is incompatible with the predicate of the second sentence from which it was extracted, the second false information detection unit (272) can detect the first sentence as second false information. For example, if the first sentence and the second sentence consist of similar words in which the subject and object are identical or belong to a superior-subordinate relationship, and the predicate of the first sentence is "rises" and the predicate of the second sentence is "descends," the words included in both predicates are incompatible with each other, so in this case, the second false information detection unit (272) can detect the first sentence containing the predicate "rises" as second false information.

[0050] The document data update unit (209) can update the document data by deleting the text area or table area corresponding to the second false information and transmit the updated document data to the user terminal (100).

[0051] The metadata extraction unit (211) can extract metadata from image data of a recognized image area. Metadata refers to data describing other data structured in a consistent manner so that a large amount of data can be collected, stored, and analyzed over time.

[0052] The third false information detection unit (273) analyzes the production date, time, modification history, and software used in the extracted metadata to determine whether deepfake technology has been applied to the image data of the recognized image area, and if it is determined that deepfake technology has been applied to the recognized image area, it can detect the image area as the third false information and transmit it to the user terminal (100). That is, the third false information detection unit (273) can detect from the extracted metadata whether the image data of the recognized image area in the document data has been manipulated, and if it is detected as manipulated image data, it can detect the image area as the third false information and transmit it to the user terminal (100).

[0053] The document data update unit (209) can update the document data by deleting the image area corresponding to the third false information and transmit the updated document data to the user terminal (100).

[0054] As such, the present invention can improve the accuracy of document data by determining whether the sentences of summary paragraphs included in each paragraph group are false through web crawling, rapidly deleting text areas or table areas corresponding to summary paragraphs determined to be false, and determining whether the image data of image areas recognized through metadata is false, and rapidly deleting image areas determined to be false.

[0055] The embodiments described above are for illustrative purposes only, and those skilled in the art will understand that the embodiments described above can be easily modified into other specific forms without altering the technical concept or essential features of the embodiments described above. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0056] The scope of protection sought through this specification is defined by the claims set forth below rather than by the detailed description, and should be interpreted to include all modifications or variations derived from the meaning and scope of the claims and the concept of equivalents.

Claims

1. A document false information detection device that detects false information in a document by analyzing the content area of ​​a document, A document data receiving unit that receives document data from a user terminal used by a user; Area recognition unit that distinguishes and recognizes text areas, table areas, and image areas in received document data; A paragraph recognition unit that distinguishes between sentences and non-sentences based on terminal words in a recognized text area, recognizes an area containing at least two consecutive sentences without line breaks as a paragraph, and recognizes the remaining area excluding the recognized paragraphs as non-sentences; A paragraph group generation unit that generates a paragraph group by grouping paragraphs, a table area, and an image area placed between an adjacent pair of non-paragraphs; A summary paragraph generation unit that uses a natural language processing algorithm to generate a summary paragraph containing at least one sentence of the content described by each paragraph included in each paragraph group and the table area; Summary paragraph similarity calculation unit that compares summary paragraphs using all possible 2 combinations within each paragraph group, determines similarity, and calculates the mutual similarity between 2 summary paragraphs; and A document false information detection device through content area analysis comprising a first false information detection unit that extracts a corresponding pair of summary paragraphs when the similarity calculated within each paragraph group is less than a set value, and detects a text area or table area corresponding to the duplicate summary paragraph as the first false information and transmits it to a user terminal when there is a duplicate summary paragraph among the extracted summary paragraphs.

2. In Paragraph 1, A matching information transmission unit that, if a result of the paragraph similarity calculation unit comparing the summary paragraph of the detected first false information with each of the summary paragraphs in the paragraph groups adjacent to the upper and lower sides containing the paragraph group containing the summary paragraph of the detected first false information and calculating the similarity, and if there exists a target paragraph group composed of summary paragraphs where all values ​​are greater than or equal to a set value, transmits matching information that matches the detected first false information to the target paragraph group to a user terminal; and A document false information detection device through content area analysis, further comprising a document data update unit that updates document data by reflecting the input information and transmits the updated document data to the user terminal when a user inputs a location to place a text area or table area corresponding to the first false information detected through the user terminal within a target paragraph in response to matching information.

3. In Paragraph 1, A first sentence extraction unit that extracts first sentences included in a summary paragraph when the summary paragraph included in each paragraph group is an explanation paragraph; A second sentence extraction unit that analyzes the extracted first sentences to recognize words included in each of the sentences as subjects, objects / complements, or predicates, and extracts a second sentence composed of at least two or more identical or similar words, including the subject and the first sentence extracted through web crawling; and A document false information detection device through content area analysis, further comprising a second false information detection unit that compares a first sentence with a matching second sentence, detects the first sentence as non-false information if two components including the subject are composed of identical or similar words and the remaining component, the target component, is composed of different words, and if the word of the target component of the first sentence and the word of the target component of the second sentence are compatible words, and detects the first sentence as second false information if they are incompatible words, and transmits the result to a user terminal.

4. In Paragraph 3, The second sentence extracted by the second sentence extraction unit is a sentence composed of words in which at least two or more components, including the subject, are identical to the first sentence or belong to a hierarchical semantic relationship, and A document false information detection device through content area analysis, characterized in that the document data update unit updates the document data by deleting a text area or a table area corresponding to the second false information and transmits the updated document data to a user terminal.

5. In Paragraph 4, A metadata extraction unit that extracts metadata from image data of a recognized image region; and It further includes a third false information detection unit that analyzes the production date, time, modification history, and software used in the extracted metadata to determine whether deepfake technology has been applied to the image data of a recognized image area, and if it is determined that deepfake technology has been applied to the recognized image area, detects the image area as third false information and transmits it to a user terminal. A document false information detection device through content area analysis, characterized in that the document data update unit updates the document data by deleting the image area corresponding to the third false information and transmits the updated document data to a user terminal.

Images