Similar document search device and program

Abstract

To provide a similar document search device and a program which allow search of a document which fits a context.SOLUTION: A similar document search server 1 includes: an item-set storage unit 23 for storing, in association with stored documents, a plurality of item sets each being a combination of one item name and one item value corresponding thereto; an item-set extraction unit 14 for extracting a plurality of item sets from a target document image; and a similarity calculation unit 15 for calculating document similarity by using the plurality of item sets extracted by the item-set extraction unit 14 and the plurality of item sets corresponding to the plurality of stored documents stored in the item-set storage unit 23.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a similar document search device and a program. [Background technology]

[0002] Similar document search is a technique for searching for other documents with similar content or themes to a target document, and is used in various fields such as information retrieval and document classification. With the spread of the Internet, a huge amount of documents exists online, and there is a need to efficiently search for documents related to a specific theme or content. In this situation, since online documents exist in various data formats, it is necessary to digitize the documents so that they can be read as text. In addition, if the document is handed over on paper or as PDF data, for example, it must also be digitized so that it can be read as text. As a method for converting a document image into text, for example, OCR (Optical Character Recognition / Reader) can be used. As a method for searching for documents similar to a target document, a device has been disclosed that, by inputting a document, searches for documents similar to the document based on the importance of target words, which are important words in the document (for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6190904 Summary of the Invention [Problem to be solved by the invention]

[0004] In the method described in Patent Document 1, if the target word is used, for example, in a different context, the document output as the search result may be different from the one originally desired.

[0005] SUMMARY OF THE INVENTION It is therefore an object of the present invention to provide a similar document search device and program that enable document search based on context. [Means for solving the problem]

[0006] The present invention solves the above problems by the following means. A first invention is a similar document search device comprising: an item set storage unit that stores a plurality of item sets, which are combinations of a plurality of item names and item values ​​corresponding to each of the plurality of item names contained in a stored document, in correspondence with the stored document; an item set extraction means that extracts the plurality of item sets from a target document image; and a similarity calculation means that calculates document similarity using the plurality of item sets extracted by the item set extraction means and the plurality of item sets corresponding to each of the plurality of stored documents stored in the item set storage unit. A second invention is a similar document search device according to the first invention, wherein the similarity calculation means calculates the document similarity indicating the degree of match between the item name of the plurality of item sets extracted by the item set extraction means and the item name of the plurality of item sets corresponding to each of the plurality of saved documents stored in the item set storage unit. A third invention is a similar document search device according to the first or second invention, wherein the similarity calculation means calculates the document similarity indicating the degree of match between the plurality of item sets extracted by the item set extraction means and the plurality of item sets corresponding to each of the plurality of saved documents stored in the item set storage unit. A fourth invention is a similar document search device according to any one of the first to third inventions, further comprising a similar document output means for outputting the stored document corresponding to the plurality of item sets having the highest document similarity calculated by the similarity calculation means. A fifth invention is a similar document search device according to the fourth invention, wherein the similar document output means outputs the stored documents corresponding to the plurality of item sets whose document similarity calculated by the similarity calculation means is equal to or greater than a threshold value. A sixth invention is a similar document search device according to any one of the first to fifth inventions, further comprising: a document receiving means for receiving an input of the target document image; a character string obtaining means for obtaining a character string from the target document image received by the document receiving means; and an estimation means for estimating attribute information and correspondence of the character string from the position and content of the character string obtained by the character string obtaining means, wherein the item set extraction means extracts the plurality of item sets from the target document image based on the estimation result by the estimation means. A seventh invention is a similar document search device according to the sixth invention, wherein the estimation means estimates the attribute information and correspondence of the character string based on rules regarding the presence or absence of keywords included in the character string and / or the position of the character string, or based on a learning model that has learned about combinations of item names and item values. An eighth invention is a similar document search device according to any one of the first to seventh inventions, comprising: a stored document item set extraction means for extracting the plurality of item sets from the stored documents; and an item set registration means for registering the plurality of item sets extracted by the stored document item set extraction means in the item set storage unit in association with the stored documents. A ninth invention is a similar document search device according to any one of the first to eighth inventions, wherein the item set includes a combination of multiple item values ​​for one item name. A tenth aspect of the present invention is a program for causing a computer to function as any one of the similar document search devices according to the first to ninth aspects of the present invention. [Effects of the Invention]

[0007] According to the present invention, it is possible to provide a similar document search device and program that enable a document search that is in line with the context. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a diagram showing the overall configuration of a similar document search system according to an embodiment of the present invention and a functional block diagram of a similar document search server. [Figure 2] 10 is a diagram illustrating an example of an item set storage unit of the similar document search server according to the embodiment. FIG. [Figure 3] 10 is a flowchart showing a pre-processing of the similar document search server according to the embodiment; [Figure 4] 10 is a flowchart showing a similar document search process of the similar document search server according to the embodiment; [Figure 5] 10A and 10B are diagrams for explaining an item set extraction process in the similar document search server according to the embodiment; [Figure 6] 10 is a flowchart showing a similarity process of the similar document search server according to the embodiment; [Figure 7] 10 is a diagram for explaining a similarity process in the similar document search server according to the embodiment; FIG. [Figure 8] 10 is a diagram for explaining a similarity process in the similar document search server according to the embodiment; FIG. [Figure 9] 10 is a flowchart showing a similarity process of the similar document search server according to a modified example of the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, this is merely an example, and the technical scope of the present invention is not limited to this example. (Embodiment) <Overall Configuration of Similar Document Search System 100> FIG. 1 is a diagram showing the overall configuration of a similar document search system 100 according to this embodiment and a functional block diagram of a similar document search server 1. As shown in FIG. FIG. 2 is a diagram showing an example of the item set storage unit 23 of the similar document search server 1 according to this embodiment.

[0010] 1, the similar document search system 100 includes a similar document search server 1 (similar document search device), a document server 4, a user terminal 5, and an OCR device 6. The similar document search server 1, the document server 4, the user terminal 5, and the OCR device 6 are connected via a communication network N. In the similar document search system 100, the similar document search server 1 accepts document image data (target document image) input from the user terminal 5, converts the document image data into text, and extracts item sets consisting of item names and item values ​​from the text. Then, in the similar document search system 100, the similar document search server 1 calculates document similarity indicating the degree of match between the item sets of the extracted document image data and the item sets of saved document data (saved documents) saved in the document server 4, and identifies saved document data similar to the document image data based on the document similarity.

[0011] <Similar Document Search Server 1> The similar document search server 1 is a server that performs processing to identify stored document data similar to document image data and output the identified stored data. The similar document search server 1 includes a control unit 10, a storage unit 20, and a communication interface unit 29. The control unit 10 is a CPU (Central Processing Unit) that controls the entire similar document search server 1. The control unit 10 appropriately reads and executes the OS (Operating System) and various application programs stored in the storage unit 20, thereby cooperating with the above-mentioned hardware and executing various functions.

[0012] The control unit 10 includes a document image acquisition unit 11, a character string acquisition unit 12, a relationship estimation unit 13, an item set extraction unit 14, a similarity calculation unit 15, a similar document output unit 16, a saved document item set extraction unit 17, and an item set registration unit 18. The document image acquisition unit 11 functions as a document receiving means. The document image acquisition unit 11 receives document image data from the user terminal 5. The document image data may be, for example, an image of a paper document converted by a scanner (not shown) or the like, and may be image data in PDF format, JPEG format, IMG format, or the like. More specifically, the document image data may be an image of a document such as an estimate, invoice, or health certificate, and may include not only text but also a detailed area in a table format.

[0013] The character string acquisition unit 12 functions as a character string acquisition unit. The character string acquisition unit 12 detects characters from the document image data received by the document image acquisition unit 11, and acquires the detected character blocks as a character string. More specifically, the character string acquisition unit 12 transmits document image data to the OCR device 6, for example, and detects characters from the text data acquired from the OCR device 6. The character string acquisition unit 12 then groups the detected characters into a single character string based on the character spacing between them. The character spacing between characters that make up the character string may vary depending on, for example, the character size. For example, when the character size is large, the character string acquisition unit 12 groups adjacent characters into a character string even if the character spacing is wider than when the character size is small. Furthermore, the character string acquiring unit 12 acquires the text as a character string, and also acquires the coordinates (position information) of the character string.

[0014] The relationship estimation unit 13 functions as estimation means. The relationship estimation unit 13 estimates the attribute information and correspondence of character strings from the positions and contents of the character strings detected by the character string acquisition unit 12. The relationship estimation unit 13 can estimate the attribute information and correspondence of character strings, for example, based on the presence or absence of keywords included in the character strings and / or rules regarding the positions of the character strings. Furthermore, the relationship estimation unit 13 can estimate the attribute information and correspondence of character strings, for example, based on a learning model stored in a learning model storage unit 22 described later. The item set extraction unit 14 functions as an item set extraction means. Based on the estimation results by the relationship estimation unit 13, the item set extraction unit 14 extracts a plurality of item sets, which are combinations of a plurality of item names and item values ​​corresponding to each of the plurality of item names, from the document image data.

[0015] The similarity calculation unit 15 functions as a similarity calculation means. The similarity calculation unit 15 calculates document similarity using a plurality of item sets for the document image data extracted by the item set extraction unit 14 and a plurality of item sets for the document data stored in the document server 4 and stored in the item set storage unit 23. More specifically, the similarity calculation unit 15 calculates the document similarity based on the degree of agreement between the item names of the multiple item sets for the document image data and the item names of the multiple item sets for the saved document data stored in the item set storage unit 23. Here, there is an empirical rule that documents of the same type as the document image have similar item names. Therefore, when searching for documents of the same type as the document image, the document similarity can be calculated using the degree of agreement between the item names.

[0016] The similarity calculation unit 15 also calculates document similarity based on the degree of match between a plurality of item sets for the document image data and a plurality of item sets for the saved document data stored in the item set storage unit 23. When searching for documents of the same type as the document image that also have similar content, the document similarity can be calculated using the degree of match between the combinations of item sets in addition to the degree of match between the item names. Note that various known techniques can be used to calculate document similarity. For example, one method for calculating document similarity based on item names is to vectorize documents using item names and then use cosine similarity.

[0017] The similar document output unit 16 functions as a similar document output means. The similar document output unit 16 extracts from the document server 4 stored document data corresponding to a plurality of item sets having the highest document similarity calculated by the similarity calculation unit 15, and outputs the data to the user terminal 5. The similar document output unit 16 may also extract from the document server 4 stored document data corresponding to a plurality of item sets, for which the document similarity calculated by the similarity calculation unit 15 is equal to or greater than a preset threshold, and output the data to the user terminal 5.

[0018] The saved document item set extraction unit 17 functions as a saved document item set extraction means. The saved document item set extraction unit 17 extracts a plurality of item sets from saved document data stored in the document server 4. Here, when the saved document data is image data, the saved document item set extraction unit 17 can extract a plurality of item sets by subjecting the saved document data to processing by the character string acquisition unit 12 and processing by the relationship estimation unit 13 as described above. The item set registration unit 18 functions as an item set registration means. The item set registration unit 18 registers a plurality of item sets for the archived document data extracted by the archived document item set extraction unit 17 in the item set storage unit 23 in association with the archived document data. The details of each of these functions will be described later.

[0019] The storage unit 20 is a storage device such as a hard disk or semiconductor memory device for storing programs, data, etc. required for the operation of the similar document search server 1. The storage unit 20 includes a program storage unit 21, a learning model storage unit 22, and an item set storage unit 23. The program storage unit 21 is a storage area for storing various programs executed by the similar document search server 1. The program storage unit 21 stores a similar document search program 21a (program). The similar document search program 21a is, for example, a program for executing various functions of the control unit 10. The programs for executing the various functions of the control unit 10 are not limited to those realized by a single similar document search program 21a as described above. For example, each function of the control unit 10 or each of multiple functions may be provided as a separate program.

[0020] The learning model storage unit 22 is a storage area for storing learning models. A model that has been learned about combinations of item names and item values ​​is stored in the learning model storage unit 22. The learning model is, for example, a model that has been trained using a large amount of document data using deep learning technology.

[0021] The item set storage unit 23 is a storage area for storing item sets, which are combinations of a plurality of item names and their item values ​​contained in the saved document data. FIG. 2 shows an example of the item set storage unit 23. The item set storage unit 23 stores an item set, which is a combination of an item name and an item value, for each document name indicating saved document data. The document name is identification information for identifying the stored document data, and can specify the stored document. The document name may be, for example, the data name of the stored document data. The item name and item value are the item name included in the saved document data and the item value corresponding to the item name. The item set storage unit 23 stores, as the document name or in addition to the document name, for example, the storage location (directory name) of the document server 4 as the storage destination of the stored document data.

[0022] The communication interface unit 29 in FIG. 1 is an interface for communicating with the document server 4, the user terminal 5, the OCR device 6, and the like via the communication network N. Here, a computer refers to an information processing device equipped with a control unit, a storage device, etc., and the similar document search server 1 is an information processing device equipped with a control unit 10, a storage unit 20, etc., and is included in the concept of a computer. Also, there is no limit to the number of hardware components that make up the similar document search server 1. It may be composed of one or more as needed. Additionally, the hardware of the similar document search server 1 may include various servers such as a web server, a DB server, an application server, etc. as needed, and it may be composed of one server, or each may be composed of separate servers. Also, the similar document search server 1 may be, for example, a cloud.

[0023] <Document server 4> The document server 4 shown in FIG. 1 is a data server that stores the saved document data used in the similar document search. The saved document data is, for example, images of documents such as estimates, invoices, medical examination reports, etc., and may include tabular detail areas in addition to text. In FIG. 1, only one document server 4 is shown, but there may be multiple document servers 4 on the communication network N. Also, the document server 4 may be, for example, a cloud. Although not shown, the document server 4 includes a control unit, a storage unit, a communication interface unit, etc.

[0024] <User terminal 5> The user terminal 5 is, for example, a terminal used by a user who wants to search for saved document data similar to the document image data. The user terminal 5 can be composed of, for example, a personal computer (PC), a tablet terminal, etc. Although not shown, the user terminal 5 includes a control unit, a storage unit, a display unit, an input unit, a communication interface unit, etc.

[0025] <OCR device 6> The OCR device 6 is a device that performs processing related to OCR for recognizing characters from document image data. The OCR device 6 receives the document image data from the similar document search server 1, performs character recognition processing on the document image to obtain text data, and transmits the text data to the similar document search server 1. The OCR device 6 is composed of, for example, a server or the like. The OCR device 6 may also be composed of a PC or the like. Although not shown, the OCR device 6 includes a control unit, a storage unit, a communication interface unit, etc.

[0026] The communication network N is a network between the similar document search server 1, the document server 4, the user terminal 5, and the OCR device 6, and is, for example, a communication network such as the Internet. The communication network N may include a LAN (Local Area Network) or the like. Furthermore, the communication network N may be wired or wireless.

[0027] <Processing Description> Next, the processing of the similar document search server 1 will be described. First, as a pre-processing, a process of registering item sets of a plurality of stored document data registered in the document server 4 in the item set storage unit 23 will be described. This process must be performed at least when constructing the similar document search system 100. After a saved document is added to the document server 4, this process can be performed at an appropriate timing. FIG. 3 is a flowchart showing the pre-processing of the similar document search server 1 according to this embodiment.

[0028] 3 (hereinafter, "step S" will be simply referred to as "S") 11, the control unit 10 of the similar document search server 1 acquires unprocessed stored document data from the document server 4. Here, "unprocessed" means that no item set has been registered in the item set storage unit 23, and the control unit 10 can confirm whether the stored document data is unprocessed by referring to the item set storage unit 23. In S12, the character string obtaining unit 12 detects characters from the obtained saved document and obtains a character string from the detected characters. The character string obtaining unit 12 can perform this process using the OCR device 6. In S13, the relationship estimation unit 13 estimates attribute information and correspondence relationships of the character strings from the positions and contents of the character strings acquired in the process of S12. The relationship estimation unit 13 may make the estimation based on rules regarding the presence or absence of keywords included in the character strings and / or the positions of the character strings, or may make the estimation using a learning model stored in the learning model storage unit 22. The relationship estimation unit 13 may also use both estimation methods.

[0029] In S14, the item set extraction unit 14 extracts item sets based on the estimation result obtained in the process of S13. In S15, the item set registration unit 18 registers the item sets extracted in the process of S14 in the item set storage unit 23 in association with the document names of the saved documents.

[0030] In S16, the control unit 10 determines whether or not there is still unprocessed stored document data that has not been processed in the document server 4. If there is still unprocessed stored document data (S16: YES), the control unit 10 moves the process to S11, extracts item sets for the unprocessed stored document data, and repeats the process of registering them in the item set storage unit 23. On the other hand, if there is no unprocessed stored document data (S16: NO), the control unit 10 ends this process. When the pre-processing is completed, the item set storage unit 23 stores a plurality of item sets included in the stored document for each of the plurality of stored document data stored in the document server 4.

[0031] Next, a process of searching for a saved document similar to document image data by the user terminal 5 will be described. FIG. 4 is a flowchart showing the similar document search process of the similar document search server 1 according to this embodiment. FIG. 5 is a diagram for explaining the process of extracting an item set in the similar document search server 1 according to this embodiment. FIG. 6 is a flowchart showing the similarity process of the similar document search server 1 according to this embodiment. 7 and 8 are diagrams for explaining the similarity process in the similar document search server 1 according to this embodiment.

[0032] First, the user terminal 5 connects to the similar document search server 1 and specifies and sends document image data, which is image data of the document to be searched for, such as an estimate or invoice, on a search screen not shown.In S21 of Figure 4, the document image acquisition unit 11 of the similar document search server 1 accepts the document image data. In S22, the character string obtaining unit 12 detects characters in the obtained document image and obtains a character string from the detected characters. The character string obtaining unit 12 can detect characters in the document image and obtain a character string from the detected characters, for example, by transmitting document image data to the OCR device 6 and obtaining text data from the OCR device 6. The processes of S23 and S24 are the same as the processes of S13 and S14 in FIG.

[0033] 5(A) shows an example of an estimated document image 30 showing the estimation result obtained by estimating attribute information and correspondence relationships of character strings in a document image. This processing is performed as an internal processing of the similar document search server 1, and therefore the estimated document image 30 shown in FIG. 5(A) is not output to the user terminal 5. The relationship estimation unit 13 estimates the character strings into item names, item values, headers, and others. Therefore, the estimated document image 30 distinguishes each estimation result with different shading. For example, character string 31 is inferred as an item name, character string 32 is inferred as an item value, character string 33 is inferred as a header, and character string 34 is inferred as other. FIG. 5B shows an extraction result 40 in which item names and item values ​​corresponding to the item names are extracted as item sets from the estimated document image 30 in FIG. 5A. The extraction result 40 is obtained by scanning the estimated document image 30 horizontally from the top left and extracting item names and item values ​​as item pairs.

[0034] In S25 of FIG. 4, the similarity calculation unit 15 performs similarity processing. Here, the similarity processing will be described with reference to FIG. FIG. 6 explains a similar document search based on the degree of coincidence of item names. In S31 of Figure 6, the similarity calculation unit 15 calculates the degree of match between the item names of the item set obtained from the document image data and the item names of the item sets of each saved document data stored in the item set memory unit 23 as document similarity. Here, a method for calculating the degree of match will be described with a specific example. In FIG. 7, document image data of an estimate is shown as document A, and saved document data is shown as documents B and C. 7(A) shows a table 71 in which the item names of documents A, B, and C are extracted. The item names of document A are extracted by the process of S24 in FIG. 4. The item names of documents B and C are obtained from the item set storage unit 23.

[0035] The similarity calculation unit 15 uses the item names extracted from each document to vectorize each document. Figure 7(B) shows Table 72, in which each document is vectorized using the number of occurrences of the item names in Table 71. In this example, the vectorization is performed using the Bag of Words format. The Bag of Words format is a method of vectorizing a document using the number of occurrences of words that appear in the document, and in this case, the vectorization is performed using the number of occurrences of the item names. In table 71 shown in FIG. 7(A), item names that appear in document A and in other documents are shaded.

[0036] Next, the similarity calculation unit 15 calculates the document similarity using the cosine similarity of the vectorized result. There are various methods for calculating the similarity from the vectorized results, but here, as an example, a method using cosine similarity will be described. The cosine similarity between document x and document y can be calculated using the following formula:

[0037]

number

[0038]

number

[0039] In the above, we used cosine similarity for the number of occurrences of the extracted item names, but we can also use cosine similarity for TF-IDF (Term Frequency - Inverse Document Frequency). TF-IDF is a vectorization technique that gives a higher weight to words that do not appear in other documents. A specific example of similarity calculation using TF-IDF will be described with reference to FIG.

[0040] Figure 8(A) is a table 73 showing TF values. Since the TF value indicates the frequency of occurrence of a certain word in a document, Table 73 is the same as Table 72 shown in Figure 7(B). Next, the control unit 10 uses the TF values ​​shown in Table 73 to calculate the DF value and the IDF value. The DF value indicates how many documents the word appears in and is determined based on Table 73, which shows the TF value. The IDF value can be calculated using the following formula:

[0041]

number

[0042]

number

[0043] Next, in S32 of FIG. 6, the control unit 10 determines whether the calculated document similarity is equal to or greater than a threshold value. The threshold value can be set as appropriate. If the document similarity is equal to or greater than the threshold value (S32: YES), the control unit 10 proceeds to S33. On the other hand, if the document similarity is less than the threshold value (S32: NO), the control unit 10 proceeds to S26 of FIG. 4. If the document similarity is less than the threshold value, the control unit 10 may transmit, for example, a message stating, "No similar documents were found." If the document similarity is less than the threshold value, the similar document output unit 16 may extract, from the document server 4, stored document data having an item name with the highest threshold value among those less than the threshold value as the stored document most similar to the document image, and transmit the extracted data to the user terminal 5.

[0044] In S33, the control unit 10 acquires from the document server 4 stored document data having an item name whose document similarity is equal to or greater than a threshold value. In S34, the similar document output unit 16 transmits the stored document data acquired from the document server 4 to the user terminal 5. Thereafter, the control unit 10 moves the process to S26 in FIG. By this process, a saved image of the same type that is similar to the document image can be output to the user terminal 5.

[0045] 4, the control unit 10 determines whether or not to register the document image data as a storage document in the document server 4. The control unit 10 may determine whether or not to register the document image data as a storage document in the document server 4, for example, in response to an instruction from the user terminal 5. If the document image data is to be registered as a storage document in the document server 4 (S26: YES), the control unit 10 proceeds to S27. On the other hand, if the document image data is not to be registered as a storage document in the document server 4 (S26: NO), the control unit 10 terminates this process. In S27, the control unit 10 registers the document image data in the document server 4, and the item set registration unit 18 registers the item set extracted in the processing of S24 in the item set storage unit 23. Thereafter, the control unit 10 ends this processing.

[0046] (Variation) The above describes the search for similar documents using document similarity based on the degree of matching of item names. In the search for similar documents using document similarity based on the degree of matching of item names, it is possible to search for stored documents of the same type as a document image. Here, if you want to search for documents of the same type that also have similar content, you can search for stored documents that are similar to the document image by performing the process shown in Figure 9, which will be explained next, instead of the process shown in Figure 6.

[0047] FIG. 9 is a flowchart showing the similarity process of the similar document search server 1 according to a modified example of this embodiment. The processes of S231 and S232 in FIG. 9 are the same as the processes of S31 and S32 in FIG. In S233, the similarity calculation unit 15 calculates, as document similarity, the degree of match between the combination of item sets acquired from the document image data and the combination of item sets stored in the item set storage unit 23 for each saved document data item whose item name is equal to or greater than the threshold value as a result of the process in S232. This calculation process can also be performed by the above-mentioned process of vectorizing the document using the item values ​​and then using cosine similarity. The processes from S234 to S236 are the same as the processes from S32 to S34 in Fig. 6. However, the process of S235 differs from the process of S34 in Fig. 6 in that saved document data having an "item set" equal to or greater than a threshold is acquired from the document server 4.

[0048] In this way, the processing by the similar document search server 1 uses matching combinations of item sets to search for similar documents, and in addition to the same type of stored documents, it can search for stored documents with similar content that contain information on proper nouns such as company names, personal names, and destinations contained in the document images.

[0049] As described above, according to this embodiment, the similar document search server 1 has the following effects. (1) An item set storage unit 23 is provided that stores a plurality of item sets, which are combinations of a plurality of item names and item values ​​corresponding to each of the plurality of item names contained in a stored document, in association with the stored document. A plurality of item sets are extracted from a document image, and document similarity is calculated using the extracted plurality of item sets and a plurality of item sets corresponding to each of the plurality of stored documents stored in the item set storage unit 23. Since similar documents are similar mainly in the items that appear in them, calculating document similarity using the item sets contained in the documents enables document retrieval that is more context-sensitive.Furthermore, by narrowing down the items used in the search to the item sets, searches can be performed with high accuracy and efficiency at low computational cost.

[0050] (2) A document similarity is calculated that indicates the degree of match between the item names of the extracted multiple item sets and the item names of the multiple item sets corresponding to each of the multiple saved documents stored in the item set memory unit 23. Therefore, by searching for similar documents based on the degree of coincidence of the item names that appear, it is possible to search for stored documents of the same type as the document image.

[0051] (3) A document similarity indicating the degree of match between the extracted multiple item sets and multiple item sets corresponding to each of the multiple saved documents stored in the item set storage unit 23 is calculated. Therefore, by searching for similar documents based on the degree of match between the combinations of the item names and item values ​​that appear, it is possible to search for stored documents whose contents are similar to the document image.

[0052] (4) The stored documents corresponding to the plurality of item pairs with the highest calculated document similarity are output. Therefore, it is possible to output the stored document that is most similar to the document image.

[0053] (5) The system outputs stored documents corresponding to multiple item pairs whose calculated document similarity is equal to or greater than a threshold value. Therefore, it is possible to output one or more stored documents that are considered to be similar to the document image to a certain degree or more.

[0054] (6) A document image is input, a character string is acquired from the received document image, attribute information and correspondence of the character string are estimated from the position and content of the acquired character string, and multiple item sets are extracted from the document image based on the estimation results. Therefore, it is possible to extract a plurality of item sets from a document image by estimating the attribute information and correspondence of the character strings.

[0055] (7) The attribute information and correspondence of a string are estimated based on rules regarding the presence or absence of keywords contained in the string and / or the position of the string, or based on a learning model that has learned about combinations of item names and item values. Therefore, the item name and item value can be appropriately estimated from the character string.

[0056] (8) A plurality of item sets are extracted from the archived document, and the extracted plurality of item sets are registered in the item set storage unit 23 in association with the archived document. Therefore, the item sets can be used to search for similar documents. Also, by acquiring the item sets of the stored documents registered in the item set storage unit 23 using the same method as acquiring the item sets used when searching for document images, the item set extraction process can be standardized, thereby improving search accuracy.

[0057] Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments. Furthermore, the effects described in the embodiments are merely a list of the most preferable effects resulting from the present invention, and the effects of the present invention are not limited to those described in the embodiments. Note that the above-described embodiments and the modified embodiments described below can be used in appropriate combinations, but detailed description thereof will be omitted.

[0058] (Variations) (1) In this embodiment, document images of documents such as estimates, invoices, and medical certificates have been described as examples, but the present invention is not limited to these. It can be used for various types of documents. It can also be used for standard documents and non-standard documents, and the same processing can be performed for any type of document.

[0059] (2) In the present embodiment, a similar document search is performed by item name, followed by a similar document search by a combination of item sets. However, this is not limiting. For example, a similar document search may be performed by a combination of item sets without performing a similar document search by item name. This makes it possible to search for documents similar to a document image from various types of stored documents with similar content.

[0060] (3) In the present embodiment, the document similarity is calculated by vectorizing documents using item names and item sets and using cosine similarity, but this is not limiting. Other methods may be used to calculate the degree of agreement between item names and item sets.

[0061] (4) In the present embodiment, an item set is described as a combination of one item name and one item value, but this is not limiting. Of course, an item set may also include a one-to-many combination of an item name and an item value, where one item name corresponds to multiple item values.

[0062] (5) In the present embodiment, the similar document search server transmits a document image to an OCR device and requests text conversion, but this is not limiting. Preprocessing may be added before transmitting the document image to the OCR device. Examples of preprocessing include contrast adjustment and sharpening. This may further improve the accuracy of processing by the OCR device.

[0063] (6) Although not specifically described in this embodiment, in the rule-based character string estimation, normalization may be performed using, for example, regular expressions. In this way, in the similarity calculation, item names that have the same meaning but different spellings can be treated as the same, thereby improving search accuracy.

[0064] (7) In the present embodiment, an example has been described in which a similar document search server, a document server, a user terminal, and an OCR device are included, but this is not limiting. For example, the similar document search server and the OCR device may be implemented by the same server. Also, for example, the similar document search server and the document server may be implemented by the same server. Furthermore, for example, a standalone configuration may be provided in which only a client terminal has the functions of the similar document search server, the document server, and the OCR function.

[0065] (8) In this embodiment, the search is performed on documents stored in a document server, but this is not limiting. For example, other documents on the communication network N may be used as the search target. In this case, the search can be performed by pre-processing the target documents to obtain a set of items, and then performing a similar document search process. [Explanation of symbols]

[0066] 1. Similar document search server 4 Document Server 5. User terminal 6 OCR device 10 Control Unit 11 Document image acquisition unit 12 String acquisition part 13 Relationship Estimation Section 14 Item set extraction part 15 Similarity calculation part 16 Similar Document Output Unit 17 Archived document item set extraction part 18 Item Group Registration Department 20 Memory section 21a Similar document search program 22 Learning model memory unit 23 Item set storage 100 Similar Document Search System

Claims

1. an item set storage unit that stores a plurality of item sets, each of which is a combination of a plurality of item names and an item value corresponding to each of the plurality of item names, in association with the stored document; an item set extraction means for extracting the plurality of item sets from a target document image; a similarity calculation means for calculating document similarity using the plurality of item sets extracted by the item set extraction means and the plurality of item sets corresponding to each of the plurality of saved documents stored in the item set storage unit; A similar document search device comprising:

2. 2. The similar document search device according to claim 1, The similarity calculation means calculates the document similarity indicating the degree of match between the item names of the plurality of item sets extracted by the item set extraction means and the item names of the plurality of item sets corresponding to each of the plurality of saved documents stored in the item set storage unit.

3. 3. The similar document search device according to claim 1, The similarity calculation means calculates the document similarity indicating the degree of match between the plurality of item sets extracted by the item set extraction means and the plurality of item sets corresponding to each of the plurality of saved documents stored in the item set storage unit.

4. 2. The similar document search device according to claim 1, a similar document output unit that outputs the stored document corresponding to the plurality of item sets having the highest document similarity calculated by the similarity calculation unit;

5. 5. The similar document search device according to claim 4, The similar document output means outputs the stored documents corresponding to the plurality of item sets whose document similarities calculated by the similarity calculation means are equal to or greater than a threshold value.

6. 2. The similar document search device according to claim 1, a document receiving means for receiving an input of the target document image; a character string acquisition means for acquiring a character string from the target document image accepted by the document acceptance means; an estimation means for estimating attribute information and correspondence of the character string from the position and content of the character string acquired by the character string acquisition means; Equipped with The item set extraction means extracts the plurality of item sets from the target document image based on the estimation result by the estimation means.

7. 7. The similar document search device according to claim 6, The estimation means estimates the attribute information and correspondence of the character string based on rules regarding the presence or absence of keywords included in the character string and / or the position of the character string, or based on a learning model that has learned about combinations of item names and item values.

8. 2. The similar document search device according to claim 1, a stored document item set extraction means for extracting the plurality of item sets from the stored document; an item set registering means for registering the plurality of item sets extracted by the archived document item set extracting means in the item set storage unit in association with the archived document; A similar document search device comprising:

9. 2. The similar document search device according to claim 1, The item set includes a combination of a plurality of the item values ​​for one of the item names.

10. a computer including an item set storage unit that stores a plurality of item sets, each of which is a combination of a plurality of item names and an item value corresponding to each of the plurality of item names, in association with the stored document; an item set extraction means for extracting the plurality of item sets from a target document image; a similarity calculation means for calculating document similarity using the plurality of item sets extracted by the item set extraction means and the plurality of item sets corresponding to each of the plurality of saved documents stored in the item set storage unit; A program to make it function as such.

Images