Document filtering device

Abstract

A document filtering device (1) includes a conditional probability calculation unit (14) for calculating a first conditional probability and a second conditional probability for each of words included in a well-known document that is input, a prior probability calculation unit (15) for calculating a prior probability, a probability calculation unit (16) for calculating a probability that the input well-known document is an updated well-known document, using Bayes' theorem based on the first conditional probability, the second conditional probability, and the prior probability, a filtering unit (17) that filters the input well-known document on the basis of the calculated probability, and a classification unit (13) that classifies well-known documents stored in a storage unit into updated well-known documents and well-known documents that are not updated well-known documents, on the basis of update history of software, wherein the prior probability calculation unit (15) calculates the prior probability on the basis of the ratio between the updated well-known documents and the well-known documents that are not updated well-known documents, that are classified by the classification unit (13).

Description

Document Filtering Device

[0001] The embodiment relates to a document filtering device.

[0002] A Bayesian filter, which utilizes Bayes' theory (Bayes' theorem), is known as a method for filtering documents of a specific category from a large number of documents. In a Bayesian filter, for each word contained in an input document, 1) the conditional probability that this document contains the corresponding word in the input document when the document is a specific document, and 2) the conditional probability that this document contains the corresponding word in the input document when the document is not a specific document, are calculated. Then, using Bayes' theorem based on these conditional probabilities, the probability that the input document is a specific document is calculated. If this probability is high, the input document is classified as a specific document. This classification result is used to determine whether a newly input document is a specific document. In this way, a Bayesian filter uses previously accumulated information to determine new phenomena, and repeated determinations are expected to improve the accuracy of the determination.

[0003] Such a Bayesian filter is used, for example, to filter out spam from received emails. A Bayesian filter can also be used to filter out public documents that contain specific information that should be made known to users from among a large number of public documents that contain information that should be made known to users.

[0004] Naruhodo Statistical Academy > Reference > TOP > 15 Statistical Episode Collection > Identifying Spam Email [Online], [Retrieved January 18, 2023], Internet <URL https: / / www.stat.go.jp / naruhodo / 15_episode / toukeigaku / meiwaku.html>

[0005] In order to actually use Bayes' theorem to calculate the probability that a document containing a given word is a specific document, pre-training is required. For pre-training, documents stored in the past must be correctly classified into specific documents and other documents. The task of classifying documents for such pre-training imposes a heavy workload on the user.

[0006] The embodiment provides a document filtering device with a reduced workload for pre-training.

[0007] One embodiment of the document filtering device filters update notification documents that notify users to update their software. The document filtering device includes a conditional probability calculation unit, a prior probability calculation unit, a probability calculation unit, a filtering unit, and a classification unit. The conditional probability calculation unit calculates, for each word contained in an input public document, a first conditional probability, which is the probability that the word is contained when the public document is an update public document, and a second conditional probability, which is the probability that the word is contained when the public document is not an update public document. The prior probability calculation unit calculates a prior probability, which is the ratio of update public documents to non-update public documents among the public documents stored in the first storage unit. The probability calculation unit calculates the probability that the input public document is an update public document using Bayes' theorem based on the first conditional probability, the second conditional probability, and the prior probability. The filtering unit filters the input public document based on the probability that the input public document is an update public document. The classification unit classifies the public documents stored in the first storage unit into update public documents and public documents that are not update public documents based on the update history of the software stored in the second storage unit. The prior probability calculation unit calculates a prior probability based on the ratio of the update public documents to the public documents that are not update public documents classified by the classification unit.

[0008] According to the embodiment, a document filtering device is provided in which the workload for pre-learning is reduced.

[0009] Fig. 1 is a block diagram showing the configuration of a document filtering device according to an embodiment. Fig. 2 is a diagram showing an example of a configuration management history. Fig. 3 is a flowchart showing the pre-learning operation of the document filtering device 1 according to an embodiment. Fig. 4 is a diagram showing a specific example for identifying an update notification document. Fig. 5 is a flowchart showing the document filtering operation of the document filtering device 1 according to an embodiment. Fig. 6 is a block diagram showing an example of the hardware configuration of the document filtering device.

[0010] Hereinafter, embodiments will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a document filtering device according to an embodiment. The document filtering device according to the embodiment filters, for example, a specific publicly known document from a large number of publicly known documents. The specific publicly known document in the embodiment is, for example, an update notification document that notifies a user to perform a software update. The document filtering device according to the embodiment can be applied to, for example, a computer device that manages the configuration of an IT system. The IT system can include, but is not limited to, computer systems such as a personnel management system and a time attendance management system, as well as network management systems.

[0011] The document filtering device 1 includes a memory unit 11, a memory unit 12, a classification unit 13, a conditional probability calculation unit 14, a priori probability calculation unit 15, a probability calculation unit 16, a filtering unit 17, and a notification unit 18.

[0012] The storage unit 11 stores data of publicly known documents D input to the document filtering device 1. Publicly known documents D may include updated publicly known documents. The storage unit 11 may store data of updated publicly known documents in a folder or the like separate from data of other publicly known documents. Here, the publicly known documents D may be data in any format that includes publicly known text, such as data in the body of an e-mail or document data attached as an attachment.

[0013] The storage unit 12 stores a configuration management history 12a of the IT system. The configuration management history 12a is a history of changes in the status of management items related to elements such as hardware, software, and networks that make up the IT system. In the embodiment, the configuration management history 12a includes at least a history of software updates. The configuration management history 12a may also include other information such as a history of errors in the IT system. Furthermore, the storage units 11 and 12 may be configured as a single storage unit.

[0014] FIG. 2 is a diagram showing an example of a configuration management history 12a. The configuration management history 12a includes, for example, items such as date and time and content. The date and time is the date and time when a change in the status of a management item occurred. The content is the content of the change in status. For example, the content of a software update includes update information such as information such as the name and control number for identifying the updated software, and information on the software version and revision for identifying the applied update. Furthermore, the content of a system error includes information such as the name and control number for identifying the element in which the error occurred, and error information such as the content of the error and information on the scope of the error. Here, in the configuration management history 12a, the update history and the system error history may be managed separately.

[0015] The classification unit 13 classifies the notification documents D stored in the storage unit 11 into update notification documents and other notification documents based on the configuration management history 12a stored in the storage unit 12. The classification unit 13 then stores information indicating whether each notification document D stored in the storage unit 11 is an update notification document or other notification document as classification result information for each notification document D. If the notification document is an update notification document, the classification result information may further include information indicating which version or revision of the software the corresponding update notification document is an update for. Details of the classification by the classification unit 13 will be described in detail later.

[0016] The conditional probability calculation unit 14 calculates conditional probabilities for calculating the probability that the input publicly known document D is an updated publicly known document using Bayes' theorem. The conditional probabilities include, for each word contained in the input publicly known document D, the probability that the publicly known document contains the corresponding word when the publicly known document is an updated publicly known document, and the probability that the publicly known document contains the corresponding word when the publicly known document is not an updated publicly known document. The conditional probability calculation unit 14 performs morphological analysis on the input publicly known document D to divide sentences contained in the input publicly known document D into words. The conditional probability calculation unit 14 then calculates conditional probabilities for each divided word as a corresponding word. The conditional probability calculation unit 14 calculates the probability that the publicly known document contains a corresponding word when the publicly known document is an updated publicly known document, for example, as the rate at which the corresponding word appears among all publicly known documents D stored in the memory unit 11 that have been classified as updated publicly known documents by the classification unit 13. On the other hand, the conditional probability calculation unit 14 calculates the probability that a publicly known document contains the relevant word when the publicly known document is not an updated publicly known document, for example, as the rate at which the relevant word appears among all publicly known documents D stored in the memory unit 11 that have been classified by the classification unit 13 as not being updated publicly known documents.

[0017] The prior probability calculation unit 15 calculates a prior probability for calculating the probability that the input publicly known document D is an updated publicly known document using Bayes' theorem. The prior probability is, for example, the ratio between the probability that the input publicly known document is an updated publicly known document and the probability that the input publicly known document is a publicly known document that is not an updated publicly known document. The prior probability calculation unit 15 calculates the prior probability as, for example, the ratio between the updated publicly known document stored in the storage unit 11 and the publicly known document that is not an updated publicly known document.

[0018] The probability calculation unit 16 calculates the probability that the input publicly known document D is an updated publicly known document using Bayes' theorem based on the two types of conditional probabilities calculated by the conditional probability calculation unit 14 and the prior probability calculated by the prior probability calculation unit 15.

[0019] The filtering unit 17 filters the input publicly known document D into an updated publicly known document and other publicly known documents. For example, if the probability calculated by the probability calculation unit 16 is equal to or greater than a threshold, the filtering unit 17 classifies the input publicly known document D as an updated publicly known document, and if the probability is less than the threshold, the filtering unit 17 classifies the input publicly known document D as a publicly known document other than an updated publicly known document. Then, based on the results of filtering the input publicly known document, the filtering unit 17 stores the input publicly known document D in the storage unit 11. For example, when the input publicly known document D is an updated publicly known document, the filtering unit 17 stores the input publicly known document D in a folder for storing updated publicly known documents.

[0020] The notification unit 18 notifies the user of the document filtering device 1 that the input public notice document D is an updated public notice document. For example, if the public notice document D was received by email, the notification unit 18 displays, for example, on a display, a message indicating that an email containing an updated public notice document has been received.

[0021] Next, the operation of the document filtering device 1 will be described. Fig. 3 is a flowchart showing the pre-learning operation of the document filtering device 1 according to the embodiment. The operation in Fig. 3 is performed every certain period, such as one day, one week, one month, or one year. Note that the pre-learning operation in Fig. 3 does not need to be performed after a sufficient amount of pre-learning operation has been performed.

[0022] In step S1 , the classification unit 13 acquires the configuration management history 12 a from the storage unit 12 .

[0023] In step S2, the classification unit 13 extracts the software update history from the configuration management history 12a in order to classify the update notification document. Here, the extracted software update history only needs to be from the period during which the pre-learning operation was performed. For example, if the pre-learning operation is performed once a week, the classification unit 13 only needs to extract the update history for the most recent week.

[0024] In step S3, the classification unit 13 selects one piece of update information from the software update history, for example, in order of oldest to newest date and time.

[0025] In step S4, the classification unit 13 extracts notification documents D stored in the storage unit 11 within a predetermined period going back from the date and time of the selected update information. The predetermined period may be a fixed period such as one day or one week. Alternatively, the predetermined period may be set based on, for example, the average time it takes for a user to actually perform the update after receiving the update notification document. The user here may be an IT system administrator or a user of an individual terminal that constitutes the IT system.

[0026] In step S5, the classification unit 13 identifies the update notification document corresponding to the selected update information by comparing the update information selected from the software update history with the notification document D extracted from the storage unit 11. The process of step S5 will be described in detail below with reference to FIG.

[0027] As shown in Fig. 2, the configuration management history 12a includes update information such as the date and time when the software was updated, and the name and management number for identifying the software to be updated. For example, in Fig. 2, "software S" was updated at "date and time Ti." Fig. 4 shows an example of identifying an update notification document for "software S" that was updated at "date and time Ti."

[0028] Here, software updates are implemented by users who receive the update notification document. Therefore, the actual software update is always implemented after the update notification document is received. Furthermore, since the update notification document is a document that notifies users that they should update their software, it is highly likely that it will contain at least a description such as the name of the software to be updated.

[0029] Based on this concept, the classification unit 13, as shown in FIG. 4, identifies public notice documents Dn, Dn-1, ..., whose reception times are within a predetermined period p prior to date and time Ti, as update notice documents if the public notice documents contain words identifying the document as a public notice document for notifying updates to software S, such as "software S" or "update." The description identifying software S is extracted, for example, by performing morphological analysis on public notice document D. In the example of FIG. 4, public notice document Dn contains the words "software S" and "update," and therefore is identified as an update notice document. Note that if public notice document Dn is identified as an update notice document, the classification unit 13 may further extract other information, such as the version of the updated software, from the corresponding update information.

[0030] Here, it is possible that multiple update notification documents for the same software exist during the predetermined period p. In this case, the classification unit 13 may identify the update notification document corresponding to the date and time Ti by using information other than the software name, such as the version, that may be written in the update notification document.

[0031] Returning now to the description of FIG. 3 , in step S6, the classification unit 13 classifies the extracted notification documents for the predetermined period into update notification documents and other notification documents. For example, the classification unit 13 stores the update notification documents in a folder for storing update notification documents. The classification unit 13 may associate information such as the software version extracted from the update information with the update notification document and store it in the storage unit 11.

[0032] In step S7, the classification unit 13 determines whether classification is complete, i.e., whether classification of the update information and the corresponding update notification documents for all dates and times in the update history extracted in step S2 is complete. If it is determined in step S7 that classification is not complete, the process returns to step S3. In this case, the classification unit 13 selects the update information for the next date and time and performs processing to identify the update notification document corresponding to this update information. If it is determined in step S7 that classification is complete, the process of FIG. 3 ends.

[0033] 5 is a flowchart showing the document filtering operation of the document filtering device 1 according to the embodiment. The operation of FIG. 5 is performed every time a new publicly known document D is input to the document filtering device 1. The new publicly known document D is input by receiving an email, for example.

[0034] In step S11, the conditional probability calculation unit 14 calculates, for each word contained in the input publicly known document D, the probability that the publicly known document contains the corresponding word when the publicly known document is an updated publicly known document, and the probability that the publicly known document contains the corresponding word when the publicly known document is not an updated publicly known document. As described above, the conditional probability calculation unit 14 performs morphological analysis on the input publicly known document D to divide the sentences contained in the input publicly known document D into words. The conditional probability calculation unit 14 then calculates the conditional probability for each divided word as a corresponding word. Through the pre-learning operation described with reference to FIG. 3, the publicly known document D stored in the memory unit 11 is automatically classified into updated publicly known documents and other publicly known documents. Therefore, the conditional probability calculation unit 14 can accurately calculate the probability that the publicly known document contains the corresponding word when the publicly known document is an updated publicly known document and the probability that the publicly known document contains the corresponding word when the publicly known document is not an updated publicly known document.

[0035] In step S12, the prior probability calculation unit 15 calculates the prior probability. As described above, the prior probability is the ratio of the probability that the document is a publicly known document that is not an updated publicly known document. By the pre-learning operation described in FIG. 3, the publicly known document D stored in the memory unit 11 is automatically classified into updated publicly known documents and other publicly known documents. Therefore, the prior probability calculation unit 15 can calculate the prior probability with high accuracy. The prior probability may be calculated when it is determined in step S7 of FIG. 3 that classification is complete. In this case, the processing of step S12 may be omitted.

[0036] In step S13, the probability calculation unit 16 calculates the probability that the input publicly known document D is an updated publicly known document using Bayes' theorem based on the two types of conditional probabilities calculated by the conditional probability calculation unit 14 and the prior probability calculated by the prior probability calculation unit 15.

[0037] In step S14, the filtering unit 17 determines whether the probability calculated by the probability calculation unit 16 is equal to or greater than a threshold. If it is determined in step S14 that the probability is equal to or greater than the threshold, the process proceeds to step S15. If it is determined in step S14 that the probability is not equal to or greater than the threshold, the process proceeds to step S17.

[0038] In step S15, the filtering unit 17 classifies the input publicly known document D as an updated publicly known document. At this time, the filtering unit 17 stores the input publicly known document D in a folder for storing updated publicly known documents in the memory unit 11. The publicly known document D stored by the filtering unit 17 can be used in the next pre-learning operation and the next document filtering operation.

[0039] In step S16, the notification unit 18 notifies the user of the document filtering device 1 that the input notice document D is an update notice document. For example, the notification unit 18 displays a message on a display indicating that an email containing an update notice document has been received. Thereafter, the processing of FIG. 5 ends.

[0040] In step S17, the filtering unit 17 classifies the input public notice document D into public notice documents other than update public notice documents. At this time, the filtering unit 17 stores the input public notice document D in a folder different from the folder for storing update public notice documents in the storage unit 11. Then, the processing in FIG. 5 ends.

[0041] Fig. 6 is a block diagram showing an example of the hardware configuration of the document filtering device 1. As shown in Fig. 6, the document filtering device 1 includes, for example, a processor 101, a read only memory (ROM) 102, a random access memory (RAM) 103, a communication module 104, a display 105, and a storage 106. The document filtering device 1 may be, for example, a terminal device such as a smartphone or a personal computer (PC). Alternatively, the document filtering device 1 may be a device built on a cloud environment.

[0042] The processor 101 is a processing circuit that controls the overall operation of the document filtering device 1. The processor 101 operates as a classification unit 13, a conditional probability calculation unit 14, a priori probability calculation unit 15, a probability calculation unit 16, a filtering unit 17, and a notification unit 18 by executing a document filtering program stored in the storage 106, for example. The processor 101 is, for example, a CPU. The processor 101 may be an MPU, a GPU, an ASIC, an FPGA, or the like. The processor 101 may be a single CPU or the like, or multiple CPUs or the like.

[0043] The ROM 102 is a non-volatile semiconductor memory that stores a startup program and various setting values ​​for the document filtering device 1. The RAM 103 is a volatile memory that is used, for example, as a working memory when the processor 101 performs processing.

[0044] The communication module 104 is a circuit having an interface for communication between the document filtering device 1 and an external device. The communication module 104 may be a wireless communication module or a wired communication module.

[0045] The display 105 is, for example, a liquid crystal display (LCD) or an electroluminescence (EL) display, and displays a graphical user interface (GUI) corresponding to various applications such as an email application.

[0046] The storage 106 is a non-volatile storage device such as a hard disk or flash memory. The storage 106 stores various programs such as the operating system (OS) of the document filtering device 1 and a document filtering program. The storage 106 can also function as the storage unit 11 and the storage unit 12. In this case, the storage 106 stores the publicly known document D and the configuration management history 12a. The publicly known document D and the configuration management history 12a may be stored in a storage device such as a server external to the document filtering device 1.

[0047] As described above, according to the embodiment, the software update history stored as the configuration management history 12a is used to extract publicly known documents from the storage unit within a predetermined period prior to the software update date and time. Then, based on the extracted publicly known documents and the contents of the update, the publicly known documents stored in the storage unit are classified into update publicly known documents and other publicly known documents. Then, based on the classification results, the publicly known documents are filtered using Bayes' theorem.

[0048] In this manner, in the embodiment, classification of updated publicly known documents for pre-learning from other publicly known documents is automatically performed, i.e., in the embodiment, no user work is required in pre-learning for filtering publicly known documents using Bayes' theorem.

[0049] In the embodiment, the publicly known documents are classified based on the update information stored as the configuration management history 12 a. Therefore, information that is included in the update information but not included in the publicly known documents can be associated with the publicly known documents.

[0050] In the embodiment, the specific public documents to be filtered are update public documents. However, the specific public documents to be filtered are not limited to update public documents. In other words, as long as chronological information representing changes in the situation due to the content notified by the public documents is stored in the storage unit 12, the technology of the embodiment can be applied even when the specific public documents to be filtered are not update public documents. For example, the specific public documents to be filtered may be public documents notifying the user that a software update has been performed or that an error has occurred. In the case of public documents notifying the user of an update or an error, the user receives the respective public documents after the update or the error has actually been performed. Therefore, the classification unit 13 extracts public documents having a reception time within a predetermined period after the update and error history stored as the configuration management history 12a from the storage unit 11 and classifies the corresponding public documents.

[0051] Furthermore, in the embodiment, the document filtering device 1 is assumed to include a classification unit 13. Alternatively, the classification unit 13 may be provided in a pre-learning device of a document filtering device separate from the document filtering device 1. In this case, the classification unit 13 classifies the publicly known documents stored in the storage unit 11 into updated publicly known documents and other publicly known documents based on the above-described technique, and notifies the document filtering device 1 of the results. The document filtering device 1 calculates the conditional probability and the prior probability based on the notification from the classification unit 13.

[0052] The present invention is not limited to the above-described embodiments, and various modifications can be made in the implementation stage without departing from the spirit of the invention. Furthermore, the embodiments may be implemented in appropriate combinations, in which case the combined effects can be obtained. Furthermore, the above-described embodiments include various inventions, and various inventions can be extracted by combining selected elements from the disclosed elements. For example, if the problem can be solved and the desired effect can be obtained even if some elements are deleted from all elements shown in the embodiments, the configuration from which these elements are deleted can be extracted as an invention.

[0053] REFERENCE SIGNS LIST 1 document filtering device 11 memory unit 12 memory unit 12a configuration management history 13 classification unit 14 conditional probability calculation unit 15 priori probability calculation unit 16 probability calculation unit 17 filtering unit 18 notification unit 101 processor 102 ROM 103 RAM 104 communication module 105 display 106 storage

Claims

1. A document filtering device that filters update notification documents that notify users to update their software, comprising: a conditional probability calculation unit that calculates, for each word contained in an input public document, a first conditional probability, which is the probability that each word is contained when the public document is an update public document, and a second conditional probability, which is the probability that each word is contained when the public document is not an update public document; a prior probability calculation unit that calculates a prior probability, which is the ratio of update public documents to non-update public documents among the public documents stored in a first memory unit; a probability calculation unit that calculates the probability that the input public document is an update public document using Bayes' theorem based on the first conditional probability, the second conditional probability, and the prior probability; a filtering unit that filters the input public document based on the probability that the input public document is an update public document; and a classification unit that classifies public documents stored in the first memory unit into update public documents and non-update public documents based on the software update history stored in a second memory unit. The document filtering device, wherein the prior probability calculation unit calculates the prior probability based on a ratio of update public documents classified by the classification unit to public documents that are not update public documents.

2. The document filtering device of claim 1, wherein the classification unit extracts public notices stored in the first memory unit during a predetermined period prior to the software update date and time registered in the update history, and classifies the extracted public notices into update public notices and public notices that are not update public notices by comparing the extracted public notices with the contents of the software update.

3. The document filtering device according to claim 1, wherein the classification unit classifies, from among the extracted publicly known documents, publicly known documents that include the name of the software that has been updated as updated publicly known documents.

Images